Publication Record
Evidence current through: Thursday, November 19, 2026, at 12:39 a.m. UTC (7:39 p.m. ET and 4:39 p.m. PT on Wednesday, November 18, when every source had been re-read).
Microsoft’s cost controls announced at Ignite on November 17, 2026, are in preview or the Frontier program, and several credit rates are not yet published. The note reads prices, rates, and statuses as Microsoft and its rivals gave them by the time above. Developments after that time are not reflected. Times in the text are given in UTC.
Research Note In Two MinutesNote synopsis · AI drafted · Analyst reviewed
Drafted by AI from the full research note and reviewed by a Greyhound Research analyst before publication. The analysis and conclusions are Greyhound Research’s own.
At Microsoft Ignite on November 17, 2026, Microsoft left its seat prices where they were and widened the meter beside them. Microsoft 365 Copilot stays at USD 30 per user a month, Agent 365 at USD 15, and Microsoft 365 E7 at USD 99. Copilot Credits, Microsoft’s usage-based currency, now also pays for Windows 365 for Agents, a new billing option in Fabric, and the Today Feed in Outlook and the Copilot app. Agent 365 Cost Management and Business Analytics, which Microsoft calls FinOps for AI, are meant to show and control that spend. Most of the new pieces are in preview or the Frontier early-access program.
Greyhound Research’s position is that predictability has moved from Microsoft’s price list to the spending policy the customer writes. A seat costs the same every month. A credit buys a task whose price is a range, drawn from a pool shared across the whole company. The policy that decides who may spend, how much, and on which services is now the control a finance team owns. Greyhound Research reads this as a familiar stage in any move to metered pricing, and a manageable one. Microsoft has published more of its meter than its rivals have, and most of the variance that remains comes from settings and habits the buyer controls.
Three findings carry the argument. First, Copilot Credits are one currency but not yet one bill: some services are set up in the Microsoft 365 admin center, others in the Power Platform admin center or an Azure subscription. Second, a spending limit is a ceiling, not a budget. It reserves nothing, it stops the user’s work for the rest of the month when it is reached, and by default every new metered service joins existing policies. For Copilot Business bought through Microsoft’s partners, Microsoft plans to switch usage-based billing itself on by default from early 2027, with a limit of 4,000 credits per user a month. Third, Microsoft now meters the spend and also scores the value, so the business should keep its own measure of what the work was worth.
The note sets out a ten-move Greyhound CXO Runbook, with eight exhibits, seven tables, and six questions for Microsoft. None of the moves waits on Microsoft. The first three are simple to state. Switch off automatic coverage of new services and write one policy per business unit. Budget the credit line as a range. Decide what happens at the limit for each critical workflow before it goes live.
The evidence has limits. Several credit rates are not yet published, including those for the Today Feed, Fabric, and Windows 365 for Agents. Microsoft’s licensing guide says all pricing is subject to change. Greyhound Research names the signals that would change its view within twelve months. Until they arrive, the advice holds: buy prepaid credits from measured use, and negotiate the terms for credits at renewal as hard as the price of seats.
In This Research NoteSection guide · AI drafted · Analyst reviewed
Eighteen sections in eight parts, each with a short summary. Select a title to go straight to that section. Each section opens with a Greyhound Spotlight, the one fact in it a reader would not expect, and thirteen end with a block named The Deeper Read, which holds the fuller argument and the passages it rests on. Select any line in that block to open its full text.
- Part 1What Ignite Changed
- 1.1What Microsoft Announced: Seat Prices Held, The Meter Widened
No seat price changed at Ignite. Copilot Credits now reach Windows 365 for Agents, Fabric, and the Today Feed, and FinOps for AI arrives in preview and Frontier.
- 1.2Two Ways To Pay: The Seat Buys Everyday AI, The Credit Buys The Hard Work
Microsoft now sells AI in two parts: a fixed seat for everyday use and credits for long, multi-step work. Which part pays for which feature is the first thing a budget must map.
- Part 2The Meter
- 2.1One Currency, Several Bills: Copilot Credits Across Microsoft
Copilot Credits now pay for Microsoft 365, Copilot Studio, Windows 365 for Agents, and Fabric work. The currency is shared, and the places where it is set up and watched are not yet one.
- 2.2Pooling And Caps: A Limit Is A Ceiling, Not A Budget
Credits are pooled across the whole tenant and prepaid credits are drawn first. A spending limit reserves nothing, stops the user’s work for the month, and covers new services by default.
- 2.3Predictability: Seats Are Flat, Tasks Are Ranges
Microsoft publishes credit costs as light, medium, and heavy bands, with the top band open-ended. A credit budget is a range, and several Ignite rates are not yet published.
- Part 3Value And Control
- 3.1FinOps For AI: Cost Management And Business Analytics
Agent 365 Cost Management sets limits, alerts, model access, and chargeback. Business Analytics forecasts spend and measures value. Most of it is in preview or Frontier, and its coverage list leaves gaps.
- 3.2Value Scored By The Seller: Who Measures What The Work Was Worth
Business Analytics measures AI value in hours assisted and value assisted. The spend is measured, the value is estimated, and the estimate comes from the company that sends the bill.
- Part 4The Contract
- 4.1The License Gate: Agent 365 And E7 Still Hold Enterprise Governance
Ignite loosened two license gates at the edges and kept the main one. Enterprise-grade agent governance and insight into other companies’ agents still need Agent 365 or E7.
- 4.2Renewal Leverage: Negotiate The Credit, Not Only The Seat
Seats renew yearly at list prices that held at Ignite. Prepaid credits are bought a year ahead, expire unused, and carry rates Microsoft says can change. The leverage is in the credit terms.
- 4.3Three Vendors Compared: Every Agent Platform Now Has A Meter
Microsoft, Salesforce, and Google each pair seats with a usage meter. They differ on what happens at the limit, whether prepaid units expire, and how commitments are discounted.
- Part 5The Risks And The Counter-Argument
- 5.1Risks To Watch: Where A Metered Estate Can Go Wrong
Five risks, from policies that widen without a decision to value scored by the seller, each with an early warning and an owner.
- 5.2The Strongest Objection: Paying By Use Is Fairer Than Paying By Seat
The fair counter-case: metering matches cost to use, protects light users, and comes with real controls. It holds on fairness and falls short on predictability.
- Part 6What To Do, By Role
- 6.1The Greyhound CXO Runbook: Ten Moves Before The First Large Bill
Ten moves for finance, FinOps, the CIO, procurement, business leaders, architects, and the board, each naming the assumption it corrects and who decides.
- 6.2Questions For Microsoft: What Buyers Will Ask Next
Six questions only Microsoft can answer, on behavior at the limit, coverage, missing rates, rate protection, the value method, and the license gate. Each is tied to a Runbook move.
- Part 7The Final Greyhound Standpoint
- 7.1Agent Economics: The Final Greyhound Standpoint
The seat is flat, the meter is a range, and the spending policy is the control. The limits of the evidence, and the signals that would change the view within twelve months.
- Part 8Appendix, Sources, And Disclosures
- 8.1Appendix: The Evidence Register, Source By Source
Every source behind this research note, with when it was read and where it is used. Passages quoted in The Deeper Read are not repeated here.
- 8.2Sources And References: Every Document Behind Every Claim
Thirty-four sources in four classes, from Microsoft’s announcements, documentation, licensing guidance, and price lists to Salesforce, Google, a Microsoft partner, and the research this note builds on.
- 8.3Disclosures: Independence, Financial, Intended Use, Methodology, Authoring, And Copyright
Independence first, then the Financial Disclosure, intended use, research method and its limits, authoring, and copyright terms.
Where To Start, By RoleReader guide · AI drafted · Analyst reviewed
Find your role and read its sections in the order shown. Each line takes you straight to the section.
Chief financial officer
FinOps lead
Head of procurement
Chief information officer
Business leader who owns an AI program
Part 1What Ignite Changed
What Microsoft Announced: Seat Prices Held, The Meter Widened
Greyhound SpotlightMicrosoft changed no seat price at Ignite. The changes to how customers pay for agents came through Copilot Credits, the meter that sits beside the seats.
Microsoft announced no new price for its seats at Ignite. A seat is a license for one named person, paid every month at a fixed price. Microsoft 365 Copilot is listed at USD 30 per user a month, Agent 365 at USD 15, and Microsoft 365 E7, which bundles both with Microsoft 365 E5 and more security tools, at USD 99 on its plans page. What changed is the meter beside the seats.
Holding list prices while widening the meter is a common move in enterprise software, and it draws attention away from the number buyers watch most closely. For three decades, Microsoft budgets have been built on seat counts multiplied by list prices, with the annual true-up settling the difference. A metered line does not wait for the true-up. It moves every month, and the people who move it are users, not the procurement team. Greyhound Research expects more Microsoft services to join the meter over the coming year, as Microsoft’s documentation already anticipates. The effect most coverage misses is organizational. The AI budget now carries a line that no single contract signature controls.
Table 1 sets the money items of Ignite 2026 by status. Frontier is Microsoft’s early-access program, and a preview is a test release without full service terms.
Greyhound Research · Table 1 · Status ledger
Seat prices held, and the new meters have no rates
The money items of Microsoft Ignite 2026 by status, with the seat prices and credit rate a buyer can sign today, November 19, 2026.
Sign on the four published prices. Carry every meter without a rate as a placeholder, not a zero.
| Item | Status | What Microsoft has published | Not yet published |
|---|---|---|---|
| Microsoft 365 Copilot seat | Generally available | USD 30 per user a month, paid yearlyIncluded in Microsoft 365 E7 | No change announced at Ignite |
| Agent 365 | Generally available | USD 15 per user a month, paid yearlySince May 1, 2026 | No change announced at Ignite |
| Microsoft 365 E7 | Generally available | USD 99 per user a month with Teams, USD 90.45 withoutSince May 1, 2026 | Whether any Copilot Credits are included |
| Copilot Credits, pay-as-you-go | Published rate | USD 0.01 per credit, billed after the monthPrepaid P3 plan, 5% to 20% off by volume | Rate protection for a term |
| Agent 365 Cost Management | Preview | Limits, alerts, model access, chargeback dataForecasting in Frontier this month, value insights in December | General availability date, full coverage list |
| Agent 365 Business Analytics | Frontier | Spend forecasts and value signals for business leaders | How hours and value are estimated |
| Windows 365 for Agents in Copilot Credits | Eligible, no rate | “now eligible for Copilot Credits” | Credit rate, place in Cost Management |
| Copilot Credits for Fabric | Announced, no rate | A usage-based option beside capacity pricing | Credit rate, status, release date |
| Today Feed | Private preview, no rate | “Recurring updates throughout the day require Copilot credits.” | Credit rate |
Table 1 lists the money items of Ignite 2026 by status, and shows the seat prices and credit rate unchanged while the new meters have no published rate. Unchanged: Microsoft 365 Copilot at USD 30 per user a month, Agent 365 at USD 15, Microsoft 365 E7 at USD 99 with Teams or USD 90.45 without, and Copilot Credits at USD 0.01 each on pay-as-you-go terms, with a prepaid plan discounted 5% to 20% by volume. New controls: Agent 365 Cost Management in preview and Agent 365 Business Analytics in Frontier. New meters without a rate: Windows 365 for Agents in Copilot Credits, Copilot Credits for Fabric, and the Today Feed. Each row names what is not yet published. Statuses and prices are as Microsoft published them through November 19, 2026.
Only the three seat prices and the Copilot Credits rate are published in full, and none of them changed. The new items are previews, Frontier releases, or eligibility statements without a rate. The Book of News says Windows 365 for Agents is now eligible for Copilot Credits. The Azure blog calls credits a new billing option for Fabric, beside its existing capacity pricing. For a finance team, the year’s AI budget now has a fixed line and a variable line, and only the fixed line has a full rate card.
Greyhound Standpoint:
At Greyhound Research, we believe Ignite 2026 was a pricing event without a price change, and that holding the seats while widening the meter was the right commercial choice for Microsoft and, on balance, for buyers. It keeps everyday AI on a line a finance team already knows how to forecast, and it charges for costlier work as that work happens. The consequence is less visible. Every service added to Copilot Credits moves a share of AI spend from a figure agreed at signature to a figure set by use, and the work of predicting it moves from Microsoft’s price list to the buyer.
Enterprise technology has made this turn before. When infrastructure moved to the public cloud, many organizations kept budgeting it like the hardware it replaced and met the difference in the bill, and FinOps grew as a practice from that experience. Microsoft deserves credit for publishing a unit rate and task bands before most buyers need them. The limit is that several new rates are unpublished and most of the new pieces are in preview, so the size of the shift cannot yet be measured. Published rates and a few quarters of steady use would ease this view. Chief financial officers should set up Copilot Credits in the next budget cycle as a spending category with its own owner, forecast, and approval rules, rather than as an extra inside the Microsoft 365 renewal.
The Deeper Read · About 2 minutes
Select a question to open the full answer
1What exactly did Microsoft announce about money?
In Microsoft’s words. The Book of News frames the theme plainly: AI spending is growing fast and moving beyond IT. It places Agent 365 Cost Management in the Microsoft admin center as a single place for usage-based AI services, and Agent 365 Business Analytics in the Frontier program. Under Windows, it says Windows 365 for Agents is now eligible for Copilot Credits, which it calls a common currency. Under Today, it says the morning briefing is included with a Copilot license while updates through the day need credits.
Fabric and the Azure side. Arun Ulag, Executive Vice President of Microsoft Azure, wrote that Copilot Credits arrive as a new usage-based billing option for Microsoft Fabric, alongside the existing capacity-based options. The post gives no rate and no release status.
What the wording tells a buyer. Microsoft chose the phrase common currency, and its Book of News describes AI spending as moving beyond IT. Both descriptions are accurate, and together they show where Microsoft expects the budget conversation to go: out of the IT line and into the business units. A buyer should read the missing rates as a matter of timing rather than a pricing decision. Eligibility statements come first and rates follow, so a budget built now should carry a placeholder for each eligible service rather than a zero.
In the source’s own words
- Microsoft’s Book of News, FinOps for AIAI spending is growing fast and moving beyond IT, and leaders need to see what it costs and delivers.Source
- Microsoft’s Book of News, Windows 365 for Agentsa common currency for eligible Microsoft services with usage-based billing.Source
- Microsoft Azure Blog, Ulag on November 17Copilot Credits as a new usage-based billing option for Microsoft FabricSource
- Microsoft Azure Blog, Ulag on November 17providing greater flexibility alongside existing capacity-based optionsSource
2Are the seat prices really unchanged?
The March baseline holds. Judson Althoff, Chief Executive Officer of Microsoft’s commercial business, announced in March 2026 that Microsoft 365 E7 and Agent 365 would be generally available on May 1, at USD 99 and USD 15 per user. Microsoft’s pricing pages re-read on November 19 still list those prices, and Microsoft 365 Copilot at USD 30, each paid yearly. None of the Ignite posts read for this note announces a change.
In the source’s own words
- The Official Microsoft Blog, Althoff on March 9General availability of the new Microsoft 365 E7: The Frontier Suite on May 1 for $99 per userSource
- The Official Microsoft Blog, Althoff on March 9Priced at $15 per user, Agent 365 gives IT and security leaders a single place to observe, govern, manage and secure agentsSource
Part 1What Ignite Changed
Two Ways To Pay: The Seat Buys Everyday AI, The Credit Buys The Hard Work
Greyhound SpotlightCopilot’s own router can suggest moving a user from Chat, which the seat covers, to Cowork, which is billed in credits.
Microsoft now sells its AI at work in two parts. Jared Spataro, Chief Marketing Officer for Copilot, describes a predictable seat for everyday use and a usage-based model for the hardest tasks. Summaries and drafts sit in the seat. Cowork, which carries out long, multi-step work across files and apps, is paid for in credits on top of the seat.
The two-part model is new for Microsoft 365, though not for enterprise software. Database and middleware licensing has long separated a base entitlement from premium options, and buyers learned to track which features sat in which tier. What differs here is that the boundary is crossed in the flow of work, often after a suggestion from the product itself, rather than at a purchasing decision. Greyhound Research expects Microsoft to keep redrawing the line as features mature, which Spataro signals when he says capability will move into the subscription over time. The less visible consequence is that the feature map becomes a living document, and someone in finance has to keep it current.
Exhibit 1 maps the main features against the three ways to pay: the Copilot seat, Copilot Credits, and an Agent 365 or E7 license. A feature can need a seat and still be paid for in credits.
Greyhound Research · Exhibit 1 · What pays for what
The line between seat and meter runs through the product
Main Microsoft 365 AI features against the three ways to pay for them, as Microsoft documented them by November 19, 2026.
A seat-only budget misses the features Microsoft promotes most. Map each feature to the part that pays for it.
Exhibit 1 is a matrix of eight Microsoft AI features against three ways to pay, the Copilot seat, Copilot Credits, and an Agent 365 or E7 license, and shows the line between seat and meter running through the product. Everyday chat and the Today Brief are included in the seat. The Today Feed is metered in credits. Cowork, Code, and advanced work in SharePoint require a seat and are metered in credits, shown as a tinted row. Windows 365 for Agents is eligible for credits and Fabric has credits as a new option. Business Analytics for Microsoft services needs one Copilot license in the tenant. Governance and insight for other platforms’ agents need Agent 365 or E7. Cells marked not stated are not documented in the sources read. The reading is Greyhound Research analysis.
The exhibit shows the line between the two parts running through the middle of the product. Everyday chat and the Today Brief, the morning summary, are included in the seat. The Today Feed, which refreshes through the day, needs credits. Cowork and Code need a seat and credits. Insight into agents built on other companies’ platforms needs Agent 365 or E7. A budget built on seats alone will miss the features Microsoft is promoting most. A companion Greyhound Research note on Copilot as the new front door for work covers what Home, Cowork, and Autopilot change for users. This note covers what they cost.
Greyhound Research wrote in August 2026 that Microsoft had kept a USD 30 Copilot license “carrying no entitlement to its agentic layer at all.” For the agentic work Microsoft promotes most, that still holds after Ignite, because Cowork, Code, and Autopilot draw on credits above the seat. One qualification is due. Microsoft’s licensing guide shows agents built in Copilot Studio with its Standard and Copilot Chat harnesses running in Microsoft channels inside the license, subject to fair-use limits. The line between seat and meter therefore sits a little closer to the seat than that sentence implied, and Microsoft decides where it sits.
Greyhound Standpoint:
Greyhound Research’s view is that the two-part model is fair in design and hard to forecast in use. Charging for heavy work by use is reasonable, because long, multi-step work costs more to run than a summary does. The difficulty lies in where the line sits. It runs through the middle of the product, and it moves as Microsoft ships features and as Auto recommends Cowork for complex work. A user who accepts that suggestion moves spend from a fixed line to a variable one without anyone making a purchasing decision.
Practitioners will recognize the pattern from software editions and mobile plans. The base plan covers ordinary use, the premium tier is one click away, and the bill tracks how often people take the click. That is a feature of the design rather than a fault in it, and Microsoft says it will move more capability into the seat over time, which would lower the credit line without a new negotiation. The evidence does not yet show how often the router recommends Cowork, so the scale of the effect is unknown, and first-quarter usage data would settle it. Finance and IT leaders should map each feature their people use to the part that pays for it before the next budget cycle and refresh it each quarter. At renewal, they should ask for the current split in writing, with notice before any feature moves from the seat to the meter.
The Deeper Read · About 2 minutes
Select a question to open the full answer
1How does Microsoft describe the two parts?
A seat and a meter. Spataro’s Ignite post sets out the commercial model in two steps, a per-user subscription and a usage-based consumption model, and adds that Microsoft will keep moving capability into the subscription. Microsoft’s support page for users says everyday AI is included with a Microsoft 365 Copilot license and that the organization sets limits on credits for advanced tasks.
What the seat already covers. Microsoft’s licensing guide says runtime in Microsoft channels for agents built with the Standard harness, and for agents built with the Copilot Chat harness, is included with the Microsoft Copilot license, with fair usage limits. Copilot Studio agents built in the GitHub Copilot harness are metered in credits.
The router’s role. The same post says Auto, Copilot’s model selector, chooses models and effort for each task, and that Auto may recommend moving from Chat to Cowork for more complex work. That recommendation crosses from the seat to the meter. The Learn overview lists Cowork, Code, and advanced work in SharePoint and OneDrive among the services that use credits and require a Copilot license.
What the user-facing page tells a buyer. Microsoft’s support page tells employees that everyday AI is included and that their organization sets the limits for advanced tasks. That wording places the limit, in the user’s eyes, with the employer rather than with Microsoft. When a user reaches a limit in the middle of the month, the request for more credits and the complaint that goes with it will come to the company’s own IT team. Rollout plans for Cowork should therefore include a short note to users on what the limit is, who set it, and how to ask for more.
In the source’s own words
- Microsoft Copilot Blog, Spataro on November 17First, our predictable, per-user subscription makes powerful Copilot capabilities broadly accessible.Source
- Microsoft Copilot Blog, Spataro on November 17Second: a usage-based consumption model that gives you access to the most powerful AI for the very hardest tasks.Source
- Microsoft Copilot Blog, Spataro on November 17For more complex work, Auto may recommend moving from Chat to Cowork.Source
- Microsoft Support, usage-based billing for AI at workEveryday AI is included with a Microsoft 365 Copilot license.Source
- Microsoft Licensing, Copilot Credits guidanceand agents built using the Copilot Chat harness are included with the Microsoft Copilot USL (fair usage limits apply).Source
2Where does the Today feature fall?
Split in two. The Book of News says the Today Brief is included with a Microsoft 365 Copilot license and that recurring updates through the day require Copilot Credits. Today begins in private preview through the Frontier program. No credit rate for the Today Feed was found in the documents read.
Part 2The Meter
One Currency, Several Bills: Copilot Credits Across Microsoft
Greyhound SpotlightWindows 365 for Agents can now be paid for in Copilot Credits, while its billing page still sends cost tracking to Azure Cost Management.
Copilot Credits are Microsoft’s unit for AI work paid by use. Microsoft’s licensing guide prices one credit at USD 0.01 on pay-as-you-go terms, billed after the month ends. The currency is now common to Microsoft 365 work, Copilot Studio agents, Windows 365 for Agents, and Fabric. A common currency helps a finance team compare costs across products, but only if the spend lands in one place.
A shared unit is a meaningful change for finance teams that have long reconciled Microsoft spend across more than one portal. In earlier pricing, each product line arrived with its own meter and its own report, and comparing them took manual work. A common currency removes the conversion step but not the reconciliation step, because the spend still surfaces in three places and the usage report counts credits rather than money. Greyhound Research expects the consoles to converge, given Microsoft’s stated aim of one cost-management experience, though the timing is Microsoft’s to set. Until then, the second-order effect falls on the month-end close. The team that books AI spend will need a reconciliation routine that most did not need before.
Exhibit 2 sets out the three places where credit spending is set up or watched today. An admin center is the web console where administrators manage a product.
Greyhound Research · Exhibit 2 · Where credits are set up and watched
One currency, three places to control it
Where each service that uses Copilot Credits is set up, limited, and reported, as Microsoft documented it by November 19, 2026.
One currency is not yet one bill. Until the consoles join up, the buyer’s own map is the control.
Exhibit 2 groups the places where Copilot Credits spending is set up or watched into three cards. The Microsoft 365 admin center holds spending policies and Cost Management for Cowork, Code, advanced work in SharePoint and OneDrive, Managed Runtime, and Work IQ APIs, with policy and user limits, alerts, and credit requests, new services joining existing policies by default, and usage reported in credits rather than money. The Power Platform admin center holds Copilot Studio and prepaid packs of 25,000 credits a month, with credit policies for Copilot Chat only. An Azure subscription carries Windows 365 for Agents, whose costs show in Azure Cost Management and whose pay-as-you-go use is billed to the subscription, and Fabric, where credits are a new option beside capacity pricing. Neither of the last two is in Cost Management’s named coverage list, though Microsoft’s Windows post says credit eligibility brings Windows 365 for Agents into one billing, governance, and cost-management experience. The grouping is Greyhound Research analysis.
The currency is shared, but the controls are spread across three places. Spending policies live in the Microsoft 365 admin center. Copilot Studio and the prepaid packs for Copilot Chat are managed through the Power Platform admin center. Windows 365 for Agents bills through an Azure subscription and reports in Azure Cost Management. Agent 365 Cost Management is Microsoft’s answer, and the services it names do not yet include Fabric or Windows 365 for Agents. Microsoft’s own pages read two ways on this point. Pavan Davuluri, Executive Vice President of Windows and Devices, wrote that credit eligibility brings Windows 365 for Agents into one billing, governance, and cost-management experience. The Book of News list of services covered by Cost Management does not name it, and its billing page still points to Azure.
Greyhound Standpoint:
Greyhound Research holds that a common currency is a real step for buyers and that it is not yet a single bill. Microsoft has given finance teams one unit to plan in, which no earlier Microsoft AI price list did, and that deserves credit. The places where the unit is set up, limited, and reported still follow the product lines: spending policies in the Microsoft 365 admin center, Copilot Studio in the Power Platform admin center, and Windows 365 for Agents in an Azure subscription. Each console shows part of the spend, and the usage report counts credits rather than money, so the finance team carries the work of joining them.
This is a familiar stage in a large vendor’s pricing. A shared unit usually arrives before a shared console, because billing systems are built product by product and joined later. The gap tends to close, but at the vendor’s pace rather than the buyer’s. Microsoft has named Agent 365 Cost Management as its answer and says Windows 365 for Agents now shares one cost-management experience. Documented coverage of it and of Fabric in that console would change this reading. FinOps leads should draw the map for their own tenant now, showing which service bills where, which policy covers it, and which report shows it. At each architecture review, a new credit-eligible service should be placed on that map before anyone is allowed to use it.
The Deeper Read · About 3 minutes
Select a question to open the full answer
1What does Microsoft’s documentation say about the currency?
One unit, several methods. Microsoft’s Learn overview, updated November 3, calls Copilot Credits a common currency and says organizations can use fixed licensing, the Copilot Credit Pre-purchase Plan, known as P3, pay-as-you-go billing, prepaid capacity packs, or combinations of these. It asks administrators to connect an Azure subscription to support billing. It also says Microsoft will add more agents and services over time.
Usage, not money. The same page says the reporting interface counts usage in credits and does not include pricing, invoices, currency costs, or purchasing forecasts. A finance team converting credits to money must apply its own rates.
In the source’s own words
- Microsoft Learn, usage-based billing overviewCopilot Credits are a common currency for eligible Microsoft services with usage-based billing.Source
- Microsoft Learn, usage-based billing overviewMicrosoft will add more agents and services over time and provide notifications and updated content as they become available.Source
- Microsoft Learn, usage-based billing overviewThe API reports usage in AI credits and does not include pricing, invoices, currency costs, or purchasing forecasts.Source
- Microsoft Licensing, Copilot Credits guidancePricing: $0.01/Copilot CreditSource
2Where is each service billed and watched?
Windows 365 for Agents. Its Learn billing page, updated June 3, says the service mixes pay-as-you-go pricing with monthly charges for each always-available Cloud PC, billed through an Azure subscription, and that its costs can be tracked in Azure Cost Management. The Ignite announcement of credit eligibility post-dates that page.
Copilot Chat and Copilot Studio. The Learn page on capacity packs says each prepaid pack includes 25,000 credits a month for the whole tenant, and that credit policies there are currently available only for Copilot Chat. Copilot Studio is billed through the Power Platform admin center unless the customer opts into the Microsoft 365 admin center, per the overview.
Two readings of one service. The Book of News says the eligibility of Windows 365 for Agents enables a unified billing, governance, reporting, and cost-management experience, and Davuluri’s post says the same in shorter words. The Learn billing page, dated June 3, has not been updated to show it.
A further meter. The security post on Microsoft Perception mentions new cost management dashboards and a cost estimator for AI usage. It gives no list price.
What the dated pages tell a buyer. The Learn billing page for Windows 365 for Agents is dated June 3, and the announcement that brings the service onto credits came in November. Documentation that trails an announcement is normal in a release cycle this fast, and it carries a practical message. Until the page changes, the documented billing route is the Azure subscription, and a FinOps team should reconcile on that basis while tracking the announced one. The dated page is the better evidence for an invoice. The announcement is the better evidence for the direction of travel.
In the source’s own words
- Microsoft Learn, Windows 365 for Agents billingWindows 365 for Agents uses a mix of consumption-based (pay-as-you-go) pricing and per-Cloud-PC monthly charges (always available), billed through your existing Azure subscription.Source
- Microsoft Learn, capacity packsEach Prepaid Capacity pack is a tenant license that includes 25,000 Copilot Credits per month.Source
- Microsoft Learn, capacity packsCopilot credit policies are currently only available for Microsoft Copilot Chat.Source
- Windows Experience Blog, Davuluri on November 17now eligible for Copilot Credits, bringing it into one billing, governance and cost-management experienceSource
- Microsoft Security Blog, Holeček on November 17New cost management dashboards and a cost estimator help you track AI usageSource
Part 2The Meter
Pooling And Caps: A Limit Is A Ceiling, Not A Budget
Greyhound SpotlightEvery spending policy covers new metered services automatically unless an administrator switches the setting off. Microsoft made that the default in September 2026.
Credits are pooled. Microsoft’s licensing guide says they are pooled at the tenant level, which means the whole company draws on one supply. A spending policy then sets how much a user, group, or agent may draw. The policy is a limit, not an allocation. It does not set credits aside for anyone, so a busy team can use up prepaid credits that a quieter team was counting on.
Pooling changes the politics of the budget as much as its size. Under seats, each department paid for its own people and could see what it had bought. Under a tenant-wide pool, departments draw on one supply, and the first team to scale an agent can use up what others had planned on. Shared cloud accounts showed the same pattern before chargeback became common. Greyhound Research expects internal disputes over who used the credits to appear before any dispute with Microsoft over the bill. What tends to be overlooked is that a limit set per user does not divide the pool between business units, so that allocation still has to be decided and written down.
Exhibit 3 follows a credit from the moment a user starts a task to the monthly report. Pay-as-you-go means billed after use, at the list rate, with no commitment.
Greyhound Research · Exhibit 3 · The path of a credit
A limit stops the work, and the policy can grow by default
Five steps from a user starting a metered task to the monthly report, read against Microsoft’s documentation, November 19, 2026.
A spending limit is a ceiling, not a budget. Write its scope and its behavior at the limit before the meter runs.
Exhibit 3 sets out five steps a Copilot Credit passes through as a numbered list, and shows that Microsoft has written the first three in full and the last two in part. Step one, a spending policy covers the user, with limits and alerts and no reserved credits, and new services join by default. Step two, prepaid credits from one tenant pool are drawn first, and unused P3 credits expire after a year. Step three, further use is billed at USD 0.01 per credit after the month, with no ceiling unless a limit is set. Step four, a user who reaches a limit loses access for the rest of the month and can request more, while the fate of a half-done task is not written. Step five, spend is reported by user, service, agent, and model in credits, without Fabric or Windows 365 for Agents named. The steps and grades are Greyhound Research analysis.
Microsoft has written most of the path. Prepaid credits are used first, then pay-as-you-go, and a user who reaches a limit loses access for the rest of the month and can ask for more. Two gaps matter. Microsoft has not said what happens to a task that is halfway done when the limit is reached. And because new services join existing policies by default, the scope of a policy can grow without anyone deciding it should.
Two other defaults sit beside that one, and they point in different directions. In Microsoft’s setup guide, usage-based billing is a step an administrator takes to unlock Cowork and the other metered experiences, so the buyer makes the first decision, and Microsoft deserves credit for that. For Microsoft 365 Copilot Business bought through Microsoft’s Cloud Solution Provider partners, a Partner Center notice of November 3 says usage-based billing will be on by default for new subscriptions from January to March 2027, moved from December 1, 2026, with a default spending limit of 4,000 Copilot Credits per user a month that administrators can adjust. That limit is a ceiling Microsoft sets, not a budget the customer wrote.
Greyhound Standpoint:
Greyhound Research’s reading is that a spending limit protects the bill and can interrupt the work, and that buyers should plan for both. Spataro is right that a limit prevents runaway spend, and Microsoft has documented the path in more detail than many vendors do. The mechanism still carries a business consequence. Credits are pooled across the tenant and a policy reserves none of them, so a busy team can draw down what a quieter team expected to use. When a user reaches the limit, access stops for the rest of the month, which turns a finance control into an operating decision.
Shared pools are not new. Pooled data plans and cloud commitments taught buyers that default settings, more than headline rates, decide where surprises come from, and here the default adds every new metered service to existing policies unless an administrator switches it off. From early 2027, Copilot Business customers buying through partners will also start with a limit Microsoft chose. The evidence has a clear limit. Microsoft has not said what happens to a task that is halfway done when the limit is reached, and documentation of that behavior would soften this view. FinOps leads should switch off automatic coverage and add each new service only after a recorded decision. Before any critical workflow passes its pilot gate, the CIO should decide whether reaching the limit means a pause, a request for more, or pay-as-you-go spend.
The Deeper Read · About 3 minutes
Select a question to open the full answer
1How do pooling and limits work?
Pooled, then limited. The licensing guide says credits are pooled at the tenant level and, for P3, deducted automatically from the prepaid balance. Microsoft’s Learn page on managing usage-based billing, updated November 11, says spending policies do not reserve or allocate credits, that prepaid credits are always used first, and that credits reset on the first of the month.
A floor under the ceiling. The same page says Microsoft’s guidance for a monthly user budget is at least 2,000 credits, that a policy must meet a required minimum, and that these recommendations may evolve. The page does not say which services the minimum covers.
At the limit. The same page says users who reach a limit lose access to agents and services for the rest of the month, and that they can request more credits but do not fall through to other policies. Spataro’s post presents the limit as protection against surprises.
In the source’s own words
- Microsoft Licensing, Copilot Credits guidanceCopilot Credits are pooled at the tenant level.Source
- Microsoft Learn, managing usage-based billingSpending policies are limit-based controls and don't reserve or allocate Copilot Credits to users or groups.Source
- Microsoft Learn, managing usage-based billingthey lose access to agents and services for the rest of the monthSource
- Microsoft Learn, managing usage-based billingSet at least 2,000 credits per user per monthSource
- Microsoft Copilot Blog, Spataro on November 17In Cowork, your team sets a spending limit, so there's no runaway spend and no surprises when the bill arrives.Source
2Why does the default matter?
Coverage that grows. Microsoft’s Message Center notice MC1470425, dated September 10, 2026, says the setting that applies policies to new services is on by default for all spending policies. The Learn overview repeats it and asks administrators to review whether automatic coverage is appropriate. A policy written in October can therefore cover a service released at Ignite in November, at whatever rate that service carries.
The partner channel. The Partner Center notice says Microsoft will create the required Azure resources and apply the 4,000-credit default to all users in the customer’s organization, that limits can be adjusted in Partner Center or the Microsoft admin center, and that turning usage-based billing off might limit access to AI experiences. Nikhil Fernandes, a product manager at Sherweb, a Microsoft partner, wrote that the risk lies less in consumption than in consumption without ownership, guardrails, or a measure of outcomes.
What the two defaults tell a buyer. The default for automatic coverage and the default limit for partner customers come from two notices, issued in September and November, and each is reasonable on its own terms. Read together, they show Microsoft setting the starting position for both the scope and the size of a policy, and leaving the buyer to change either. Fernandes’s point about ownership fits that reading. A default is a decision made in advance by someone else, and the buyer’s task is to replace it with a recorded decision before the first bill rather than after it.
In the source’s own words
- Microsoft 365 Message Center, MC1470425The Auto-apply new services setting is enabled by default for all spending policies.Source
- Microsoft Learn, usage-based billing overviewThe Auto-apply new services setting is enabled by default for spending policies.Source
- Microsoft Partner Center, November 3 noticeThe default spending limit is 4,000 Copilot Credits per user per month, which admins can adjust.Source
- Sherweb blog, Fernandes, updated November 5The real risk is not consumption itself.Source
Part 2The Meter
Predictability: Seats Are Flat, Tasks Are Ranges
Greyhound SpotlightMicrosoft’s own rate card lists a heavy Cowork task as more than 700 credits, with no upper figure.
A seat costs the same every month whether it is used or not. A credit is spent only when work is done, and the amount depends on the work. Microsoft’s licensing guide publishes its rates as bands, light, medium, and heavy, for each kind of task. That is more than many vendors publish. It also means the realistic forecast for a credit budget is a range, not a number.
Bands are a fair way to price variable work, and they move the forecasting problem rather than solve it. A seat forecast needs one input, the headcount. A credit forecast needs three: how many tasks people run, how heavy each one is, and which model handles it. Neither of the last two appears in a human resources system or a license report. Greyhound Research expects the first year of credit budgets to rest partly on assumption and to be corrected quarter by quarter, much as early cloud budgets were. Less noticed is that forecast accuracy now depends on usage data held by IT, so finance and IT will need to share it on a set schedule.
Exhibit 4 shows Microsoft’s published bands for four kinds of metered work, on one scale, and marks the services whose rates are not yet published. A scenario is one piece of work a user or agent completes.
Greyhound Research · Exhibit 4 · What a task costs in credits
Every published band is open at the top
Microsoft’s published Copilot Credit bands for four kinds of metered work, and the Ignite services with no published rate, November 19, 2026.
Published bands explain spend after the fact. Budget the credit line as a range with a stated top before it.
Exhibit 4 shows Microsoft’s published Copilot Credit bands as horizontal range bars on a scale from 0 to 1,000 credits, and shows that every published band is open at the top. A Cowork task runs from 100 credits for light work to more than 700 for heavy. A Copilot Studio agent scenario in the GitHub Copilot harness runs from 100 to more than 500. A Copilot Code task runs from 25 to more than 200. A Teams Phone Agent call runs from 5 to more than 40. A short hatched end on each bar shows that the heaviest band has no upper figure. The Today Feed, Fabric, and Windows 365 for Agents have no credit rate in the guide, shown as a dashed bar with a question mark. One credit is USD 0.01 on pay-as-you-go terms.
Each published band has a floor and a ceiling, except the heaviest, which is open. The same kind of task can sit at either end, depending on how much the agent reads, how many steps it takes, and which model Auto picks. Three of the services Ignite added have no published rate at all. A forecast built today on the light band will run over as soon as users find the heavy uses.
This marks a change in Greyhound Research’s own reading. It wrote in November 2025 that new pricing units such as credits were “deliberately designed to obscure value exchange” and were often “metered without transparency or pre-defined ceilings.” Copilot Credits answer much of that charge. Microsoft publishes the unit rate and bands by task, and buyers can set their own ceilings. The view has therefore moved from opacity to predictability. The open question is no longer what a credit is but how many a task will use, and Microsoft calls its own bands rough planning estimates.
Greyhound Standpoint:
In Greyhound Research’s assessment, Microsoft has made credit costs more visible than its rivals have, and buyers should give it credit for publishing bands by task. Bands make spend explainable after the fact. They do not make it predictable beforehand, because the mix of light and heavy work is set by what users ask for and by the model Auto selects, not by the price list. A forecast built on the light band holds until people find the heavy uses, and the heaviest Cowork band has no upper figure.
The pattern is familiar from metered infrastructure. Early cloud budgets were often built from list rates and a guess at volume, and the variance came from the shape of use rather than from price. The organizations that brought it under control measured first and committed later. This agrees with a companion Greyhound Research note on Azure agent infrastructure, which held that savings from Context Cache should be booked only from measured reuse. The limit here is that three Ignite rates are not yet published, so any range drawn today is partial, and their publication would narrow it. Chief financial officers should budget the credit line as a range with a stated top at the next planning round, and review the mix of light and heavy work monthly for the first two quarters.
The Deeper Read · About 2 minutes
Select a question to open the full answer
1What does the rate card say?
Bands by task. Microsoft’s licensing guide lists Cowork tasks at 100 to 300 credits for light, 300 to 700 for medium, and more than 700 for heavy. It lists Copilot Code at 25 to 60, 75 to 100, and more than 200 credits per task, and Copilot Studio agents running in the GitHub Copilot harness at 100 to 300, 300 to 500, and more than 500. A Teams Phone Agent call is listed at 5 to 15 credits for calls under a minute, and more than 40 for calls over two minutes. The guide gives no credit rate for the Today Feed, Fabric, or Windows 365 for Agents.
Estimates, not quotes. The guide calls its figures rough planning estimates and says actual pricing will vary with workflows and usage patterns. It does not say whether a task that fails, or is run again, is charged.
Prices can move. The same page says all pricing is in US dollars and subject to change, and the Learn page on managing billing says recommendations and policy requirements may evolve.
What “rough planning estimates” tells a buyer. Microsoft’s own label for its bands is careful and fair. It tells buyers that the figures are a starting point for a forecast, not a quote the vendor can be held to. Two gaps in the same guide matter for the range: whether a failed or repeated task is charged, and the missing rates for three new services. A buyer drawing a range today should widen it for both, and narrow it as measured use and published rates arrive.
In the source’s own words
- Microsoft Licensing, Copilot Credits guidancePricing: $0.01/Copilot CreditSource
- Microsoft Licensing, Copilot Credits guidanceAll pricing is in USD and subject to change.Source
- Microsoft Licensing, Copilot Credits guidanceThese figures are rough planning estimates. Actual pricing will vary based on workflows and usage patterns.Source
- Greyhound Research, published commentary, November 26, 2025In many cases, these units are metered without transparency or pre-defined ceilings.Source
2Who decides how heavy a task is?
The router, in part. Spataro writes that Auto chooses models and effort for each task, balancing quality, speed, and cost. Agent 365 Cost Management lets administrators control which models teams can use. Between them, those two settings decide much of where a task lands in its band.
Greyhound Research’s earlier finding. Greyhound CIO Pulse 2025 found that 67% of CIOs globally identify escalating inference costs as their most pressing barrier to AI adoption, and that 58% of Fortune 1000 CIOs in North America had paused or scaled back agent pilots, with 41% reporting overspend of 25% or more. Greyhound Research’s July 2025 note on Copilot’s research agents recorded CIOs’ concerns about budgeting predictability when Microsoft first metered agent queries.
In the source’s own words
- Microsoft Copilot Blog, Spataro on November 17Under the hood, Auto dynamically selects the right models and effort for the task, balancing quality, speed, and cost.Source
- Microsoft Tech Community, Agent 365 Ignite postControl which models teams can use so organizations can reserve premium intelligence for the work that requires it.Source
- Greyhound Research, procedural memory note67% of CIOs globally identify escalating inference costs as their most pressing barrierSource
- Greyhound Research, Copilot research agents noteCIOs have flagged concerns about budgeting predictabilitySource
Part 3Value And Control
FinOps For AI: Cost Management And Business Analytics
Greyhound SpotlightThe services Microsoft names for Agent 365 Cost Management include GitHub Copilot CLI and Autopilot, and do not include Fabric or Windows 365 for Agents.
FinOps is the practice of managing cloud spending as a shared job between finance, technology, and the business. Microsoft’s FinOps for AI has two parts. Agent 365 Cost Management, in the Microsoft admin center, shows what drives spend and lets administrators set policies, limits, and alerts. Agent 365 Business Analytics, in the Frontier program, gives business leaders forecasts and a view of value.
Placing FinOps inside the vendor’s own console is a notable step. In cloud infrastructure, the discipline grew up largely outside the provider, and native cost tools came later. Microsoft is offering its native tools as the meter widens, which shortens the period in which buyers manage spend without them. The trade-off is that a vendor’s console reports best on the vendor’s own estate, and Copilot Credits now reach products the console does not yet list. Greyhound Research expects most large buyers to run Microsoft’s tools beside their existing cost systems for some time. Less discussed is the reporting line. Once business leaders hold delegated budgets, FinOps sits with finance and the business as much as with IT.
Table 2 sets each control Microsoft has described beside its status and what is not yet written. Chargeback means billing each department for the AI it uses.
Greyhound Research · Table 2 · FinOps for AI, control by control
The right controls, mostly in preview, with gaps in coverage
Agent 365 Cost Management and Business Analytics controls as Microsoft described them at Ignite, with status and gaps, November 19, 2026.
The right controls, not yet the full scope. Prove them against your own invoices before retiring other cost tools.
| Control | What Microsoft has written | Status | Not yet written |
|---|---|---|---|
| Spending limits and alerts | At user, group, organization, or resource level | Preview | Task behavior at the limit |
| Model access | Control which models teams can use | Status not stated | Credit cost by model |
| Delegated budgets | Business leaders and delegates manage their own budgets | Planned, December | Scope of delegation |
| Forecasts and recommendations | Forecast spend, ask in plain language, act on recommendations | Frontier, this month | Forecast method |
| Chargeback data | Graph APIs, exports, and Fabric connectors | Preview | Money values in the usage API |
| Coverage of new services | New services join existing policies by default | In use since September | Notice before a new service is added |
| Services covered | Cowork, Code, GitHub Copilot CLI, Copilot Studio, Managed Runtime, Autopilot, Work IQ APIs | Preview | Fabric and Windows 365 for Agents |
| Value insights across agents | Microsoft and other companies’ agents | Planned, December | Method and audit rights |
Table 2 lists eight controls of Agent 365 Cost Management and Business Analytics with what Microsoft has written, their status, and what is not yet written, and shows the right controls, mostly in preview, with gaps in coverage. Spending limits and alerts at four levels are in preview. Model access has no stated status, and delegation to business leaders is planned for December. Forecasts are in Frontier this month. Chargeback data through Graph APIs, exports, and Fabric connectors is in preview. Automatic coverage of new services has been the default since September. The named services include Cowork, Code, GitHub Copilot CLI, Copilot Studio, Managed Runtime, Autopilot, and Work IQ APIs, and not Fabric or Windows 365 for Agents. Value insights across agents are planned for December. Statuses are as Microsoft published them through November 19, 2026.
The controls are the right ones. Limits at four levels, model access, delegated budgets, forecasts, and data for chargeback cover most of what a FinOps team asks for. Most of them are in preview or Frontier, and value insights for agents on other platforms, with delegation to business leaders, arrive in December. The gap a CFO will notice is coverage. The named services stop short of two products Ignite has just put on credits.
Greyhound Standpoint:
Greyhound Research’s judgment is that Agent 365 Cost Management is the most useful announcement Ignite made for finance teams. Limits at four levels, model access, delegated budgets, and chargeback data answer what buyers have asked for since the first metered agents. Model access matters most for money. It lets administrators keep premium models for the work that needs them, and the model chosen largely decides where a task lands in its band. Greyhound Research said in April 2026 that dashboards “merely describe the bill while it arrives,” and that architecture lowers it. Cost Management earns this judgment for the controls that act before the spend, not for its reports.
Its worth depends on coverage and status. The named services stop short of Fabric and Windows 365 for Agents, two products Ignite has just put on credits, and most of the controls are in preview or Frontier. Preview programs often change scope and behavior before general availability, and a vendor’s cost console tends to see that vendor’s own estate best. Both are reasons to test it, not faults in the design. Documented full coverage at general availability would remove most of this qualification. FinOps leads should run it in the Frontier program against their own invoices, list every credit-using service it does not show, and keep their existing cost tools until that coverage is written down. Retiring those tools belongs at the first review after general availability.
The Deeper Read · About 2 minutes
Select a question to open the full answer
1What does Cost Management do?
Four levels of limits. The Agent 365 Ignite post says administrators can apply spending limits and alerts at the user, group, organization, or resource level, track consumption across users, services, agents, and models, control which models teams can use, and let business leaders and their delegates manage their own budgets. It names Microsoft Graph APIs, data exports, and Fabric connectors as routes into chargeback workflows.
Coverage and status. The Book of News names Copilot Cowork, Copilot Code, GitHub Copilot CLI, Copilot Studio, Copilot Managed Runtime, Autopilot, and Work IQ APIs as services covered. It marks the Fabric chargeback analysis as in preview, says forecasting and recommendations reach the Frontier program this month, and says value insights across Microsoft and other agents follow in December, together with delegation to business leaders.
What the coverage list tells a buyer. Naming the covered services is good practice, because it lets a buyer see exactly what the console will show. The list also marks the edge of the console. GitHub Copilot CLI and Autopilot are inside it, while Fabric and Windows 365 for Agents, both newly on credits, sit outside it for now. A FinOps team should treat the published list as the scope of the tool on the day it was read, and check it again at each release, since Microsoft has already scheduled further capabilities for December.
In the source’s own words
- Microsoft Tech Community, Agent 365 Ignite postApply spending limits and alerts at the user, group, organization, or resource level to keep usage aligned with budget.Source
- Microsoft Tech Community, Agent 365 Ignite postLet business leaders and their delegates manage their own AI budgets and policies within Cost Management.Source
- Microsoft’s Book of News, FinOps for AIextend analysis through Microsoft Fabric for custom chargeback and reporting.Source
- Microsoft’s Book of News, FinOps for AIIn December, these capabilities will expand to include value insights across Microsoft and non-Microsoft agentsSource
- Greyhound Research, published commentary, April 28, 2026The dashboards merely describe the bill while it arrivesSource
2What does Business Analytics need?
Two license conditions. The Book of News says insights for Microsoft services need at least one Microsoft 365 Copilot license in the tenant, and insights for other companies’ services need an Agent 365 license. The Agent 365 post words the second condition as Agent 365, ME7, or both, and says Business Analytics begins rolling out in Frontier.
Part 3Value And Control
Value Scored By The Seller: Who Measures What The Work Was Worth
Greyhound SpotlightMicrosoft’s value measure starts from an estimate of how long a task would take a person, and Microsoft makes the estimate.
Microsoft wants customers to judge AI by its results. Spataro says Business Analytics measures investment in work completed, not simply tokens, the units in which AI models count text. That is the right question. The tool that answers it is supplied by the company that also sends the bill.
Value measurement has moved closer to the vendor with each generation of workplace software. Productivity tools once reported adoption, then usage, and Microsoft now offers an estimate of the hours and value its AI assisted. Each step is more useful to a buyer and more dependent on the vendor’s assumptions. The resulting figure is likely to travel from a dashboard into a business case and then a board pack, losing its caveats on the way. Greyhound Research expects internal audit teams to ask, sooner rather than later, where AI return figures come from. The second-order effect falls on the business case itself, since a return figure supplied by the vendor is hard to defend once spend comes under review.
Exhibit 5 follows the chain from spend to value, and marks where measurement ends and estimation begins. Assisted hours are hours of human work that the AI is credited with saving.
Greyhound Research · Exhibit 5 · From spend to value
The spend is counted, and the value is estimated
The chain from AI spend to AI value in Agent 365 Business Analytics, marked where measurement ends and estimation begins, November 19, 2026.
Half the chain is counted and half is estimated. The business, not the seller, should own the value figure.
Exhibit 5 follows the chain from AI spend to AI value in Agent 365 Business Analytics along a bar in four equal parts, with a marker at the midpoint where estimation begins. The first two parts are counted from Microsoft’s records: spend in credits by user, service, agent, and model, and activity, meaning tasks, success rate, and satisfaction. The last two parts are estimated by Microsoft’s model: assisted hours, the time the work would take a person and the share the AI did, and assisted value, what those hours are worth, whose method is not stated in the posts read. The split between counted and estimated is Greyhound Research analysis.
The first half of the chain is counted. Spend, tasks, and success rates come from Microsoft’s own logs. The second half is estimated. How long a task would have taken a person, and what those hours were worth, are model outputs, and the documents read do not set out the method. The figure at the end will appear in budget reviews as if it were measured.
Greyhound Standpoint:
Greyhound Research’s position is that measuring AI by completed work is the right test, and that the measure belongs to the buyer. Microsoft’s signals are useful inputs, especially for comparing teams inside one company, and moving the conversation from tokens to tasks is a change finance leaders should welcome. The mechanism needs care. Spend, tasks, and success rates are counted from Microsoft’s logs. Assisted hours and assisted value rest on an estimate of how long a person would take, and that estimate is made by the company that sends the bill.
Outsourcing contracts taught the same lesson. Where the supplier reported both the service and the savings, buyers that kept their own baseline negotiated from their own records, and buyers that did not negotiated from the supplier’s account. Nothing in the record suggests Microsoft’s estimate is biased. It is unaudited, and its method is not set out, so published method documentation that customers can test would raise the weight it can carry. Greyhound CIO Pulse 2026 found that just 32% of enterprise technology leaders say most of their materially funded AI initiatives have delivered measurable business outcomes against their approved business cases, and a measure the vendor supplies will not close that gap alone. Before the next budget review, chief financial officers and program owners should set value measures from the business’s own records, such as cases closed, days saved, or errors avoided, and use Microsoft’s figures as a cross-check.
The Deeper Read · About 1 minute
Select a question to open the full answer
1What does Business Analytics measure?
Signals named. The Agent 365 post says Business Analytics assesses value through signals such as assisted hours, assisted value, success rate, and satisfaction, and offers spend forecasts and scenario planning. Spataro’s post says it analyzes the tasks, how long that work typically takes a person, and the share Cowork performed. Neither post sets out how the time a person would take is estimated.
What the signal names tell a buyer. Assisted hours and assisted value are careful names. They claim help rather than replacement, and Microsoft deserves credit for not calling them savings. A finance team should keep that distinction when the figures move into a business case. An assisted hour is not an hour removed from a payroll or a schedule, and treating it as one would turn a reasonable signal into an overstated return.
2Has Greyhound Research made this point before?
On outcomes. Greyhound Research reported in its July 17, 2026, note on IBM’s results that Greyhound CIO Pulse 2026, based on responses from more than 1,000 leaders, finds just 32% of enterprise technology leaders saying a majority of their materially funded AI initiatives have met their approved business cases. In its August 2025 note on procedural memory, it described a two-tier cost model, learning with the largest model and executing with the smallest one that will do. Business Analytics measures the result of that choice from the seller’s side of the table.
In the source’s own words
- Greyhound Research, July 17, 2026 noteGreyhound CIO Pulse 2026 finds that just 32 per cent of enterprise technology leaders say a majoritySource
Part 4The Contract
The License Gate: Agent 365 And E7 Still Hold Enterprise Governance
Greyhound SpotlightUsage insights for Microsoft’s own agents no longer need an Agent 365 license. Governing agents built on other companies’ platforms still does.
A license gate is the point where a feature needs a paid license. Ignite moved two gates outward. Usage insights for Microsoft agents in the admin center are now available whatever the Agent 365 status, and agent insights in Viva Insights no longer need Agent 365. The main gate stayed where it was. Enterprise-grade management, security, and governance of agents still need Agent 365 or E7.
Ignite moved this gate in the direction buyers would want. Making usage insights for Microsoft’s own agents available without Agent 365 lets a company see part of its estate before it pays to govern it, which lowers the cost of a first inventory. The commercial effect tends to follow the inventory. Once an estate is visible, the agents built on other platforms come into view too, and governing them still needs the paid license. Greyhound Research expects the gate to keep moving at the edges as features reach general availability. The less obvious consequence is timing. An inventory taken before renewal gives procurement a count to negotiate with, and one taken after renewal does not.
Table 3 sets out what each money and governance feature needs, from a single Copilot license in the tenant to Agent 365 or E7 for every user who needs it.
Greyhound Research · Table 3 · The license gate
Visibility into Microsoft agents is free, control over others is not
What each money and governance feature needs, from no extra license to Agent 365 or Microsoft 365 E7, November 19, 2026.
Visibility is free and control is paid. Size governance seats to the people who govern other platforms’ agents.
| What the buyer wants | What it needs | Status |
|---|---|---|
| Everyday Copilot in Microsoft 365 apps | A Copilot seat, or E7 | Generally available |
| Cowork, Code, advanced work in SharePoint | A Copilot seat plus Copilot Credits | Available |
| Usage insights for Microsoft agents in the admin center | No Agent 365 license needed | Generally available, loosened at Ignite |
| Agent insights in Viva Insights | At least 50 assigned Copilot licenses, no Agent 365 | Generally available, loosened at Ignite |
| Business Analytics for Microsoft services | At least one Copilot license in the tenant | Frontier |
| Business Analytics for other companies’ services | Agent 365, E7, or both | Frontier |
| Enterprise-grade agent management, security, and governance | Agent 365 or E7 | Generally available, unchanged |
Table 3 lists seven features with the license each needs, and shows visibility into Microsoft’s own agents freed at Ignite while control over other platforms’ agents still needs Agent 365 or E7. Everyday Copilot needs a Copilot seat or E7. Cowork, Code, and advanced work need a seat plus credits. Usage insights for Microsoft agents in the admin center need no Agent 365 license. Agent insights in Viva Insights need at least 50 Copilot licenses and no Agent 365. Business Analytics for Microsoft services needs one Copilot license in the tenant. Business Analytics for other companies’ services needs Agent 365, E7, or both. Enterprise-grade agent management, security, and governance need Agent 365 or E7. Statuses are as Microsoft published them through November 19, 2026.
The pattern is consistent. Microsoft gives away visibility into its own agents and charges for control over everyone’s. For a buyer weighing Agent 365 at USD 15 against E7 at USD 99, the question is how many people need governance of agents built elsewhere. A companion Greyhound Research note on agent identity covers what that governance does. This note covers what it costs.
Greyhound Standpoint:
Greyhound Research concludes that Ignite left the Agent 365 and E7 gate where it was, and that this is the price of Microsoft’s aim to govern agents built on other vendors’ platforms. Making usage insights for Microsoft’s own agents available without Agent 365 lowers the cost of getting started, and Microsoft deserves credit for it. The control that matters in a mixed estate, governance of agents from other platforms, still sits behind a paid seat. For buyers, the cost of that control depends less on the price per seat than on how many seats they buy.
Free visibility and paid control is a long-standing way to price enterprise software, and a reasonable one. The exposure for buyers sits in the bundle. Suites priced below the sum of their parts have a long record of producing shelfware, licenses paid for across the headcount and used by a few. E7 at USD 99 is good value where most of its seats use most of its parts, and less so where Agent 365 alone is needed. A feature-by-license table for the Ignite releases at general availability would sharpen this reading, and a move of governance features below the gate would change it. At the next renewal, procurement should price Agent 365 and E7 against the number of people who govern agents on other platforms, not against total headcount, and record the count it used.
The Deeper Read · About 1 minute
Select a question to open the full answer
1What exactly did Ignite change in the gates?
Two loosened, one held. The Book of News says that usage insights for Microsoft agents are now available regardless of Agent 365 licensing, that Viva Insights keeps a requirement of at least 50 assigned Copilot licenses, and that customers will still need an Agent 365 or ME7 license for full governance. ME7 is Microsoft’s short name for Microsoft 365 E7.
What the word “still” tells a buyer. The Book of News uses still twice in a short passage, once for the Agent 365 or ME7 requirement and once for enterprise-grade governance. On Greyhound Research’s reading, the repetition is deliberate. Microsoft is signaling that the loosening at Ignite happened at the edges and that the governance gate is a settled part of its commercial design. Buyers counting on the gate moving further at general availability should treat that as unlikely on the current wording, and size their licenses on the gate as it stands.
In the source’s own words
- Microsoft’s Book of News, management insightsretaining the requirement for at least 50 assigned Copilot licenses.Source
- Microsoft’s Book of News, management insightscustomers will still require an Agent 365 or ME7 license.Source
- Microsoft’s Book of News, management insightsAn Agent 365 license is still required for customers to get enterprise-grade management, security and governance.Source
2What is in E7?
The bundle. Althoff’s March post says E7 unifies Microsoft 365 E5, Microsoft 365 Copilot, and Agent 365, adds the Entra Suite and advanced Defender, Intune, and Purview capabilities, and is priced below buying those parts separately. It does not mention Copilot Credits. Microsoft’s plans page lists E7 at USD 99 with Teams and USD 90.45 without, each paid yearly.
In the source’s own words
- The Official Microsoft Blog, Althoff on March 9Microsoft 365 E7 unifies Microsoft 365 E5, Microsoft 365 Copilot and Agent 365 into a single solution powered by Work IQSource
- The Official Microsoft Blog, Althoff on March 9At $99 per user, E7 is priced below purchasing these capabilities à la carteSource
Part 4The Contract
Renewal Leverage: Negotiate The Credit, Not Only The Seat
Greyhound SpotlightPrepaid Copilot Credits bought under Microsoft’s one-year plan expire at the end of the term, and the largest volume discount listed is 20%.
Renewal leverage is the buyer’s power to win better terms when a contract comes up for renewal. For a decade, Microsoft renewals have turned on the price per seat. Credits change the subject. Microsoft’s prepaid plan, P3, is paid a year in advance for a pool of credits, with discounts from 5% to 20% by volume. Unused credits expire when the year ends.
For most Microsoft customers, the renewal team has spent years refining one negotiation, the price and count of seats. Credits bring a second negotiation with different mechanics: commitment size, expiry, rate stability, and the treatment of unused balance. Those mechanics are familiar to anyone who has negotiated a cloud commitment, but that experience often sits with a different team. Greyhound Research expects that, in many organizations, the people who negotiate consumption commitments will not be the people who renew the Microsoft agreement. The consequence that tends to be missed is internal. Leverage at renewal depends less on the vendor than on whether the buyer brings both sets of skills to the same table.
Exhibit 6 sets out one year under P3 next to annual seats, and marks the points where a buyer gains or loses leverage. Carry-over means unused credits move into the next term.
Greyhound Research · Exhibit 6 · One year of prepaid credits
The calendar rewards the buyer who measures first
One twelve-month term of Microsoft’s Copilot Credit Pre-purchase Plan beside annual seats, with the points where buyers gain or lose leverage.
The leverage sits in the credit terms. Measure for six months, then negotiate the rate, carry-over, and notice.
Exhibit 6 lays out one twelve-month term of Microsoft’s prepaid Copilot Credit plan as a month grid, and shows Microsoft’s terms and Greyhound Research’s advice side by side. In month one, the plan is paid up front for the year and annual seats renew. Spending limits reset on the first of every month. In month six, Greyhound Research advises having six months of measured use in hand. In month nine, it advises negotiating the next term: a held credit rate, carry-over of unused credits, and notice of changes. In month twelve, unused prepaid credits expire and seats auto-renew. Notes beneath give the volume discount of 5% at 300,000 credits to 20% at 300 million, the monthly packs of 25,000 credits, and Microsoft’s statement that pricing is subject to change.
The calendar favors the buyer who measures. A team that buys P3 on a forecast either pays pay-as-you-go rates for the overrun or loses what it did not use. A team that buys after six months of measured use can buy close to its real need. Because the rates can change, the strongest terms to seek are a held rate for credits through the term and at renewal, not only a discount on seats. Buyers that hold an Azure consumption commitment have one more lever. Microsoft’s guidance says those committed funds can be applied to eligible Copilot consumption when billing runs through the subscription linked to the commitment.
Greyhound Standpoint:
Greyhound Research finds that renewal leverage has moved from the seat to the credit. Seat prices held, and they will be negotiated as they have been for years. What is new to most Microsoft agreements is a metered line whose rates Microsoft says are subject to change, whose coverage widens by default, and whose prepaid form expires unused at the end of the year. A buyer who negotiates only the seat leaves the fastest-moving part of the spend on standard terms.
Buyers of public cloud learned that committed-spend discounts reward accurate forecasts and cost the buyer when forecasts run high, and that rate protection can be worth more than a headline discount. P3 follows that shape, with discounts from 5% to 20% and no carry-over. Microsoft has published more of these terms than many vendors, and the discount tiers are real. The limit is that no rate protection is documented, and a held rate through a term would change this reading most. Greyhound Research’s July 2025 note warned that bundling weakens renegotiation leverage, and that holds more strongly when part of the bundle is metered. Procurement should bring three terms to every renewal that includes credits: a held credit rate for the term, a carry-over or credit-back for unused prepaid credits, and written notice before any rate or coverage change.
The Deeper Read · About 2 minutes
Select a question to open the full answer
1What are the P3 terms?
One year, paid up front. Microsoft’s licensing guide describes P3 as a one-year, pay-up-front purchase option, with credits deducted automatically from the balance and unused credits expiring at the end of the annual term. It lists nine volume tiers, from 300,000 credits at 5% off to 300 million credits at 20% off. If use runs past the balance, customers can buy more or be billed pay-as-you-go.
Committed Azure funds. Microsoft’s guidance for partners and commitment holders says that funds under an Azure consumption commitment apply to eligible Copilot consumption only when the selected subscription links to the billing account that holds the commitment.
Packs reset monthly. The Learn page on capacity packs says each pack’s 25,000 credits are replenished at the start of every monthly billing period, and that without pay-as-you-go, Copilot Chat stops for users covered by the policy when prepaid credits run out.
What the three clocks tell a buyer. The terms run on three different cycles. Spending limits and capacity packs reset monthly, P3 runs for a year and then expires, and seats renew on the agreement’s own calendar. Each clock creates a moment when unused value is lost or a decision falls due. A procurement team that lines the three up on one calendar, as Exhibit 6 does, can time its negotiation for the point at which it holds the most measured data and the least unused balance.
In the source’s own words
- Microsoft Licensing, Copilot Credits guidanceCopilot Credit P3 is a one-year pay-upfront purchase option for Copilot Credits.Source
- Microsoft Licensing, Copilot Credits guidanceUnused credits expire at the end of the annual term.Source
- Microsoft Learn, CSP, partner, and MACC guidanceIf your organization has an Azure Consumption Commitment (MACC), you can apply those committed funds toward eligible Copilot consumption.Source
- Microsoft Learn, capacity packsWhen prepaid credits are exhausted, Copilot Chat is unavailable for users covered by the policy until credits replenish at the start of the next monthly billing period.Source
2What did Greyhound Research say before?
Bundling and leverage. Greyhound Research’s July 2025 note on Copilot’s research agents expected advanced agents to be folded into higher plans and found that renegotiation leverage is materially diminished when they are. Its November 15 note on the Microsoft 365 outage argued that consolidating on one suite can remain right while its standard terms fall short. Credits are a new case of the same pattern.
In the source’s own words
- Greyhound Research, Copilot research agents noterenegotiation leverage is materially diminished.Source
Part 4The Contract
Three Vendors Compared: Every Agent Platform Now Has A Meter
Greyhound SpotlightGoogle’s caps pause an agent’s calls when a project hits its limit. Microsoft’s caps stop the user’s access for the rest of the month.
Microsoft is not alone. Salesforce sells Agentforce with seats and Flex Credits, and Google sells Gemini Enterprise with seats and a pay-as-you-go edition it announced in August. Every large agent platform now mixes a fixed seat with a meter. For a buyer, the comparison has moved from the price per seat to the rules around the meter.
That every large agent platform now pairs a seat with a meter is itself the finding. It suggests the industry has settled on a hybrid model, and that the next round of competition will turn on terms rather than on whether to meter at all. In the move to subscriptions, rival vendors also converged on one model and then competed on the fine print. Greyhound Research expects the rules at the limit and for unused units to become the main points of difference over the coming year. For buyers running more than one platform, the overlooked consequence is that each meter carries its own rules, so a single spending policy will need a clause for each vendor.
Table 4 compares the three on each vendor’s own pricing page or announcement. A commitment discount is a lower rate in return for agreeing to spend a set amount.
Greyhound Research · Table 4 · Three vendors, term by term
Ahead on published task costs, behind on the cap
Microsoft, Salesforce, and Google compared on the terms around the meter, from each vendor’s own pricing page or announcement, November 19, 2026.
No vendor leads on every rule. Choose on behavior at the limit and on unused units before comparing rates.
| Term | Microsoft | Salesforce Agentforce | Google Gemini Enterprise |
|---|---|---|---|
| Seat for everyday AI | Copilot, USD 30 per user a month | Agentforce add-ons, USD 125 per user a month | Fixed monthly fee per user with pooled quota |
| Unit of use | Copilot Credit | Flex Credit, 20 per standard action | Consumption at pay-as-you-go rates |
| List rate | USD 0.01 per credit | USD 500 per 100,000 credits | Not stated in the post read |
| Pooling | Pooled at tenant level | Applied across teams, channels, and use cases | Pooled across users in an edition, project, and location |
| At the limit | User loses access for the rest of the month | No overage penalty, billed at contracted rate | Agent calls pause, resume with a click or allow overages |
| Unused prepaid units | P3 credits expire at year end | Do not roll over into later terms | Not stated in the post read |
| Commitment discount | P3, 5% to 20% by volume | Pre-Commit with true-up, later this year | Savings plans, 10% for one year, 20% for three |
| Published cost by task type | Bands for Cowork, Code, Studio, phone | Not stated | Not stated |
Table 4 compares Microsoft, Salesforce, and Google on eight terms around the usage meter, and shows Microsoft ahead on published task costs and behind Google on cap behavior. Seats: Copilot at USD 30, Agentforce add-ons at USD 125, and a fixed Gemini Enterprise fee with pooled quota. Units: Copilot Credits at USD 0.01, Flex Credits at USD 500 per 100,000 with 20 per action, and Google consumption with no rate in the post read. Pooling is tenant-wide at Microsoft, across teams at Salesforce, and across users in a project at Google. At the limit, Microsoft stops the user for the month, Salesforce bills at the contracted rate with no penalty, and Google pauses agent calls. Unused prepaid units expire at Microsoft and do not roll over at Salesforce. Commitments: Microsoft 5% to 20%, Salesforce pre-commit with true-up later this year, Google 10% or 20%. Only Microsoft publishes cost bands by task type. Two cells are not stated.
Microsoft publishes the most detail on what a task costs, through its bands. Google is ahead on the behavior of its caps and offers a three-year commitment, though its post does not say whether unused commitments lapse. Salesforce states that it charges no overage penalty. None of the three gives a single figure for what an agent’s month will cost. Two cells are not stated in the documents read.
Greyhound Standpoint:
Greyhound Research places Microsoft level with its rivals on metering, ahead on published task rates, and behind Google on what happens at the cap. Microsoft’s bands give buyers a better starting point for a forecast than either rival publishes. Google’s caps pause an agent’s calls, while Microsoft’s stop a user’s access for the month, and Salesforce states that it charges no overage penalty. Google’s post does not say whether unused commitments lapse, so this note does not rank the two on commitments. Each vendor leads on a different rule, and on every platform the meter is now the larger negotiation.
As markets for a new kind of computing mature, the comparison tends to move from the list rate to the terms around it, because rates converge faster than terms do. Cloud infrastructure went through that movement, and the same one is visible here. The limit is that this comparison rests on each vendor’s own pages, two cells are not stated in the documents read, and Google’s pay-as-you-go edition is still rolling out. A rival changing its cap behavior or its expiry rules would reset the order. CIOs choosing or renewing a platform should compare the rules at the limit and the treatment of unused credits before they compare rates, because a low rate with a cap that stops work, or prepaid units that expire, can cost more than a higher rate without them.
The Deeper Read · About 2 minutes
Select a question to open the full answer
1What do Salesforce and Google publish?
Salesforce. Its Agentforce pricing page lists Flex Credits at USD 500 per 100,000, with 20 credits for each standard action. It says credits can be applied across teams, channels, and use cases, that unused credits do not roll over into later terms, and that there is no overage penalty. Agentforce add-ons are listed at USD 125 per user a month with unmetered employee agent use.
Google. Its August 26 post by Michael Gerstenhaber and Pravir Gupta says seat quotas are pooled across users in the same edition within a project and location, that agent calls pause when a project hits its limit, and that alerts go out at 50%, 80%, and 100% of the budget. Spend-based savings plans give 10% off for one year or 20% off for three. The pay-as-you-go edition is available to select customers and rolling out.
In the source’s own words
- Salesforce, Agentforce pricingUnused Flex Credits do not roll over into subsequent subscription terms.Source
- Salesforce, Agentforce pricingThere is no overage penalty.Source
- Google Cloud Blog, August 26, 2026If a project hits its limit, the agent's API calls temporarily pauseSource
- Google Cloud Blog, August 26, 2026Usage quotas are pooled across users in the same edition within a project and location.Source
2Where is Microsoft ahead?
Published bands. Microsoft’s licensing guide gives credit bands by task type for Cowork, Code, Copilot Studio agents, and phone calls. Neither the Salesforce page nor the Google post read for this note publishes cost bands by task type. Microsoft’s Cost Management also promises limits at the user, group, organization, and resource level, which is finer than a project-level cap. Greyhound Research’s November 5 note on assistants as the front door to business applications used Microsoft’s Dynamics 365 documentation, which names both bills for one task on one page, as its standard for how the cost of one task should be written.
What the differences tell a buyer. The three vendors have each given a different answer to the same question: what should happen when a meter reaches its limit. Microsoft stops the user, Google pauses the agent, and Salesforce bills at the contracted rate without an overage penalty. None of the three is wrong. Each suits a different kind of work, from tasks that can wait to processes that cannot stop. A buyer comparing platforms should first decide which behavior its own critical workflows need, and then read each vendor’s terms against that requirement.
In the source’s own words
- Google Cloud Blog, August 26, 2026Receive 10% off for 1-year or 20% off for 3-year commitmentsSource
Part 5The Risks And The Counter-Argument
Risks To Watch: Where A Metered Estate Can Go Wrong
Greyhound SpotlightNone of the five risks needs a price rise. Each comes from how the meter is set up.
Greyhound Research sees five risks in a metered Microsoft estate. Each shows first inside the buyer’s own organization, in a report, a setting, or a contract it already holds, and each can be caught before the first large bill.
These risks differ from the pricing risks finance teams usually watch. A price rise arrives in a renewal quote and can be negotiated. The five here arrive through settings, defaults, and habits, and they show up in reports and policy screens well before they show up in money. That is why the early warnings matter as much as the risks themselves. Greyhound Research expects the first of them to appear in the monthly reconciliation, where a credit report and an Azure invoice do not agree. The point most coverage misses is ownership. Only one of the five sits with the CIO, so a register held by the technology team alone will leave most of them without an owner.
Table 5 sets each risk beside its early warning and its owner. Every early warning can be checked without help from Microsoft.
Greyhound Research · Table 5 · Risk register
Every risk comes from how the meter is set up
Five risks in a metered Microsoft AI estate, each with its early warning and its owner.
Each risk starts in a setting or a habit, not a price. Check every early warning this month, without Microsoft.
| Risk | Early warning in the buyer’s own estate | Owner |
|---|---|---|
| Policies that widen by defaultNew metered services join existing policies without a decision. | Automatic coverage switched on in any spending policy. | FinOps lead |
| A cap that stops critical workA user or agent loses access mid-month on a workflow that cannot stop. | A critical workflow list with no rule for the limit. | Chief information officer |
| Prepaid credits that expireP3 is bought on a forecast and part of it lapses at year end. | A P3 order placed before two quarters of measured use. | Head of procurement |
| Value scored by the sellerThe vendor’s estimate becomes the business case’s return figure. | A budget review that cites assisted value as its only measure. | Chief financial officer |
| Spend outside the consoleCredit spend on Fabric or Windows 365 for Agents sits outside Cost Management. | A monthly report that does not reconcile to the Azure invoice. | FinOps lead |
Table 5 lists five risks in a metered Microsoft AI estate, each with an early warning in the buyer’s own estate and an owner. Policies that widen by default, warned by automatic coverage switched on, owned by the FinOps lead. A cap that stops critical work, warned by a critical workflow list with no rule for the limit, owned by the CIO. Prepaid credits that expire, warned by a P3 order placed before two quarters of measured use, owned by procurement. Value scored by the seller, warned by a budget review citing assisted value as its only measure, owned by the CFO. Spend outside the console, warned by a monthly report that does not reconcile to the Azure invoice, owned by the FinOps lead. The risks are Greyhound Research analysis.
The five share a cause: treating credits as a line item inside the Microsoft renewal rather than as a spending category with its own controls. Each warning sits in something the buyer already has, such as a policy setting, a monthly report, or a contract. The table names owners, not likelihoods, and ranks none of the risks above another. None of the five needs Microsoft’s help to check.
Greyhound Standpoint:
In Greyhound Research’s reading, none of the five risks is unusual, and each is cheaper to prevent than to unwind. They share a cause, which is treating credits as a line item inside the Microsoft renewal rather than as a spending category with its own controls. A policy that widens by default, a limit that stops a critical workflow, prepaid credits that expire unused, a value figure taken on trust, and spend outside the console are each small in the first month. They grow with use, and by the time they show in an invoice the decisions behind them are months old.
Experienced FinOps and procurement teams will recognize the shape from cloud sprawl and from unused licenses in enterprise agreements. In both, the cost came less from prices than from settings and ownership left undefined, and the remedy was to give each one a name and a date. To its credit, Microsoft has given buyers most of the settings needed to act. The limit of this reading is that the table names owners, not likelihoods, and ranks none of the risks above another. This is a judgment about order, not a prediction of overspend. Each owner named in Table 5 should check the early warning this month and report it at the first quarterly spend review, before the first large bill arrives.
Part 5The Risks And The Counter-Argument
The Strongest Objection: Paying By Use Is Fairer Than Paying By Seat
Greyhound SpotlightThe strongest objection to this note is that seats carry their own unpredictability, because companies pay for many that go unused.
The objection deserves its full weight. Seats charge the same for a person who uses Copilot every hour and one who rarely opens it, and companies have long paid for unused licenses. Credits charge for work done. Microsoft has published rates in bands, offered limits at four levels, and built a cost console. On this view, metering is fairer than seats, and the controls make it safe.
The fairness argument carries real weight with business units, which have long paid for licenses their people rarely open. Metering moves cost toward the heaviest users, and that change tends to be welcomed by light users and questioned by the teams whose work is most automated. Greyhound Research expects the debate inside companies to turn less on whether metering is fair than on who pays for the heavy work. The less visible effect is on incentives. A team charged by use has a reason to hold back the very tasks the investment was meant to encourage, so the design of chargeback deserves as much care as the spending policy.
Exhibit 7 tests each part of the objection against the evidence. Three parts stand. The claim that the controls make spending predictable does not hold on the evidence.
Greyhound Research · Exhibit 7 · The objection held to the evidence
Three parts of the objection stand, and one falls
The fair counter-case to this research note, part by part, against the evidence gathered for each.
Fair, yes. Predictable, not yet. Move a program onto the variable line only once its use is measured.
Exhibit 7 holds four parts of the strongest objection to this research note against the evidence, as four cards, and shows that three stand and one falls. Paying by use is fairer than paying for idle seats: stands. Microsoft publishes what tasks cost, in bands in its licensing guide: stands. The controls are real, with limits at four levels, alerts, model access, and a cost console: stands. The controls make spend predictable: falls on the evidence, because limits stop work rather than forecast it, the top bands are open, and three rates are not published. The verdicts are Greyhound Research analysis.
Greyhound Standpoint:
Greyhound Research accepts the objection on fairness and rejects it on predictability. Paying for work done is a better match of cost to value than paying for idle seats, and Microsoft’s controls are real. Unused licenses have been one of the most persistent costs in enterprise software for decades, and credits remove that waste for light users. Three parts of the objection stand on the evidence, and this note’s case should be read with that weight.
The part that does not hold is the claim that the controls make spend predictable. Limits keep spend under a ceiling by stopping work, bands give a range rather than a figure, and several rates are not yet published. Each control reduces the size of a surprise without saying where spend will land. The history of usage-based pricing points the same way, since buyers have tended to accept metering once they can forecast it, and forecasting depends on stable rates and documented behavior at the limit. Both are open today, and both are within Microsoft’s power to close. Fair pricing and predictable pricing are different properties, and a chief financial officer needs both. Finance leaders should apply both tests to each credit-funded program at its pilot gate, and move it onto the variable line only once the predictability test has a measured answer.
The Deeper Read · About 1 minute
Select a question to open the full answer
1Why do the controls not make spend predictable?
A ceiling is not a forecast. Microsoft’s Learn page describes limits that block access once reached and notifications as users approach them. Those keep spend below a ceiling. They do not say where under the ceiling spend will land, and the published bands are open at the top. Microsoft’s own overview says it will add more services over time, each joining existing policies by default.
What would close the gap. Rates for every credit-eligible service, documented behavior when a limit is reached during a task, and a held rate for a term would turn the ceiling into a forecast a CFO can sign.
What the request route tells a buyer. The quoted line shows that a user at the limit can ask for more credits and does not fall through to another policy. That is a sensible guard against spend leaking between policies. It also means each request lands with a person, so the speed of approval becomes part of the cost of a limit. Where a workflow cannot wait, the approver and the expected response time belong in the policy alongside the number.
In the source’s own words
- Microsoft Learn, managing usage-based billingWhen users reach their limit within a policy, they can request more credits but they don't default to other policies.Source
Part 6What To Do, By Role
The Greyhound CXO Runbook: Ten Moves Before The First Large Bill
Greyhound SpotlightEvery move uses something the buyer already controls, from a policy setting, a budget, and a contract to a model setting, a feature map, and the risk register the board already reviews.
Ten moves follow from the evidence, at least one for each leader who will own part of the AI budget. Each names the assumption it corrects, the evidence behind it, and the executive who decides. None waits for Microsoft, because each uses controls the buyer already holds.
Most of the ten can start with tools the buyer already has. Switching off automatic coverage, deciding what happens at the limit, and setting model access are settings and decisions, not projects. Budgeting as a range and owning the value measure fit the next planning round. The billing map belongs to the architecture review and the risk register to the board’s risk review. The credit terms, the feature map, and the license count belong to the renewal. That spread reflects a lesson from earlier licensing shifts: settings are cheap to change early and harder to change once users depend on them. Greyhound Research expects the first and third moves to give most of the early protection. The overlooked benefit is portability, since a spending policy written for Copilot Credits can later carry a clause for a second vendor’s meter.
Table 6 sets the ten moves beside the assumption each corrects, its evidence, and its owner. Two of the ten finish only when Microsoft publishes more terms.
Greyhound Research · Table 6 · The Greyhound CXO Runbook by owner
Ten moves, and none waits on Microsoft to start
Ten moves for the leaders who will own part of the AI budget, with the assumption each corrects, its evidence, and its owner.
All ten moves can start this quarter with controls the buyer holds. Only two wait on Microsoft to finish.
Select a move to open its full text in The Deeper Read below
| Move | The assumption it corrects | Evidence anchor | The decision sits with |
|---|---|---|---|
| 1/Switch off automatic coverage and write one policy per business unit | A policy stays the size it was written | Exhibit 3, Table 2 | FinOps lead, with the CIO |
| 2/Budget the credit line as a range with a stated top | The light band is the forecast | Exhibit 4 | CFO |
| 3/Decide what happens at the limit for every critical workflow | A cap only protects the bill | Exhibit 3 | CIO, with workflow owners |
| 4/Buy prepaid credits from measured use and negotiate the credit terms | The seat price is the renewal | Exhibit 6, Table 4 | Head of procurement, with the CFO |
| 5/Own the value measure | The vendor’s dashboard is the return | Exhibit 5 | Business leaders, with the CFO |
| 6/Count the users who need Agent 365 or E7, not the headcount | Governance seats are for everyone | Table 3 | CIO, with procurement |
| 7/Put every credit-eligible service on one billing map | One currency is one bill | Exhibit 2, Table 5 | Enterprise architect, with the FinOps lead |
| 8/Set which models each team may use before the pilot gate | The price list sets what a task costs | Exhibit 4, Table 2 | CTO, with program owners |
| 9/Map each feature to the part that pays for it | The seat pays for Copilot | Exhibit 1 | Microsoft 365 service owner, with procurement |
| 10/Put the five meter risks on the board’s risk review | AI spend is a technology risk | Table 5, Exhibit 5 | Audit committee chair, with the chief risk officer |
Table 6 sets out the ten moves of the Greyhound CXO Runbook, each with the assumption it corrects, the exhibit or table it rests on, and the executive with whom the decision sits, and shows that none waits on Microsoft to start while two complete only when Microsoft publishes more terms. The moves are: switch off automatic coverage and write one policy per business unit, budget the credit line as a range, decide what happens at the limit for every critical workflow, buy prepaid credits from measured use and negotiate the credit terms, own the value measure, count the users who need Agent 365 or E7, put every credit-eligible service on one billing map, set which models each team may use before the pilot gate, map each feature to the part that pays for it, and put the five meter risks on the board’s risk review. The moves are Greyhound Research analysis.
The two that finish later, on budgeting and on renewal terms, wait on rates and commitments Microsoft has not yet published. Those are what the questions that follow ask for, and each question maps to a move a buyer can make meanwhile. Every other move rests on an exhibit or table in this note and names who decides. The remaining eight can be finished with settings, records, reviews, and contracts the buyer already holds.
The Deeper Read · About 5 minutes
Select a move to open its full text
1FinOps leads: switch off automatic coverage and write one policy per business unit.
Exhibit 3 and Table 2 show policies that widen to new services by default and reserve no credits. This move corrects the belief that a policy written once stays the size it was written. Its action is to switch off automatic coverage of new services, write a policy for each business unit with its own limit and alerts, and add each new service only after a recorded decision. Customers buying Copilot Business through a partner should review or lower the default limit of 4,000 credits a user before the first purchase in 2027. The decision sits with the FinOps lead, together with the CIO.
Why the order matters. The first move comes first because the later moves depend on it. A range budget, a rule at the limit, and a prepaid order sized from measured use all assume a policy whose scope is known and recorded. A policy that widens by default would leave each later document describing an estate that has already changed.
2CFOs: budget the credit line as a range with a stated top.
Exhibit 4 shows credit costs published as bands, with the heaviest open-ended and three Ignite rates not yet published. This move corrects the habit of forecasting from the light band. Its action is to budget credits as a range, review the mix of light and heavy work each month for two quarters, and book savings or overruns only from measured use. The decision sits with the CFO.
3CIOs: decide what happens at the limit for every critical workflow.
Exhibit 3 shows that a user who reaches a limit loses access for the rest of the month, and that Microsoft has not said what happens to a task in progress. This move corrects the belief that a cap only protects the bill. Its action is to list the workflows that cannot stop, and for each one set in advance whether the limit means a pause, a request for more, or pay-as-you-go spend. The decision sits with the CIO, together with the business owner of each workflow.
4Procurement: buy prepaid credits from measured use and negotiate the credit terms.
Exhibit 6 and Table 4 show prepaid credits that expire after a year and rates Microsoft says can change. This move corrects the assumption that the seat price is the renewal. Its action is to buy P3 only after at least two quarters of measured use, and to seek a held credit rate for the term, a carry-over or credit-back for unused credits, and written notice of any rate or coverage change. The decision sits with the head of procurement, together with the CFO.
5Business leaders: own the value measure.
Exhibit 5 shows a value figure that starts from Microsoft’s estimate of how long work would take a person. This move corrects the belief that the vendor’s dashboard is the measure of return. Its action is to set the value measures for each AI program from the business’s own records and use Microsoft’s Business Analytics as a cross-check. The decision sits with the business leader who owns each program, together with the CFO.
6CIOs: count the users who need Agent 365 or E7, not the headcount.
Table 3 shows that governance of agents built on other platforms needs Agent 365 or E7, while visibility into Microsoft’s own agents does not. This move corrects the habit of buying governance seats for everyone. Its action is to count the people who govern agents on other platforms and license them, and to weigh E7 only where most of its parts are used. A companion Greyhound Research note on agent identity sets out what those seats control. The decision sits with the CIO, together with procurement.
7Enterprise architects: put every credit-eligible service on one billing map at the architecture review.
Exhibit 2 shows one currency set up and watched in three places: the Microsoft 365 admin center, the Power Platform admin center, and an Azure subscription. Table 5 names spend outside the console as a risk, and gives a monthly report that does not reconcile to the Azure invoice as its early warning. This move corrects the belief that one currency means one bill. Its action is to keep a one-page map of the tenant showing, for each service that uses credits, where it bills, which spending policy covers it, and which report shows it. Each new credit-eligible service is placed on that map at the architecture review before anyone may use it. Because the usage report counts credits rather than money, the map also names the rate the finance team applies. The decision sits with the enterprise architect, together with the FinOps lead, at each architecture review.
8CTOs: set which models each team may use before a credit-funded program passes its pilot gate.
Exhibit 4 shows that the same kind of task can sit at either end of its band, depending on how much the agent reads, how many steps it takes, and which model Auto picks. Table 2 lists model access among the controls Microsoft has described for Agent 365 Cost Management, with the credit cost by model not yet written. This move corrects the belief that the price list sets what a task costs. Its action is to set which models each team may use, keep premium models for the work that needs them, and record the mix of light and heavy tasks the pilot produces under those settings. A program passes its pilot gate only once that record exists. The decision sits with the CTO, together with the business owner of each program, at the pilot gate of every program funded by credits.
9Microsoft 365 service owners: map each feature to the part that pays for it, and ask for the split in writing at renewal.
Exhibit 1 shows the line between seat and meter running through the product. Everyday chat and the Today Brief sit in the seat, the Today Feed is metered, and Cowork and Code need a seat and credits. Auto may also recommend moving a user from Chat to Cowork, which crosses from the seat to the meter without anyone making a purchasing decision. This move corrects the belief that the seat pays for Copilot. Its action is to keep a map of each feature the company uses against the part that pays for it, the seat or the credits, and to refresh it each quarter. At renewal, the service owner asks Microsoft for the current split in writing, with notice before any feature moves from the seat to the meter. The decision sits with the Microsoft 365 service owner, together with procurement, at the renewal.
10Audit committee chairs: put the five meter risks on the board’s risk review, each with its named owner.
Table 5 names five risks in a metered estate, each with an early warning in the buyer’s own estate and an owner, and only one of the five sits with the CIO. This move corrects the belief that AI spend is a technology risk. Its action is to add the five risks to the risk register the board already reviews, with the owner and the early warning from Table 5, and to ask each owner to report that warning at the first quarterly spend review. Exhibit 5 adds one test for board papers. Any return figure for an AI program should say whether it was counted from records or estimated by a model, since assisted hours and assisted value are estimates. The decision sits with the chair of the audit committee, together with the chief risk officer, at the next board risk review.
Part 6What To Do, By Role
Questions For Microsoft: What Buyers Will Ask Next
Greyhound SpotlightFour of the six questions have no answer at all in what Microsoft has published, including what happens to a task when the money runs out.
Every finding in this note rests on what Microsoft has published, and Microsoft alone can close the gaps it names. Six questions follow, in the order the evidence raises them. Each gives what Microsoft has said so far, what the answer would settle, and the Runbook move it maps to. No move waits on them, and an account team can answer each from documents Microsoft already holds.
Questions put in writing tend to carry further than questions raised in a briefing. A written reply can be referenced in an order or a renewal schedule, and a verbal assurance from an account team usually cannot. Greyhound Research expects the rate questions to be answered first, because Microsoft needs published rates to sell its new metered services, and the questions on rate protection and the value method later, because those answers bind the vendor. Buyers who ask together, through user groups or partners, tend to draw answers sooner than buyers who ask alone. The questions help even before they are answered, because a documented gap is a fact a buyer can plan around.
Table 7 tracks the six questions against what Microsoft has said so far. Two are answered in part, and four not at all.
Greyhound Research · Table 7 · Questions for Microsoft, tracked
Two questions answered in part, four not at all
The six questions only Microsoft can answer, against what it had stated when re-read on November 19, 2026.
Ask for written answers and plan around the gaps meanwhile. No Runbook move waits on Microsoft.
Select a question to open its full text in The Deeper Read below
| Question | What Microsoft has said so far | The Runbook move it settles |
|---|---|---|
| Q1What happens to a running task when a user or policy reaches its limit? | Not stated Users lose access for the rest of the month and can request more. | Runbook move 3 |
| Q2When will Fabric and Windows 365 for Agents appear in Cost Management and spending policies? | Not stated Windows 365 for Agents costs show in Azure Cost Management. | Runbook move 1 |
| Q3What are the credit rates for the Today Feed, Fabric, and Windows 365 for Agents? | Not stated No rate in the licensing guide. | Runbook move 2 |
| Q4Will credit rates be held for a P3 term and at renewal? | Not stated The licensing guide says pricing is subject to change. | Runbook move 4 |
| Q5How does Business Analytics estimate hours and value, and can customers audit it? | Stated in part Signals named, method not stated. | Runbook move 5 |
| Q6Which Ignite features need Agent 365 or E7, and will that change at general availability? | Stated in part Governance and other platforms’ insight need Agent 365 or E7. | Runbook move 6 |
Table 7 tracks six questions only Microsoft can answer against what it had stated when re-read on November 19, 2026, and shows two answered in part and four not stated at all. Not stated: what happens to a running task at the limit, when Fabric and Windows 365 for Agents join Cost Management, the credit rates for the Today Feed, Fabric, and Windows 365 for Agents, and whether credit rates are held for a term and at renewal. Answered in part: how Business Analytics estimates value, since the signals are named and the method is not, and which features need Agent 365 or E7. Each question names the Runbook move it settles. The grading is Greyhound Research analysis.
Two answers would change this note’s reading most. One is documented behavior when a limit is reached in the middle of a task. The other is a held credit rate through a term and at renewal. Either would move Copilot Credits closer to a cost a CFO can forecast. None of the six is a trick question. Each asks Microsoft to write down what its billing does or will do.
The Deeper Read · About 2 minutes
Select a question to open its full text
Q1What happens to a running task when a user or policy reaches its limit?
Said so far. Not stated. Users lose access for the rest of the month and can request more credits. The answer settles. Whether a critical workflow can be left on a hard limit, under the third Runbook move. Evidence that would settle it. Documentation of task behavior at the limit. Greyhound Research would update. Exhibit 3.
Why the first question is practical. Behavior at the limit sounds like a technical detail, and it decides how a workflow is designed. If a running task finishes, a hard limit is safe for most work. If it stops partway, a critical workflow needs headroom, a fast approval, or pay-as-you-go cover. Until Microsoft documents the behavior, a careful buyer should test it in a pilot and record what it observes.
Q2When will Fabric and Windows 365 for Agents appear in Cost Management and spending policies?
Said so far. Not stated. Windows 365 for Agents costs can be tracked in Azure Cost Management. The answer settles. Whether one policy can govern every credit-eligible service, under the first Runbook move. Evidence that would settle it. An updated coverage list. Greyhound Research would update. Exhibit 2 and Table 2.
Q3What are the credit rates for the Today Feed, Fabric, and Windows 365 for Agents?
Said so far. Not stated. The answer settles. Whether the credit budget can include the Ignite services, under the second Runbook move. Evidence that would settle it. Rates in the licensing guide. Greyhound Research would update. Exhibit 4.
Q4Will credit rates be held for a P3 term and at renewal?
Said so far. Not stated. The licensing guide says pricing is subject to change. The answer settles. Whether a prepaid commitment carries a known price, under the fourth Runbook move. Evidence that would settle it. Rate protection terms in the licensing guide or the enterprise agreement. Greyhound Research would update. Exhibit 6.
Q5How does Business Analytics estimate hours and value, and can customers audit it?
Said so far. In part. Microsoft names the signals: assisted hours, assisted value, success rate, and satisfaction. The method is not stated. The answer settles. How much weight the value figure can carry, under the fifth Runbook move. Evidence that would settle it. Published method documentation. Greyhound Research would update. Exhibit 5.
Q6Which Ignite features need Agent 365 or E7, and will that change at general availability?
Said so far. In part. Enterprise governance and insight into other platforms’ agents need Agent 365 or E7. The answer settles. How many governance seats a buyer needs, under the sixth Runbook move. Evidence that would settle it. A feature-by-license table for the Ignite releases. Greyhound Research would update. Table 3.
Part 7The Final Greyhound Standpoint
Agent Economics: The Final Greyhound Standpoint
Greyhound SpotlightFour signals within twelve months would move this position, and Microsoft controls all four.
Greyhound Research’s final judgment is that Ignite 2026 changed how Microsoft’s AI is paid for more than what it costs. Seat prices held. The meter beside them widened, and Copilot Credits are now one currency across Microsoft 365, Copilot Studio, Windows 365 for Agents, and Fabric. When work is billed by the task, the spending policy becomes the control: who may spend, how much, on which services, and what happens when the money runs out. That control belongs to the buyer, and it is the one part of the AI budget that no price list can supply.
Microsoft deserves credit on several counts. It is ahead of its rivals in publishing what a task costs and level with them in pairing seats with a meter. Its FinOps for AI tools are the right tools, mostly in preview, and its move to judge AI by completed work rather than tokens asks the right question. The open points are coverage, which stops short of Fabric and Windows 365 for Agents, limits that stop work rather than forecast it, and prepaid credits that expire. Companion Greyhound Research notes on Azure agent infrastructure and confidential computing reached their verdicts on the same week’s news, and this note agrees with both.
The pattern is one enterprise buyers have met at each major shift in how technology is sold. The moves from perpetual licenses to subscriptions, and from owned hardware to cloud consumption, each opened a period in which the price model ran ahead of the buyer’s controls. In each, the organizations that fared best wrote their own rules for ownership, forecasting, and commitment before the bills grew. Metered agents are at the start of that period. The difference this time is pace, because one default setting can bring a new service under an existing policy in the month it is released.
The evidence has limits. Several Ignite credit rates are not published, Cost Management is in preview, and Business Analytics is in Frontier. Microsoft says its pricing is subject to change, and the comparison with Salesforce and Google rests on each company’s own pages. Statuses, prices, and terms are as published, re-read on November 19, 2026. Greyhound Research’s own reading of credits has moved with the evidence. In November 2025 it described such units as obscuring value. Microsoft’s published rate, bands, and limits have answered much of that, so the question now is predictability rather than opacity.
Four signals would move this position within twelve months, and each would make credit spending easier to forecast. They are published rates for every credit-eligible service, documented task behavior at the limit, a held credit rate through a term, and Cost Management coverage of every service that uses credits. Any one of them would ease the caution in this note.
Exhibit 8 sets those signals side by side with what each would change. All four sit with Microsoft.
Greyhound Research · Exhibit 8 · Signals that would move the position
Four signals in twelve months would move this position
What would move Greyhound Research’s judgment within twelve months, and what each would change.
Of the four signals, a held credit rate would move this view furthest. Ask for it at the next renewal.
Exhibit 8 lists the four signals that would move Greyhound Research’s position within twelve months, each controlled by Microsoft, and one development outside its control. A held credit rate through a prepaid term and at renewal would turn a prepaid purchase into a known price, the signal that matters most. Documented behavior at the limit would let critical workflows run under a hard limit. Published rates for the Today Feed, Fabric, and Windows 365 for Agents would let the credit budget include the Ignite services. Full Cost Management coverage would make one currency one bill. A rival changing the rules at its cap or the expiry of its prepaid units could reset the comparison. The signals are Greyhound Research analysis.
One development outside Microsoft’s control could reset the comparison: a rival changing the rules at its cap or the expiry of its prepaid units. Of the four signals, a held credit rate would matter most, because it turns a prepaid purchase into a known price. Until the signals arrive, the actions sit with named owners at known decision points. FinOps leads should write spending policies, with automatic coverage switched off, before the next service joins the meter. Chief financial officers should budget the credit line as a range in the coming planning round. Procurement should buy prepaid credits from measured use and, at the next renewal, negotiate the credit as hard as the seat.
Part 8Appendix, Sources, And Disclosures
Appendix: The Evidence Register, Source By Source
Greyhound Research sets out here the evidence behind every fact, figure, and quotation in this research note, one drop-down for each class of source, so that any sentence can be checked against the words it rests on without leaving the page. Each row gives the source, when it was read, the part of the source the words sit in, what it says, and where in this research note that material is used. Times of reading are in UTC. Every source was read or re-read on November 19, 2026, between 00:28 and 00:39 UTC. Where a passage was summarized and not quoted, the row says so. A passage quoted in The Deeper Read of a section is not repeated here, and the source’s entry names the section that quotes it.
Microsoft announcements and blogs7 sources, 56 extracts
| Source, read, and used | Where in the source | What it says |
|---|---|---|
| Microsoft (2026a)Read November 18, 2026, and re-read November 19, 2026, between 00:28 and 00:39 UTCUsed in What Microsoft Announced, Two Ways To Pay, One Currency, Several Bills, FinOps For AI, The License Gate. Tables 1, 2, and 3. Exhibits 1 and 4Quoted in The Deeper Read of What Microsoft Announced, FinOps For AI, Two Ways To Pay, The License Gate | FinOps for AI | [quoted in The Deeper Read] |
| FinOps for AI | Agent 365 Cost Management is described as "a centralized place to see and control a growing list of usage-based AI services" (fragment). | |
| FinOps for AI | Covered services listed as Copilot Cowork, Copilot Code, GitHub Copilot CLI, Microsoft Copilot Studio, Copilot Managed Runtime, Autopilot, and Work IQ APIs (summarized, not quoted). | |
| FinOps for AI | [quoted in The Deeper Read] The bullet is marked in preview. | |
| FinOps for AI | Forecasting, plain-language questions, and optimization recommendations are available in the Frontier program this month (summarized, not quoted). | |
| FinOps for AI | [quoted in The Deeper Read] | |
| FinOps for AI | [quoted in The Deeper Read] | |
| FinOps for AI | Insights for non-Microsoft services require an Agent 365 license (summarized, not quoted). | |
| Today | [quoted in The Deeper Read] | |
| Today | [quoted in The Deeper Read] | |
| Today | "Today will begin rolling out in private preview through the Frontier program." | |
| Management insights | Usage insights for Microsoft agents are available regardless of Agent 365 licensing status, generally available (summarized, not quoted). | |
| Management insights | [quoted in The Deeper Read] | |
| Management insights | [quoted in The Deeper Read] | |
| Management insights | [quoted in The Deeper Read] | |
| Windows 365 for Agents | "Windows 365 for Agents is now eligible for Copilot Credits" | |
| Windows 365 for Agents | [quoted in The Deeper Read] | |
| Windows 365 for Agents | The eligibility enables a unified AI billing, governance, reporting, and cost-management experience (summarized, not quoted). | |
| FinOps for AI | The December expansion also adds delegation to business leaders (summarized, not quoted). | |
| Reading note | Read through an extraction layer. Quotations re-checked against the live page on November 19, 2026. No credit rate for any Ignite service appears in the text read. | |
| Althoff, J. (2026)Read November 18, 2026, and re-read November 19, 2026, between 00:28 and 00:39 UTCUsed in What Microsoft Announced, The License Gate. Table 1Quoted in The Deeper Read of What Microsoft Announced, The License Gate | Post | [quoted in The Deeper Read] |
| Post | [quoted in The Deeper Read] | |
| Post | [quoted in The Deeper Read] | |
| Post | [quoted in The Deeper Read] | |
| Post | "It includes Microsoft Entra Suite and advanced Defender, Intune and Purview security capabilities" | |
| Footnote | Microsoft 365 E7 is available with and without Teams (summarized, not quoted). | |
| Reading note | The post does not mention Copilot Credits or usage-based pricing. | |
| Davuluri, P. (2026)Read November 18, 2026, and re-read November 19, 2026, between 00:28 and 00:39 UTCUsed in One Currency, Several Bills. Table 1Quoted in The Deeper Read of One Currency, Several Bills | Summary | "Windows 365 for Agents, now eligible for Copilot Credits" |
| Post | [quoted in The Deeper Read] | |
| Post | "persistent agent Cloud PCs for long-running, always-on work are coming to Frontier preview." | |
| Reading note | The post gives no credit rate or price. | |
| Holeček, A. (2026)Read November 18, 2026, and re-read November 19, 2026, between 00:28 and 00:39 UTCUsed in One Currency, Several BillsQuoted in The Deeper Read of One Currency, Several Bills | Post | [quoted in The Deeper Read] |
| Reading note | The post gives no list price for Microsoft Perception. | |
| Microsoft (2026b)Read November 18, 2026, and re-read November 19, 2026, between 00:28 and 00:39 UTCUsed in Predictability, FinOps For AI, Value Scored By The Seller, Three Vendors Compared. Tables 2 and 3. Exhibit 5Quoted in The Deeper Read of FinOps For AI, Predictability, Value Scored By The Seller | Cost Management | [quoted in The Deeper Read] |
| Cost Management | "Track consumption across users, services, agents, and models to understand where AI investment is going." | |
| Cost Management | [quoted in The Deeper Read] | |
| Cost Management | [quoted in The Deeper Read] | |
| Cost Management | Microsoft Graph APIs, data exports, and Fabric connectors integrate AI spend into reporting, budgeting, and chargeback (summarized, not quoted). | |
| Business Analytics | [quoted in The Deeper Read] | |
| Business Analytics | "optimize investments with spend forecasts and scenario planning." | |
| Business Analytics | [quoted in The Deeper Read] | |
| Business Analytics | Analytics for connected non-Microsoft AI services require Agent 365 licensing, ME7 licensing, or both (summarized, not quoted). | |
| Scale | "Organizations have registered more than 50 million agents in Microsoft Agent 365 since its May 2026 launch." Not used in the analysis. | |
| Reading note | The post gives no Agent 365 price and does not mention Copilot Credits. The author is shown by a Microsoft username. | |
| Spataro, J. (2026)Read November 18, 2026, and re-read November 19, 2026, between 00:28 and 00:39 UTCUsed in Two Ways To Pay, Pooling And Caps, Predictability, Value Scored By The SellerQuoted in The Deeper Read of Two Ways To Pay, Predictability, Pooling And Caps, Value Scored By The Seller | Commercial model | [quoted in The Deeper Read] |
| Commercial model | [quoted in The Deeper Read] | |
| Commercial model | "And as the Frontier advances, we'll continue to move more and more capability into the subscription license." | |
| Auto | [quoted in The Deeper Read] | |
| Auto | [quoted in The Deeper Read] | |
| Cowork | [quoted in The Deeper Read] | |
| Business Analytics | [quoted in The Deeper Read] (fragment) | |
| Business Analytics | It analyzes the tasks, how long that work typically takes a person, and the share Cowork performed (summarized, not quoted). | |
| Reading note | The post cites a third-party cost comparison against rival assistants. Greyhound Research does not use it. The post gives no prices. | |
| Ulag, A. (2026)Read November 18, 2026, and re-read November 19, 2026, between 00:28 and 00:39 UTCUsed in What Microsoft Announced, One Currency, Several Bills. Table 1. Exhibit 1Quoted in The Deeper Read of What Microsoft Announced | Post | [quoted in The Deeper Read] |
| Post | [quoted in The Deeper Read] | |
| Reading note | The post gives no rate or release status for Copilot Credits in Fabric. |
Microsoft documentation, licensing, and prices12 sources, 54 extracts
| Source, read, and used | Where in the source | What it says |
|---|---|---|
| Microsoft (2026c)Read November 18, 2026, and re-read November 19, 2026, between 00:28 and 00:39 UTCUsed in Two Ways To Pay, One Currency, Several Bills, Pooling And Caps, The Strongest Objection. Exhibits 1, 2, and 3. Table 2Quoted in The Deeper Read of One Currency, Several Bills, Pooling And Caps | Overview | [quoted in The Deeper Read] |
| Billing | Organizations can use fixed licensing, P3, pay-as-you-go billing, prepaid capacity packs, or supported combinations (summarized, not quoted). | |
| Billing | Administrators connect an Azure subscription to support billing (summarized, not quoted). | |
| Services | Cowork, advanced work in SharePoint and OneDrive, Work IQ API, Copilot Managed Runtime, Teams Phone Agent, and Code are listed, several requiring a Copilot license (summarized, not quoted). | |
| Services | [quoted in The Deeper Read] | |
| Policies | [quoted in The Deeper Read] | |
| Reporting | [quoted in The Deeper Read] | |
| Reading note | Updated November 3, 2026. Copilot Studio is billed through the Power Platform admin center unless opted into the Microsoft 365 admin center. | |
| Microsoft (2026d)Read November 18, 2026, and re-read November 19, 2026, between 00:28 and 00:39 UTCUsed in Pooling And Caps, Predictability, The Strongest Objection. Exhibit 3Quoted in The Deeper Read of Pooling And Caps, The Strongest Objection | Limits | [quoted in The Deeper Read] |
| Limits | [quoted in The Deeper Read] | |
| Policies | [quoted in The Deeper Read] | |
| Billing order | "The policy always uses prepaid credits first" (fragment), before pay-as-you-go. | |
| Reset | "credits reset on the first of the month." | |
| Set limits and alerts | [quoted in The Deeper Read] | |
| Set limits and alerts | A policy must meet a required minimum, and recommendations and policy requirements may evolve. The page does not say which services the minimum covers (summarized, not quoted). | |
| Reading note | Updated November 11, 2026. The page says recommendations and policy requirements may evolve. | |
| Microsoft (2026e)Read November 18, 2026, and re-read November 19, 2026, between 00:28 and 00:39 UTCUsed in One Currency, Several Bills, Renewal Leverage. Exhibits 2 and 6Quoted in The Deeper Read of One Currency, Several Bills, Renewal Leverage | Packs | [quoted in The Deeper Read] |
| Packs | At the start of every monthly billing period, the 25,000 credits are replenished (summarized, not quoted). | |
| Policies | [quoted in The Deeper Read] | |
| Exhaustion | [quoted in The Deeper Read] | |
| Reading note | Updated November 3, 2026. The page gives no price for a pack. | |
| Microsoft (2026f)Read November 18, 2026, and re-read November 19, 2026, between 00:28 and 00:39 UTCUsed in One Currency, Several Bills. Exhibit 2Quoted in The Deeper Read of One Currency, Several Bills | Pricing model | [quoted in The Deeper Read] |
| Cost monitoring | Costs can be tracked in Azure Cost Management with the Windows365foragents tag (summarized, not quoted). | |
| Reading note | Updated June 3, 2026. The page does not mention Copilot Credits, which the Ignite announcement added later. | |
| Microsoft (2026g)Read November 18, 2026, and re-read November 19, 2026, between 00:28 and 00:39 UTCUsed in Two Ways To PayQuoted in The Deeper Read of Two Ways To Pay | Page | [quoted in The Deeper Read] |
| Page | "Your organization sets limits for how many credits you can use for those advanced AI tasks." | |
| Microsoft (2026h)Read November 18, 2026, and re-read November 19, 2026, between 00:28 and 00:39 UTCUsed in Pooling And Caps. Table 2Quoted in The Deeper Read of Pooling And Caps | Notice | [quoted in The Deeper Read] |
| Notice | "These updates will begin rolling out in early September 2026." | |
| Reading note | Read through a public mirror of the Microsoft 365 Message Center, mc.merill.net, because the Message Center requires a tenant sign-in. | |
| Microsoft (2026i)Read November 19, 2026, between 00:28 and 00:39 UTCUsed in Research Note In Two Minutes, Pooling And Caps, The Greyhound CXO RunbookQuoted in The Deeper Read of Pooling And Caps | November 3 availability update | "Usage-based billing is now enabled by default for new Microsoft 365 Copilot Business subscriptions purchased in Cloud Solution Provider (CSP) during Microsoft's FY27 Q3 (January-March 2027), instead of December 1, 2026, as previously announced." |
| November 3 availability update | [quoted in The Deeper Read] | |
| November 3 availability update | "Disabling usage-based billing might limit access to AI experiences." | |
| November 3 availability update | Microsoft creates the required Azure resources and applies the default limit to all users in the customer organization. Limits can be adjusted in Partner Center or the Microsoft admin center (summarized, not quoted). | |
| Reading note | Page last updated November 13, 2026. The November 2 date on the same page refers to Partner Center sandbox environments, not to usage-based billing. | |
| Microsoft (2026j)Read November 19, 2026, between 00:28 and 00:39 UTCUsed in Pooling And Caps | Setup | "This process activates AI experiences, like Cowork, that are gated by usage-based billing." |
| Setup | "When you activate the default spending policy, you set the tenant-level limit for all users." (formatting removed) | |
| Reading note | Updated November 11, 2026. The page states no default credit limit and does not say in so many words that billing is off until activated. | |
| Microsoft (2026k)Read November 19, 2026, between 00:28 and 00:39 UTCUsed in Renewal LeverageQuoted in The Deeper Read of Renewal Leverage | MACC | [quoted in The Deeper Read] |
| MACC | MACC benefits apply only when the selected subscription links to the billing account that holds the commitment (summarized, not quoted). | |
| Reading note | Updated November 3, 2026. The page does not state the CSP default or the 4,000-credit limit. | |
| Microsoft (n.d.a)Read November 18, 2026, and re-read November 19, 2026, between 00:28 and 00:39 UTCUsed in One Currency, Several Bills, Two Ways To Pay, Pooling And Caps, Predictability, Renewal Leverage, Three Vendors Compared, The Strongest Objection. Tables 1 and 4. Exhibits 3, 4, and 6Quoted in The Deeper Read of One Currency, Several Bills, Renewal Leverage, Pooling And Caps, Predictability, Two Ways To Pay | Pay-as-you-go | [quoted in The Deeper Read] Billed in arrears at the end of the billing month. |
| P3 | [quoted in The Deeper Read] | |
| P3 | [quoted in The Deeper Read] | |
| P3 | Nine tiers from 300,000 credits at 5% off to 300,000,000 credits at 20% off (summarized, not quoted). | |
| Pooling | [quoted in The Deeper Read] | |
| Overage | "If usage exceeds the available number of credits, customers can purchase additional Copilot Credits." | |
| Rates | Cowork tasks 100 to 300, 300 to 700, and more than 700 credits. Copilot Code 25 to 60, 75 to 100, and more than 200. Copilot Studio runtime in the GitHub Copilot harness 100 to 300, 300 to 500, and more than 500. Teams Phone Agent 5 to 15, 15 to 40, and more than 40 (summarized, not quoted). | |
| Rates | [quoted in The Deeper Read] | |
| Harnesses | [quoted in The Deeper Read] | |
| Terms | [quoted in The Deeper Read] | |
| Reading note | The page shows no publication date. It gives no credit rate for the Today Feed, Fabric, or Windows 365 for Agents. A second URL for the same guidance returned an error when read. | |
| Microsoft (n.d.b)Read November 18, 2026, and re-read November 19, 2026, between 00:28 and 00:39 UTCUsed in What Microsoft Announced, The License Gate. Table 1 | Plans | Microsoft 365 E7 USD 99.00 per user a month with Teams and USD 90.45 without, paid yearly, annual commitment (summarized, not quoted). |
| Microsoft (n.d.c)Read November 18, 2026, and re-read November 19, 2026, between 00:28 and 00:39 UTCUsed in What Microsoft Announced. Tables 1 and 4 | Add-ons | Microsoft 365 Copilot USD 30.00 and Agent 365 USD 15.00 per user a month, paid yearly, annual subscription that auto-renews, each included in Microsoft 365 E7 (summarized, not quoted). |
| Add-ons | Copilot Studio is listed as pay-as-you-go with a capacity license starting at 25,000 messages a month (summarized, not quoted). |
Salesforce, Google, and a Microsoft partner3 sources, 16 extracts
| Source, read, and used | Where in the source | What it says |
|---|---|---|
| Fernandes, N. (2026)Read November 19, 2026, on the publisher’s pageUsed in Pooling And CapsQuoted in The Deeper Read of Pooling And Caps | Post | [quoted in The Deeper Read] |
| Post | "It is unmanaged consumption without clear ownership, guardrails or outcome measurement." | |
| Post | The CSP default is a spending policy capped at 4,000 Copilot Credits per user per month, which only an administrator can raise, and the start moved from December 1, 2026, to January to March 2027 (summarized, not quoted). | |
| Reading note | Sherweb is a Microsoft Cloud Solution Provider distributor. The author is its product manager for Microsoft 365, Copilot, and Google Workspace. | |
| Gerstenhaber, M. and Gupta, P. (2026)Read November 18, 2026, and re-read November 19, 2026, between 00:28 and 00:39 UTCUsed in Three Vendors Compared. Table 4Quoted in The Deeper Read of Three Vendors Compared | Caps | [quoted in The Deeper Read] |
| Caps | Alerts at 50%, 80%, and 100% of the budget. Work can resume with one click, or overages can be allowed at pay-as-you-go rates (summarized, not quoted). | |
| Pooling | [quoted in The Deeper Read] | |
| Savings plans | [quoted in The Deeper Read] | |
| Pay-as-you-go | Available for select customers and rolling out broadly, with no upfront commitment or base fee (summarized, not quoted). | |
| Reading note | Title corrected on November 19, 2026, to the heading on the page. The post gives no per-seat price and does not say whether unused savings-plan commitments lapse. | |
| Salesforce (n.d.)Read November 18, 2026, and re-read November 19, 2026, between 00:28 and 00:39 UTCUsed in Three Vendors Compared. Table 4Quoted in The Deeper Read of Three Vendors Compared | Flex Credits | USD 500 per 100,000 credits. Agentforce actions use 20 credits, voice actions 30 (summarized, not quoted). |
| Flex Credits | Credits "can be applied across teams, channels, and use cases." (fragment) | |
| Flex Credits | [quoted in The Deeper Read] | |
| Flex Credits | [quoted in The Deeper Read] | |
| Seats | Agentforce add-ons at USD 125 per user a month with unmetered employee agent usage (summarized, not quoted). | |
| Buying models | Pre-Commit with a true-up at term end, described as widely available later this year (summarized, not quoted). |
Greyhound Research12 sources, 19 extracts
| Source, read, and used | Where in the source | What it says |
|---|---|---|
| Greyhound Research (2025a)Read November 18, 2026, and re-read November 19, 2026, on greyhoundresearch.comUsed in Predictability, Renewal LeverageQuoted in The Deeper Read of Predictability, Renewal Leverage | Note | [quoted in The Deeper Read] |
| Note | [quoted in The Deeper Read] | |
| Note | Microsoft metered advanced agents at roughly USD 0.01 per agent query during preview (summarized, not quoted). | |
| Greyhound Research (2025b)Read November 18, 2026, and re-read November 19, 2026, on greyhoundresearch.comUsed in Predictability, Value Scored By The SellerQuoted in The Deeper Read of Predictability | Greyhound Flashpoint | [quoted in The Deeper Read] (Greyhound CIO Pulse 2025). |
| Greyhound Pulse | 58% of Fortune 1000 CIOs in North America have paused or scaled back agent pilots, with 41% reporting overspend of 25% or more (Greyhound CIO Pulse 2025, summarized, not quoted). | |
| Greyhound Standpoint | Procedural memory introduces a two-tier cost model: learn with the largest model, execute with the smallest viable one (summarized, not quoted). | |
| Greyhound Research (2025c)Read November 19, 2026, on the publisher’s pageUsed in Predictability | Commentary | "What was once a predictable licensing model is being replaced by vague units: credits, interactions, events, deliberately designed to obscure value exchange." |
| Reading note | Greyhound Research commentary as published by a third party on November 26, 2025. | |
| Greyhound Research (2026a)Read November 18, 2026, and re-read November 19, 2026, on greyhoundresearch.comUsed in Three Vendors Compared | Note | Microsoft’s Dynamics 365 documentation names both bills for one task on one page, and the note uses that wording as the standard for how the cost of one task should be written (summarized, not quoted). |
| Greyhound Research (2026b)Read November 18, 2026, and re-read November 19, 2026, on greyhoundresearch.comUsed in Renewal Leverage | Note | Consolidating on one suite can remain right while its standard terms fall short (summarized, not quoted). |
| Greyhound Research (2026c)Read November 18, 2026, and re-read November 19, 2026, on greyhoundresearch.comUsed in The Final Greyhound Standpoint, Predictability, Agent Economics | Note | Context Cache saves money only where agents reread the same long text, and savings should be booked only from measured reuse (summarized, not quoted). |
| Greyhound Research (2026d)Read November 18, 2026, and re-read November 19, 2026, on greyhoundresearch.comUsed in The Final Greyhound Standpoint, Agent Economics | Note | Companion note in the same series on Microsoft confidential computing at Ignite 2026 (summarized, not quoted). |
| Greyhound Research (2026e)Read November 19, 2026, on greyhoundresearch.comUsed in Value Scored By The Seller | Note | "Greyhound CIO Pulse 2026 finds that just 32 per cent of enterprise technology leaders say a majority of their materially funded AI initiatives have delivered measurable business outcomes against their approved business cases, based on responses from more than 1,000 leaders." |
| Greyhound Research (2026f)Read November 19, 2026, on the publisher’s pageUsed in FinOps For AI | Commentary | "The dashboards do not lower the bill." "The architecture lowers the bill." |
| Reading note | Greyhound Research commentary as published by a third party on April 28, 2026. | |
| Greyhound Research (2026g)Read November 19, 2026, on the publisher’s pageUsed in Two Ways To Pay | Commentary | "Microsoft has kept a $30 Copilot licence carrying no entitlement to its agentic layer at all." |
| Reading note | Greyhound Research commentary as published by a third party on August 11, 2026. | |
| Greyhound Research (2026h)Read November 19, 2026, on greyhoundresearch.comUsed in Two Ways To Pay | Note | Companion note in the same series on Copilot as the front door for work at Ignite 2026 (summarized, not quoted). |
| Greyhound Research (2026i)Read November 19, 2026, on greyhoundresearch.comUsed in The License Gate | Note | Companion note in the same series on agent identity at Ignite 2026 (summarized, not quoted). |
Part 8Appendix, Sources, And Disclosures
Sources And References: Every Document Behind Every Claim
Every fact, figure, and quotation in this research note traces to one of the sources below, presented in Harvard reference format and grouped by class. Each source is also linked in the sentence that carries it. Microsoft’s, Salesforce’s, Google’s, and Sherweb’s documents are each company’s account of its own products, prices, and terms, and prove what the company said rather than that the thing is true. Standing pages that carry no publication date are marked n.d. All sources were read or re-read on November 19, 2026.
Microsoft announcements and blogs
Althoff, J. (2026) Introducing the First Frontier Suite built on Intelligence + Trust. The Official Microsoft Blog, March 9. Available at: https://blogs.microsoft.com/blog/2026/03/09/introducing-the-first-frontier-suite-built-on-intelligence-trust/
Davuluri, P. (2026) Microsoft Ignite 2026: Building Windows for the next era of work. Windows Experience Blog, November 17. Available at: https://blogs.windows.com/windowsexperience/2026/11/17/microsoft-ignite-2026-building-windows-for-the-next-era-of-work/
Holeček, A. (2026) Security for the agentic era, powered by Microsoft Perception. Microsoft Security Blog, November 17. Available at: https://www.microsoft.com/en-us/security/blog/2026/11/17/security-for-the-agentic-era-powered-by-microsoft-perception/
Microsoft (2026a) Microsoft Ignite 2026 | Book of News. Microsoft News, November 17. Available at: https://news.microsoft.com/events/ignite/2026/book-of-news/
Microsoft (2026b) What’s new in Microsoft Agent 365: Ignite 2026. Microsoft Tech Community, Agent 365 Blog, November 17. Available at: https://techcommunity.microsoft.com/blog/agent-365-blog/whats-new-in-microsoft-agent-365-ignite-2026/4571753
Spataro, J. (2026) Copilot at Microsoft Ignite: Expanding what’s possible with intelligence built for work. Microsoft Copilot Blog, November 17. Available at: https://www.microsoft.com/en-us/copilot/blog/2026/11/17/copilot-at-microsoft-ignite-expanding-whats-possible-with-intelligence-built-for-work/
Ulag, A. (2026) Microsoft Ignite 2026: What’s new in Azure, the platform purpose-built for AI. Microsoft Azure Blog, November 17. Available at: https://azure.microsoft.com/en-us/blog/microsoft-ignite-2026-whats-new-in-azure-the-platform-purpose-built-for-ai/
Microsoft documentation, licensing, and prices
Microsoft (2026c) Understand usage-based billing and cost management for Copilot Credits. Microsoft Learn, updated November 3. Available at: https://learn.microsoft.com/en-us/microsoft-365/copilot/usage-based-billing-overview-copilot-credits
Microsoft (2026d) Managing AI experiences enabled by usage-based billing. Microsoft Learn, updated November 11. Available at: https://learn.microsoft.com/en-us/microsoft-365/copilot/usage-based-billing-manage-copilot-credits
Microsoft (2026e) Use Copilot Studio capacity packs for Microsoft Copilot Chat and SharePoint agents. Microsoft Learn, updated November 3. Available at: https://learn.microsoft.com/en-us/microsoft-365/copilot/pay-as-you-go/copilot-capacity-packs
Microsoft (2026f) Set up billing for Windows 365 for Agents. Microsoft Learn, updated June 3. Available at: https://learn.microsoft.com/windows-365/agents/billing-w365a
Microsoft (2026g) Understand usage-based billing for AI at work. Microsoft Support, October 16. Available at: https://support.microsoft.com/microsoft-365-copilot/understand-usage-based-billing-ai-at-work
Microsoft (2026h) MC1470425, Microsoft Copilot Cost Management: September Updates. Microsoft 365 Message Center, September 10, as mirrored by mc.merill.net. Available at: https://mc.merill.net/message/MC1470425
Microsoft (2026i) Availability update: Updates to usage-based billing for new Microsoft 365 Copilot Business licenses and Partner Center sandbox environments. Microsoft Partner Center announcements, November 3. Available at: https://learn.microsoft.com/en-us/partner-center/announcements/2026-november
Microsoft (2026j) Set up usage-based billing for Copilot Credits. Microsoft Learn, updated November 11. Available at: https://learn.microsoft.com/en-us/copilot/implement/usage-based-billing-copilot-credits-setup
Microsoft (2026k) Usage-based-billing guidance for CSPs, partner-managed customers, and MACC. Microsoft Learn, updated November 3. Available at: https://learn.microsoft.com/en-us/copilot/implement/usage-based-billing-copilot-credits-csp-partner-macc
Microsoft (n.d.a) Copilot Credits Licensing Guidance. Microsoft Licensing Resources. Available at: https://www.microsoft.com/licensing/guidance/CopilotCreditsGuide
Microsoft (n.d.b) Microsoft 365 Enterprise plans and pricing. Microsoft. Available at: https://www.microsoft.com/en-us/microsoft-365/enterprise/microsoft365-plans-and-pricing
Microsoft (n.d.c) Additional services plans and pricing. Microsoft. Available at: https://www.microsoft.com/en-us/microsoft-365/enterprise/additional-services-plans-and-pricing
Salesforce, Google, and a Microsoft partner
Fernandes, N. (2026) Usage-based billing for Microsoft 365 Copilot Business: What MSPs need to know. Sherweb blog, October 16, updated November 5. Available at: https://www.sherweb.com/blog/microsoft-ecosystem/copilot/copilot-business-usage-based-billing/
Gerstenhaber, M. and Gupta, P. (2026) FinOps for the AI era: New flexible billing and cost controls for agents. Google Cloud Blog, August 26. Available at: https://cloud.google.com/blog/products/ai-machine-learning/flexible-billing-and-cost-controls-for-agents-on-google-cloud
Salesforce (n.d.) Agentforce Pricing. Salesforce. Available at: https://www.salesforce.com/ca/agentforce/pricing/
Research this note builds on
Greyhound Research (2025a) How Microsoft’s Researcher And Analyst Copilot Agents Are Transforming Enterprise Workflows With AI Reasoning. Research note, July 14. Available at: https://greyhoundresearch.com/microsofts-copilot-gets-a-brain-what-researchers-and-analysts-mean-for-enterprise-workflows/
Greyhound Research (2025b) Transforming AI Economics with Procedural Memory. Research note, August 30. Available at: https://greyhoundresearch.com/transforming-ai-economics-with-procedural-memory/
Greyhound Research (2025c) Published commentary on new software pricing metrics. November 26. Available at: https://www.cio.com/article/4097012/new-software-pricing-metrics-will-force-cios-to-change-negotiating-tactics.html
Greyhound Research (2026a) The Front Door Race – Oracle, OpenAI And Five Rivals, And Why The Permission Becomes The Control. Research note, November 5. Available at: https://greyhoundresearch.com/oracle-openai-the-front-door-race-the-permission-becomes-the-control/
Greyhound Research (2026b) Microsoft 365 Outage – When The Suite Goes Dark, Restoration Order Becomes The Control. Research note, November 15. Available at: https://greyhoundresearch.com/microsoft-365-outage-when-the-suite-goes-dark-restoration-order-becomes-the-control/
Greyhound Research (2026c) Microsoft Azure Agent Infrastructure – When Agents Get Their Own Machines, Authority Becomes The Control. Research note, November 2026. Available at: https://greyhoundresearch.com/microsoft-azure-agent-infrastructure-when-agents-get-their-own-machines-authority-becomes-the-control/
Greyhound Research (2026d) Microsoft Confidential Computing – When Agents Work Inside The Seal, Attestation Becomes The Control. Research note, November 2026. Available at: https://greyhoundresearch.com/microsoft-confidential-computing-when-agents-work-inside-the-seal-attestation-becomes-the-control/
Greyhound Research (2026e) Panic Has A Shelf Life: Why Greyhound Research Remains Bullish On IBM After July 14. Research note, July 17. Available at: https://greyhoundresearch.com/panic-has-a-shelf-life-why-greyhound-research-remains-bullish-on-ibm-after-july-14/
Greyhound Research (2026f) Published commentary on GitHub Copilot usage-based billing. April 28. Available at: https://www.infoworld.com/article/4164236/github-shifts-copilot-to-usage-based-billing-signaling-new-cost-model-for-enterprise-ai-tools.html
Greyhound Research (2026g) Published commentary on premium ChatGPT Business seats. August 11. Available at: https://www.cio.com/article/4207908/openai-targets-heavy-users-with-premium-chatgpt-business-seats.html
Greyhound Research (2026h) Microsoft Copilot As The OS For Work – When The Front Door Moves, Frontier Enrollment Becomes The Control. Research note, November 2026. Available at: https://greyhoundresearch.com/microsoft-copilot-as-the-os-for-work-when-the-front-door-moves-frontier-enrollment-becomes-the-control/
Greyhound Research (2026i) Microsoft Agent Identity – When Rivals’ Agents Get A Microsoft Name, Enforcement Becomes The Control. Research note, November 2026. Available at: https://greyhoundresearch.com/microsoft-agent-identity-when-rivals-agents-get-a-microsoft-name-enforcement-becomes-the-control/
Part 8Appendix, Sources, And Disclosures
Disclosures: Independence, Financial, Intended Use, Methodology, Authoring, And Copyright
The evidence in this research note is current through November 19, 2026. It was first read on November 18, the day after Microsoft’s Ignite announcements, and every source was re-read between 00:28 and 00:39 UTC on November 19, 2026. Any written response received from an organization named in it after that time is dated where it appears.
Independence Disclosure
Research under the Greyhound Research brand is independent, and no organization holds a right of approval over its content or publication. It is not commissioned, sponsored, or funded by any organization it covers. Greyhound Research may send a research note to companies named in it for factual review before publication. Any such review is confined to the correction of fact, and no viewpoint, conclusion, or recommendation is guided by those companies. All insights and recommendations are Greyhound Research’s alone, and editorial control remains entirely with Greyhound Research. Greyhound Consilium is the B2B thought-leadership arm of The House Of Greyhound and a sister brand to Greyhound Research. It builds on Greyhound Research’s independent research through collaborations that produce thought-leadership content, research, and practical toolkits for decision-makers. Every Greyhound Consilium publication names its partner on the face of the document. This research note is published under the Greyhound Research brand.
Financial Disclosure
Greyhound Research undertakes advisory and consulting engagements with enterprise buyers and technology providers, and no such engagement confers control over its research findings or its decision to publish them. For this research note, Greyhound Research has no financial interest in, and no commercial relationship with, Microsoft, Amazon Web Services, Google, Anthropic, or OpenAI to declare. Oracle is a client of Greyhound Research. Oracle is not a subject of this research note.
Intended Use
This research note is written for chief financial officers, FinOps leads, procurement and vendor management leaders, chief information officers, business leaders who own AI programs, board members, and the press. It reads Microsoft’s pricing, licensing, and cost management announcements at Microsoft Ignite 2026, the documentation and price lists beside them, and the published pricing of Salesforce and Google, to inform decisions on AI budgets, spending policies, value measures, license counts, and renewals. This research note is not investment advice, a recommendation to buy, sell, or hold any security, or a view on any company’s share price or valuation. It is not legal or tax advice, and its reading of contract terms is to be confirmed with counsel. Its assessments rest on the documents listed in Sources And References as they stood on November 19, 2026, and can change as the companies named publish more. Product availability, status, features, and prices are described as each company published them through November 19, 2026, so confirm them with the vendor before acting on them.
Research Methodology
Greyhound Research’s analysis combines lived experience with public sources. The lived experience is years of continuously tracking the market, Greyhound Fieldnotes from ongoing advisory engagements with enterprise buyers and with technology providers, and Greyhound Pulse, the firm’s market trackers. Lived experience forms the judgment and never stands in for evidence. The public sources are the thirty-four documents listed in Sources And References, in four classes, and every factual claim is verified against the document that supports it. The passages relied on are quoted in The Deeper Read of the section that uses them, and the other words taken from each source, the passages summarized rather than quoted and the reading notes, are set out in the appendix. A finding that something is not stated or not documented is bounded to the documents the text names, as they stood when re-read on November 19, 2026. Statuses are given as each company states them. Where this note builds on or moves from a position Greyhound Research has published before, it cites that position and explains any change of view where it appears.
Microsoft’s licensing guidance on Copilot Credits carries no publication date and says its pricing is subject to change. Microsoft’s own post on the evolution of its Copilot pricing model, dated September 2026, did not render when read, so this note rests on Microsoft Learn, Microsoft Support, the licensing guidance, and the Message Center notice of September 10, 2026, read through a public mirror. Credit costs are given in credits as Microsoft lists them, at its published rate of USD 0.01 per credit, and are not converted into money per task. Greyhound CIO Pulse figures are cited as published in Greyhound Research’s research notes of August 30, 2025, and July 17, 2026. No composite figure is computed. No Greyhound Fieldnotes are quoted, and no factual claim in this research note rests on unpublished material.
The reading of the two-part model, the billing map, the credit path, the value chain, the readings in Exhibits 1 to 8 and Tables 1 to 7, the risks, the Greyhound CXO Runbook, the questions for Microsoft, the signals, and every term proposed for a contract are Greyhound Research’s analysis, not statements by Microsoft or any other company. Quotations are checked against the live source before publication.
Authoring Disclosure
This material reflects the analysis, viewpoint and editorial judgment of Greyhound Research. The position presented is developed from Greyhound Research’s research, advisory work and accumulated institutional knowledge. Depending on the assignment, AI may be used to retrieve relevant material; structure, draft or polish prose from analyst-provided concepts, dictation, notes, rough drafts and source material; support secondary research and factual cross-checking; and assist with editing, condensation and consistency with Greyhound Research’s house style. AI does not independently establish Greyhound Research’s viewpoint, analytical position or conclusions. Material factual claims are verified against source evidence before release. Greyhound Research reviews and approves the final material and retains responsibility for it.
Copyright Policy
All content in this document and on the Greyhound Research website is protected by copyright law and may not be reproduced, distributed, transmitted, displayed, published, or broadcast without the prior written permission of Greyhound Research or, in the case of third-party materials, the prior written consent of the copyright owner of that content. You may not alter, delete, obscure, or conceal any trademark, copyright, or other notice appearing in any Greyhound Research content. Quotation of up to one hundred words is permitted where it carries attribution to Greyhound Research and a link to the published page. Reproduction, redistribution or republication beyond that extent, in whole or in part, by email or in any medium including websites, newsletters and intranets, requires prior written permission, which may be requested at the address below. The published version of each content piece on the Greyhound Research website carries sharing tools for readers who wish to share it.
For questions arising from these disclosures, including licensing requests, contact the Greyhound Research Community Relations Team at connect@thofgr.com. For media inquiries, write to media@thofgr.com.

Analyst In Focus: Sanchit Vir Gogia
Sanchit Vir Gogia, or SVG as he is popularly known, is a globally recognized technology analyst, innovation strategist, digital consultant and board advisor. SVG is the Chief Analyst at Greyhound Research (part of The House Of Greyhound), a Global, Award-Winning Technology Research, Advisory, Consulting & Education firm. Greyhound Research works closely with global organizations, their CxOs and the Board of Directors on Technology & Digital Transformation decisions.
Discover more from Greyhound Research
Subscribe to get the latest posts sent to your email.
