Kesslernity · M365 Copilot Field Guide

Copilot Credits

What is included in the seat. What starts the meter. What Microsoft will not price.
Copilot Credits are the shared usage-based currency across eligible Microsoft Copilot services. Some surfaces publish a fixed rate. Most publish a range labeled a planning estimate. Several publish nothing at all. This guide separates those three categories, because a budget that treats an unpublished rate as zero is not a budget.
Source edition Credits Guide, November 2026
Editorial reference date November 8, 2026
Billing Usage-based, Copilot Credits
Scope Microsoft Copilot, Copilot Studio, Business Applications
Published by kesslernity.com
Requires before you start

To read this guide: nothing. It is a map of what Microsoft publishes, not a configuration walkthrough.
To act on it: usage-based billing for Copilot Credits enabled in the Microsoft 365 admin center, at least one spending policy, and the underlying product licenses for whichever surface you are costing. Several surfaces in this guide are in public preview and carry their own prerequisites, listed per surface below.

At a glance
One currency, many meters
Copilot Credits are a common currency across eligible services. They are not GitHub Copilot credits, and the two do not convert.
Two fixed rates are published, both 0.1 credits per API call
The Work IQ Tools API, for deterministic actions and tools, at general availability; and the Copilot Managed Runtime, for hosting, runtime and management, in public preview. Everything else is variable or a planning estimate.
Unpublished is not free
Where Microsoft publishes no rate, this guide records the meter and leaves the cost blank. A blank is exposure, not zero.

What a Copilot Credit is, and four things it is not

Microsoft describes Copilot Credits as "a common currency for eligible Microsoft services with usage-based billing." Organizations buy capacity through fixed licensing, the Copilot Credit Pre-purchase Plan (P3), pay-as-you-go, prepaid capacity packs, or supported combinations. Administrators govern access and spend from the Cost management dashboard in the Microsoft 365 admin center.

Four boundaries matter more than the definition:

It is not GitHub Copilot credits

Different product, different meter, different billing surface. Nothing in this guide converts between them.

It is not a universal dollar amount

Monetary value depends on plan, purchasing route, currency, region, and effective date. Microsoft's own Cost Management API "reports usage in AI credits and does not include pricing, invoices, currency costs, or purchasing forecasts." Price any credit against the specific agreement it will be drawn from, not against a remembered rate.

It is not charged for first-party Copilot use of Work IQ

Microsoft Copilot is natively built on Work IQ. There are no incremental charges for Work IQ accessed inside Copilot experiences such as chat, the Office apps, or built-in agents including Researcher, Facilitator, and Analyst. Charges apply only to access through the Work IQ APIs.

It is not capped by the visibility features

Spending limits, alerts, and threshold notifications exist and are real controls. The in-product /cost response and the end-user remaining-credit view are visibility, not a guarantee that spend stops at a number. Treat them as instrumentation.

The meter map

Every surface below consumes Copilot Credits. The rate column is the discipline: a fixed per-call rate appears only where Microsoft publishes one, and everywhere else the column says what Microsoft publishes instead, which is usually a planning band. Nothing here is rounded to zero.

SurfaceStatusBilling unit and rateLicense prerequisiteSource
Copilot Cowork GA; app building is Frontier Program Variable. Published task bands: light 100–300, medium 300–700, heavy over 700 credits Microsoft Copilot USL; usage-based billing Credits Guide pp. 6–8
Copilot Code Microsoft Frontier Program Variable. Published task bands: light 25–60, medium 75–150, heavy over 200 credits Microsoft Copilot USL; usage-based billing Credits Guide pp. 11–13
Work IQ Tools API GA 0.1 credits per API call, deterministic actions and tools Work IQ enabled for tenant; usage-based billing enabled. Not an entitlement of the Copilot license. Credits Guide pp. 14, 26
Work IQ Chat or Context API GA Variable. Driven by models, runtime, context, and tools used for grounding, retrieval, and reasoning As above Credits Guide pp. 14, 26
Business Application tools, including MCP Public preview Follows Work IQ. Split by call type, see section 03 Appropriate Dynamics 365 or Power Platform licenses for the underlying data and services Credits Guide pp. 25–26
Customer conversation harness Public preview Variable. Published planning bands by modality, see section 04 Copilot Studio Credits Guide pp. 15, 17
Contact Center Intelligence (CCI) Public preview Variable. Bands published per agent, in three different units, see section 05 Connected contact center environment and engagement channels Credits Guide pp. 19–21
Teams Phone Agent Named billable service Variable. Published call bands, see section 06 Copilot Studio for embedded custom voice agents Credits Guide pp. 21–22; Learn
Advanced work in SharePoint Named billable service Variable. Light tasks included in license; medium and heavy consume credits Microsoft Copilot USL; M365 subscription including SharePoint; usage-based billing; spending policy Credits Guide pp. 9–11; Learn
Advanced work in OneDrive Named billable service Variable. Same included-versus-advanced split As above, OneDrive Credits Guide pp. 9–11; Learn
GitHub Copilot harness Named in the Credits Guide Variable. Published runtime bands: 100–300, 300–500, over 500 credits. Manual, non-LLM configuration does not bill Copilot Studio Credits Guide pp. 16–17
Copilot Managed Runtime Public preview 0.1 credits per API call for hosting, runtime and management. Agents and Work IQ APIs on top bill separately Hosts apps from Cowork, Copilot Studio, Copilot Code and third-party tools. Included for Power Apps Premium USL holders, subject to that license's request limits Credits Guide pp. 24–25
Opal (Frontier) Rollout announced to begin November 9, 2026 Metered; rate not published in the cited notice. Copilot Credits usage-based billing and spending-policy assignment MC1540330, via third-party archive

Read the status column before the rate column. Six of these thirteen are public preview, Frontier Program or unreleased as of November 8, 2026. Preview pricing and availability can change without notice, and Microsoft says so explicitly in Appendix A of the source guide.

Business Application tools, including Model Context Protocol (MCP)

Status: public preview. Added to the source guide in the October 2026 revision, pages 25 to 26.

Business Application tools span Dynamics 365, Power Platform, and Dataverse. They give Copilot and agents shared context for understanding business data, the relationships between records, and the business rules that guide next steps. Microsoft notes these tools "may be exposed through Work IQ APIs, MCP servers, or other supported integration surfaces."

The single most misread fact in this release. It is not true that an MCP call costs 0.1 credits. The rate attaches to the call type, not to the protocol. Microsoft's own example of an MCP tool, "Dynamics 365 Sales or Customer Service MCP tools for non-deterministic, LLM based interactions," sits in the variable row, not the fixed one.

Query typeConsumptionMicrosoft's example scenarios
Work IQ Chat or Context API Variable. Consumption varies based on the models, runtime, context, and tools used for grounding, retrieval, and reasoning. Ground responses in business context from Dataverse. Use Dynamics 365 Sales or Customer Service MCP tools for non-deterministic, LLM based interactions, such as reviewing opportunity risks and recommending actions.
Work IQ Tools API Static: 0.1 Copilot Credits per API call for deterministic actions and tools. Use deterministic APIs to update a record in a Dataverse table. Use Dynamics 365 Sales tools for deterministic business actions such as listing cases. Use Power Apps tools to create a new record.
The billing pattern is surface-independent

Microsoft states that Business Application tools "follow the above usage-based billing in all integration surfaces whether through Work IQ APIs, MCP servers, or other supported integration surfaces." So the question to ask of any call is not "which surface did this arrive on" but "was this a deterministic action, or did a model reason about it."

A wording difference worth knowing

Page 14 of the source states the Tools API rate as "0.1 Copilot Credits per API call for actions and tools." Page 26, in the Business Applications context, states it as "0.1 Copilot Credits per API call for deterministic actions and tools." This guide uses the narrower page 26 wording, because it is the statement made where MCP is in scope.

Underlying licenses still apply

"Appropriate Dynamics 365 or Power Platform licenses are required for use rights to the underlying business data and services." Credits buy the AI work. They do not buy access to the records the AI is working on.

Customer conversation harness

Status: public preview. Added in the October 2026 revision, pages 15 and 17.

Copilot Studio exposes three harnesses. A harness is the software scaffolding that lets an agent plan, reason, maintain context, and interact with tools, systems, or other agents. The customer conversation harness is the one for customer-facing agents: agents that answer questions, complete tasks, and provide support across voice and chat.

Microsoft's statement on consumption is deliberately multi-variable: the harness "uses Copilot Credits through usage-based billing. Consumption depends on text versus voice, duration, model selection, conversation volume, and the work the agent performs."

There is no per-minute rate and no per-conversation rate. What Microsoft publishes is a set of scenario bands, split by modality. Voice costs materially more than text, and premium speech-to-speech voice costs more again. Those bands are labeled "rough planning estimates" in the source and are reproduced here as planning inputs only.

ScenarioShapeDigitalGenAI VoicePremium (S2S) Voice
LightAbout 2 minutes on voice. Simple FAQ or confirmation. Little or no tool use.5–15 credits20–30 credits34–44 credits
MediumAbout 5 minutes on voice. Moderate reasoning across turns. One or two tools.15–25 credits50–75 credits85–110 credits
HeavyAbout 10 minutes on voice. Extended tool-assisted workflow.25–50 credits100–150 credits170–220 credits

Source: Microsoft Copilot Credits Guide, November 2026, page 17. Microsoft's caption: "These figures are rough planning estimates. Actual credit consumption will vary based on workflows and usage patterns."

What this means for a forecast

The modality split is the dominant variable, not duration. Compare like with like and the modality gap is large: 5 to 15 credits digital against 34 to 44 premium speech-to-speech at the light tier, and 25 to 50 against 170 to 220 at the heavy tier. The comparison only holds tier for tier, because the bands overlap across tiers: a heavy digital conversation at 50 costs more than a light premium-voice one at 44. If you are modeling a contact center move, model the channel mix first and the conversation length second.

Contact Center Intelligence (CCI)

Status: public preview. Added in the October 2026 revision, pages 19 to 21.

CCI is a contact center solution for transforming an existing contact center "without requiring a full replacement of their current contact center platform." It supports both Microsoft and third-party contact center systems, and covers self-service, agent assistance, quality management, and operations across voice, chat, web, Teams Phone, and contact center scenarios.

Microsoft lists the billable workloads as contact center agents (Customer Assist, Quality Assurance, Service Operations), contact center workflows and orchestration, real-time coaching and evaluation, and customer conversation and voice experiences.

Read the unit column, not just the number. CCI is the one surface in this guide where the published bands are not comparable to each other. The Customer Assist Agent is costed per interaction, the Quality Assurance Agent partly per turn, and the Service Operations Agent per setup. Adding these figures together produces a number that means nothing. One band is also published as Not specified, and it stays that way here.

AgentUnitLowMediumHeavy
Customer Assist Agent
Digital and voice interactions
Per interaction 2–5 credits 10–15 credits 20–25 credits
Quality Assurance Agent
Scores and evaluates conversations
Mixed, see cells 5–10 credits 5–10 credits per turn 10–20 credits per turn
Service Operations Agent
Configures and governs the environment
Per setup 15–20 credits 35–45 credits Not specified

Source: Microsoft Copilot Credits Guide, November 2026, pages 19 to 20. Microsoft's caption on each: "These figures are rough planning estimates. Actual credit consumption will vary based on workflows and usage patterns."

Why the per-turn unit matters more than the range

A Quality Assurance Agent at the low tier evaluates a closed conversation once, at close. At the medium tier it evaluates in real time, every four turns in messaging or every ten seconds in voice. The credit figure barely moves between those two tiers, 5 to 10 either way, but the unit changes from once per conversation to once every few turns. A long conversation can therefore cost many multiples of the low tier at a band that looks identical on paper.

Prerequisites

Configure and connect the desired contact center environment and customer engagement channels. Copilot Credits usage-based billing enabled. Microsoft notes that Copilot Studio management capabilities let administrators view billed credits and configure monthly consumption limits for individual agents.

Teams Phone Agent

Status: named billable service. Added in the October 2026 revision, pages 21 to 22. Microsoft Learn lists it among the services managed by usage-based billing in the Microsoft 365 admin center.

An AI-powered receptionist for phone communications. It greets callers, answers common questions, schedules appointments, and routes calls with context. Through an integration with Copilot Studio, custom conversational voice agents can be embedded into it for workflows such as billing, order status, inventory management, and support.

Credit usage "scales with model usage, call duration, and the complexity of the tasks," driven by three factors: Models (real-time listen, reason, and speak, including translation and transcription), Tools (reasoning and actions taken to satisfy intents, such as lookups, scheduling, and transfers), and Runtime (how long the agent stays engaged on the call).

Call typeShapeEstimated credits
Light
Simple information requests
About 0 to 60 seconds. 1 to 2 agent actions. Minimal reasoning.5–15 credits
Medium
Information plus action
About 1 to 2 minutes. 3 to 5 agent actions. Multiple capabilities combined.15–40 credits
Complex
End-to-end customer journeys
Over 2 minutes. 6 or more agent actions. Multi-step workflow orchestration.Over 40 credits, no upper bound published

Source: Microsoft Copilot Credits Guide, November 2026, page 22. Microsoft's caption: "Please note that these examples are illustrative only, not necessarily typical or expected."

The complex band has no ceiling

Microsoft publishes the complex tier as "40<", that is, more than 40 credits, with no upper figure. For forecasting that is an open interval, not a number. Treat the complex band as uncosted exposure and size it by call volume rather than by multiplying 40.

Advanced work in SharePoint and OneDrive

Status: updated in the November 2026 revision, pages 9 to 11. This is the newest change in the source guide. Microsoft Learn lists "Advanced work in SharePoint" and "Advanced work in OneDrive" as separately named services managed by usage-based billing.

Microsoft's framing: "Everyday individual productivity experiences are included with the Microsoft Copilot USL. Advanced capabilities and tasks that require greater scale, automation, or real-time processing use usage-based billing and consume Copilot Credits."

Included with a Microsoft Copilot licenseRequires Copilot Credits
Ask questions, find content, and catch up
Create HTML, Word, Excel, or PowerPoint files
Create simple rules, approvals, and deterministic workflows
Rename and share individual files
Create and manage skills
Use Autofill metadata processing
Create and edit pages, libraries, or lists (SharePoint only)
Work across large or unbounded content sets, including large-scale synthesis and document review
Create and edit files
Build multi-step, complex AI workflows, such as involving skills in scheduled AI automations
Use real-time Advanced Autofill and other continuous AI-driven metadata processing
Generate new images directly into document libraries and pages, and edit existing pages
Get analytics on site and content performance across sites, pages, news, and documents (SharePoint only)
Create entire sites and solutions (SharePoint only)
Create and edit pages, libraries, and lists at scale (SharePoint only)

The Autofill distinction is the one that will surprise a site owner. Autofill in which metadata fills in over time remains included for Microsoft Copilot licensed users. Site-wide real-time Advanced Autofill is different: usage-based billing must also be enabled for the SharePoint site, not only for the tenant and the user. Same feature name, two billing states, separated by whether the processing is continuous.

The task tiers, and what the asterisks mean

The source guide grades SharePoint tasks as light, medium, and heavy. Light tasks are available through the Microsoft Copilot license. Medium and heavy tasks are available through the Microsoft Copilot license and require Copilot Credits. Light means one item or a small, clearly defined set, creating a few outputs. Medium means a defined collection, multiple AI-assisted actions, a recurring cadence, hundreds of outputs. Heavy means large or open-ended content sets, repeated AI analysis, at-scale or ongoing work, thousands of outputs. Microsoft publishes no credit figures against these tiers.

The two spending-policy controls

In the Microsoft 365 admin center, admins must create a spending policy and configure access for the applicable users by enabling Advanced work for SharePoint or Advanced work for OneDrive in Microsoft Copilot. These are separate toggles. Enabling one does not enable the other.

Prerequisites, in full

Microsoft Copilot USL. A Microsoft 365 subscription that includes SharePoint or OneDrive. Copilot Credits usage-based billing enabled. A spending policy with the relevant Advanced work control enabled. For site-wide real-time Advanced Autofill, usage-based billing enabled for the SharePoint site.

What /cost does and does not do

End users can see how many credits a task used by prompting /cost, and can check how many credits remain until their monthly limit resets. That is per-task visibility and a running balance. It is not a spending cap, it does not block the next task, and it does not reconcile to an invoice. Administrators set the actual limits, alerts, and threshold notifications centrally.

Opal (Frontier): a separate, upcoming rollout

Opal is not one of the twelve surfaces above and should not be planned alongside them. As of the November 8, 2026 reference date for this guide, Opal is an announced rollout that begins November 9, 2026. It is not generally available and not universally available. Nothing in the cited notice says otherwise.

MC1540330 describes Opal (Frontier) as "a Computer Use Agent experience for Microsoft 365 enterprise users." Access is admin-governed and depends on usage-based billing configuration.

FieldWhat the notice states
Status on November 8, 2026Announced. "Rollout begins on November 9th, 2026." No completion date is given.
AudienceMicrosoft 365 enterprise users. No specific SKU or seat entitlement is named.
Billing requirement"Opal usage requires Copilot Credits with usage-based billing and spending-policy assignment."
Access failure mode"Users without an assigned spending policy may be unable to access the experience."
Credit rateMetered; rate not published in the cited notice.
Admin checklist before broad rollout
1
Verify billing
Confirm Copilot Credits usage-based billing is enabled and a funding method is attached. Without it, the experience is not reachable.
2
Assign intended users and groups
Attach the intended users or groups to a spending policy. The notice is explicit that an unassigned user may simply be unable to access Opal, so this is an availability control, not only a cost control.
3
Review covered services and spending controls
Check which services each policy covers and what limits, alerts, and thresholds apply. Note that the Auto-apply new services setting is enabled by default on spending policies, so a new service can inherit an existing policy without a deliberate decision. Turn it off where new services should be reviewed first.
4
Confirm tenant availability
Confirm the experience has actually reached your tenant before communicating it. A rollout start date is the beginning of a deployment, not a guarantee of same-day availability.
Provenance of this section

MC1540330 was read on a third-party archive, mc.merill.net, which is not operated by Microsoft and makes no verbatim-reproduction guarantee. The original notice sits behind tenant authentication in the Microsoft 365 admin center Message Center and could not be opened for this edition. Verify against the original before acting on it. Corroborating signal: as of November 8, 2026, Microsoft Learn's list of services managed by usage-based billing does not include Opal, which is consistent with a rollout that has not yet begun.

Worksheet: separating costed exposure from uncosted exposure

The purpose of this worksheet is not to produce a number. It is to produce two numbers and keep them apart: what you can cost from published rates, and what you are exposed to but cannot cost. Most forecasting errors in this area come from collapsing the second into the first by treating it as zero.

Rules for filling this in. An unknown rate stays blank or reads "not costed." It never becomes 0. Known-cost subtotals and uncosted exposure are reported separately and never summed into a single headline figure. Every monetary conversion is stamped with its plan, purchasing route, currency, region, and effective date, because none of those is universal.

MeterVolume driverYour volumePublished rateCredits
Copilot Cowork tasksTasks per month________Planning band only, 100–300 / 300–700 / >700band, not a rate
Copilot Code tasksTasks per month________Planning band only, 25–60 / 75–150 / >200band, not a rate
GitHub Copilot harness runtimeRuns per month________Planning band only, 100–300 / 300–500 / >500band, not a rate
Copilot Managed RuntimeHosting API calls per month________0.1 credits per call________
Work IQ Tools API, deterministicAPI calls per month________0.1 credits per call________
Work IQ Chat or Context APIQueries per month________Variablenot costed
Business Application tools, deterministicCalls per month________0.1 credits per call________
Business Application tools, LLM basedInteractions per month________Variablenot costed
Customer conversation harness, digitalConversations per month________Planning band onlyband, not a rate
Customer conversation harness, voiceConversations per month________Planning band onlyband, not a rate
CCI, Customer Assist AgentInteractions per month________Planning band onlyband, not a rate
CCI, Quality Assurance AgentTurns per month, not conversations________Planning band, mixed unitsband, not a rate
CCI, Service Operations AgentSetups per month________Heavy tier: Not specifiednot costed
Teams Phone Agent, light and mediumCalls per month________Planning band onlyband, not a rate
Teams Phone Agent, complexCalls per month________Open interval, over 40, no ceilingnot costed
Advanced work, SharePointMedium and heavy tasks per month________Not publishednot costed
Advanced work, OneDriveMedium and heavy tasks per month________Not publishednot costed
Opal (Frontier)Not yet observable________Metered; rate not published in the cited noticenot costed
Worked example, synthetic inputs

Every number in this example is invented for illustration. It is not a benchmark, not an observed tenant, and not a Microsoft figure. Replace all of it.

Synthetic volumes: 40,000 deterministic Work IQ Tools API calls per month, and 12,000 deterministic Business Application tool calls per month.

Costed subtotal: 52,000 deterministic calls at 0.1 credits each = 5,200 credits per month. This is the only part of the example that derives from a published rate.

Uncosted exposure, reported separately and deliberately not added: every other meter in the table above. That is all Chat and Context API consumption, all LLM based Business Application tool interactions, the entire customer conversation harness footprint, all three CCI agents, Teams Phone Agent calls, all medium and heavy Advanced work in SharePoint and OneDrive, and any Opal usage after November 9.

Excluded from this example: monetary conversion of credits, because that requires the plan, purchasing route, currency, region, and effective date of a specific agreement. Also excluded: the underlying Dynamics 365, Power Platform, SharePoint, Copilot Studio, and Copilot license costs, which are seat costs and sit on a different line of the bill.

The honest output of this worksheet is a sentence of the form: "We can cost 5,200 credits a month against a published rate. Every other meter we are running is either a planning band or unpublished, and we cannot cost it from public information." That sentence is more useful to a finance partner than a single confident total would be.

Two traps this layout is designed to stop. First, the planning bands are not rates, and three surfaces publish them in incompatible units, so summing a band column produces a number with no meaning. Second, two entries have no upper bound at all: the CCI Service Operations heavy tier reads "Not specified" and the Teams Phone Agent complex tier reads "over 40". An open interval multiplied by a volume gives a floor and no ceiling: a hundred complex calls are more than 4,000 credits, with nothing published above that. Carry both as named exposure with a floor, never as a point estimate.

Spending controls, and what they actually control

Administrators configure these from Copilot > Cost management in the Microsoft 365 admin center.

ControlWhat it doesWhat it does not do
Identity-based spending policiesControl which users and groups can access supported servicesDoes not itself price anything
Spending limitsSet limits for supported users, environments, and agentsNot a billing guarantee; review how your funding method behaves at the limit
Alerts and threshold notificationsNotify administrators and users as thresholds approachNotification is not prevention
Auto-apply new servicesApplies a policy to future supported services and agentsEnabled by default. A future service can inherit a policy without a deliberate decision
Reader-based rolesLet finance, operations, and governance review consumption read-onlyNo permission to change policies or billing
/cost and remaining balancePer-task credit visibility for end users, plus credits remaining until monthly resetNot a cap and not an invoice
One scope boundary to check in your own tenant

Copilot Studio environment and agent policies are opt-in. Existing Power Platform admin center controls continue to apply until a Microsoft 365 spending policy covers that scope. If your Copilot Studio estate predates your Microsoft 365 spending policies, the two are probably not yet governed from the same place.

Change log

DateChangeSource and verification
November 8, 2026 Guide created. Added Business Application tools including MCP, the customer conversation harness, Contact Center Intelligence, Teams Phone Agent, and Advanced work in SharePoint and OneDrive. Added a separate Opal (Frontier) callout and admin checklist. Added the meter map and the costed-versus-uncosted worksheet. Microsoft Copilot Credits Guide, November 2026 edition, 30pp, fetched and read November 8, 2026. Microsoft Learn usage-based billing overview, page metadata ms.date 2026-10-30, updated_at 2026-11-03, read November 8, 2026. MC1540330 via third-party archive, read November 8, 2026.

Upstream change log, reproduced from Appendix B of the source guide, for the items covered here: Copilot in SharePoint and OneDrive, update, November 2026, pages 9 to 11. Microsoft Contact Center Intelligence, new, October 2026, pages 19 to 21. Customer conversation harness, new, October 2026, pages 15 and 17. Teams Phone Agent, new, October 2026, pages 17 to 20. Business Application tools including MCP, new, October 2026, pages 25 to 26. The source guide also records that Microsoft 365 Copilot was rebranded as Microsoft Copilot in August 2026, which is why this guide uses the current name throughout.

After this guide

Every guide in this set is free, and stays free. What they do not cover is the rollout itself: the operating model, the 90-day plan with owners and decision gates, the governance checklist, and the ROI case for a finance team that has already said no once.

That is the M365 Copilot Deployment Kit. $97, one payment, no subscription.

See what is in the Deployment Kit →