If you are planning, operating, or governing Copilots, the single most important business concept to understand is Copilot Credits. They turn every kind of agent “work” into a common unit you can forecast and control. Since September 1, 2025, credits replaced “messages” as the billing currency, which makes it much easier to compare scenarios, set budgets for different environments, and decide when to use PAYG for spikes.

This guide walks through what consumes credits, how to translate credits into dollars, how allocation and overage enforcement work, and how to model real‑world costs across internal and external channels.

1) What Are Copilot Credits

Copilot Credits measure the “work” an agent does: answering with classic or generative responses, grounding over Microsoft Graph, running actions or flows, or invoking AI tools. Your total cost is the sum of credits used across these features.

For PAYG, use USD $0.01 per credit as your quick conversion (enterprise agreements and currency conversions may vary).

Quick mental math

• 100 credits ≈ $1.00
• 10,000 credits ≈ $100.00
• 25,000 credits ≈ $250.00 (PAYG)

You can monitor consumption and allocate capacity by environment in the Power Platform admin center. Within each agent, Analytics breaks down billed credits by activity type and trend.

2) Credits and Event Scenarios

The table below summarizes how many credits common features consume, and whether usage is charged when the end user is licensed for Microsoft 365 Copilot and the agent runs in that user’s authenticated context inside Microsoft 365 (B2E scenarios).

Per User Licensing
These credits are NOT charged when the end user is an authenticated Microsoft 365 Copilot user (B2E). i.e. if a user has a full ($30/u/m Microsoft 365 Copilot) license, then the Copilot credits in the table are ZERO credits for ALL ROWS.

Deep reasoning note
When you use a reasoning LLM, you pay for the base feature (for example, a generative answer) plus the premium AI tools meter (100 credits per 10 responses).

3) Translating Credits to Dollars

Use PAYG $0.01 per credit as a simple conversion:

• Classic answer (1 credit) → $0.01
• Generative answer (2 credits) → $0.02
• Agent action (5 credits) → $0.05
• Tenant Graph grounding (10 credits) → $0.10
• Agent flow actions (13 credits per 100 actions) → $0.13 per 100 actions
• AI Tools
  • Basic (1 credit per 10 responses) → $0.10 per 100 responses
  • Standard (15 credits per 10 responses) → $1.50 per 100 responses
  • Premium (100 credits per 10 responses) → $10.00 per 100 responses
• Content processing (8 credits per page) → $0.08 per page

Important
If you also run Azure OpenAI directly for bespoke workloads, that token‑based pricing is separate from Copilot Studio. Don’t double‑count.

4) Worked Cost Examples

These examples are illustrative; swap in your volumes. The $ conversion uses PAYG $0.01 per credit.

Example A: Employee Helpdesk Agent in Microsoft 365 (B2E)

• A typical conversation uses 1 Tenant Graph grounding (10) + 2 generative answers (4) + 2 agent actions (10) = 24 credits per conversation.
• If 200 contractors don’t have Microsoft 365 Copilot and each runs 300 conversations/month:
  • Credits = 200 × 300 × 24 = 1,440,000
  • Dollars = $14,400 (PAYG)
• If all 500 employees are licensed for Microsoft 365 Copilot, those sessions are not charged against tenant credits. Cost = $0 for those interactions.
  • Budget context: 500 × ~ $30/month ≈ $15,000 for per‑user licensing (separate value drivers beyond this article).

Takeaway
In B2E scenarios, who is licensed can change your draw on credits by orders of magnitude.

Example B: Public Website Support Agent (B2C) with Generative Orchestration

• Average session: 3 classic (3) + 2 generative (4) + 1 action (5) = 12 credits
• 30,000 sessions/month → 360,000 credits → ~$3,600 (PAYG)
• If you add Tenant Graph grounding for signed‑in customers, add +10 credits per session → 22 credits → $6,600 for the same traffic

Optimization levers
Prefer classic answers for FAQs and reserve generative or grounding for long‑tail queries.

Example C: Ops Agent with Autonomous Triggers and Flows

• Per order: 200 flow actions → 26 credits (13 per 100) + 2 actions → 10 credits
• Total: 36 credits per order
• 10,000 orders/month → 360,000 credits → ~$3,600 (PAYG)

Observation
High‑volume, short‑flow patterns behave like a workload. Budget accordingly.

Example D: Reasoning Model Add‑On (Premium AI Tools)

• For each 10 reasoning responses, add +100 credits on top of base features
• If 1,000 sessions use 2 reasoning responses each:
  • Premium tools:1,000×2÷10×100 = 20,000 credits ($200)
  • Base features (2 generative = 4 + 1 grounding = 10 → 14 per session): 14,000 credits ($140)
  • Subtotal: $340 (PAYG)

Tip
Gate reasoning to specific intents and confidence thresholds.

5) Multi‑Environment Allocation and Enforcement

You allocate prepaid credits per environment. Unallocated consumption draws from the tenant pool. If usage exceeds available prepaid capacity, there is a grace overage, but the default enforcement threshold is 125% of the applicable prepaid capacity. After enforcement, custom agents are disabled for new interactions until you add capacity, reallocate, or rely on PAYG for that environment. Existing conversations aren’t interrupted. Admins are notified and can remediate in the Power Platform admin center.

6) Overage Scenarios Made Simple

Think like a business analyst: What gets shut off, where, and how do I keep the lights on? Here are the essential patterns without the noise.

Scenario:

You purchase 25,000 Copilot Credits, and the following allocation structure is being used:

• Environment A has 10,000 Copilot Credits allocated.
• Environment B has no allocation.
• Environment C has no allocation.
  (Env B and C therefore use the remaining tenant pool (14,500).
• Environment D has an allocation of 500 Copilot Credits and uses PAYG.

What happens …

• To Agents that are added to Environments B or C?
  They consume from the total remaining tenant pool (14,500)
• If Agents in B + C hit 125% of available capacity? (125% of 14,500 = 18125)
  All agents in Env B & C drawing from the pool are blocked for new sessions until capacity is added or reallocated. Current sessions will be able to finish. New session will get an “over capacity” user error.
• to agent in Env A? The Allocated capacity is honored.
  While Env A still has any part of its 10,000 allocation, it stays online, even if the tenant pool is in overage enforcement.
• When A runs out: After Env A consumes its 10,000, it starts drawing from the tenant pool and becomes subject to the pool’s enforcement status.
• Already in enforcement, A still fine: If the pool is already at 125% because of B and C, Env A remains unaffected as long as it still has remaining allocated capacity.
• If you dont want it to ever disable?
  enable PAYG as immunity for the spike. Env D has 500 allocated plus PAYG. When it burns through 500, it will consume from the tenant pool if there is any available. however if there is not, then PAYG auto‑kicks in, so Env D continues to serve without disabling, even if the tenant pool is in overage enforcement.

7) Budgeting Patterns That Work

• Isolate critical agents in their own environments with a small allocation plus PAYG to avoid downtime.
• Estimate B2C traffic with the usage estimator, then convert directly to $0.01 per credit.
• Minimize premium tools usage to intents that truly benefit.
• Monitor daily; alert when environments begin drawing from the tenant pool.
• Exploit B2E no‑charge for licensed users by publishing employee‑facing agents inside Microsoft 365 where appropriate.

8) Frequently Asked Consumption Questions

• Do proactive greetings count? Yes. If an agent readies or initiates a message, it consumes credits as a Classic answer, even if the user doesn’t reply.
• Do “Copilot Studio for Teams” agents consume credits? Agents created inside the Teams environment don’t. Agents created in standalone Copilot Studio and deployed to Teams do, unless the user is licensed for Microsoft 365 Copilot in eligible B2E contexts.
• Do unused credits roll over? No. Usage is tallied monthly and resets on the first day of each month.
• Do test environments consume credits? Yes. Dev, Test, and Prod all consume credits when agents run; “non‑production” isn’t free.
• Do autonomous triggers count even without a user? Yes. Background triggers, flows, and actions consume credits just like interactive queries.
• If an agent fails to answer or errors out, does it still consume credits? Usually yes. Once orchestration, grounding, or actions execute, credits are consumed even if the final output isn’t useful.
• Do custom connectors or external APIs add extra credits? The external call itself isn’t billed, but the agent action that invokes it is (and any additional reasoning/grounding steps also consume credits).
• Does retrying a question cost additional credits? Yes. Each retry is a new sequence of work and is metered again.
• Does adding knowledge sources consume credits? No. Uploading or connecting data doesn’t use credits; using those sources during answers does.
• Do cloned agents start consuming credits immediately? No. Credits begin once the clone is published and invoked.
• Does testing in Copilot Studio’s test canvas consume credits? Yes. Testing runs real orchestration and grounding.
• Are credits billed per channel? No. Billing is based on activity, not channel. The same agent across multiple channels is still metered by the work it performs.
• Will turning off generative orchestration reduce costs? Yes. It pushes more traffic to cheaper classic answers and fewer premium operations.
• Should I spread traffic across multiple environments to avoid enforcement? Sometimes. It can reduce pressure on a single pool but increases management overhead, use it deliberately.
• Can I hard‑cap credits per agent? Not directly. Control usage via environment allocation, PAYG safety nets, and by limiting premium features.
• Is consumption reporting real‑time? There can be a delay of a few hours, so running close to limits can still trigger unexpected enforcement.
• Can I forecast usage before launch? Yes. Use Microsoft’s estimator (or your own model) to simulate traffic, orchestration mix, grounding, flows, and premium tool usage. Microsoft agent usage estimator
• Can I mix prepaid and PAYG in the same tenant? Yes. It’s common; PAYG‑enabled environments continue serving even if a prepaid pool hits enforcement.

Conclusion

Treat Copilot Credits like cloud compute: forecast volumes, isolate critical workloads, and use PAYG as your safety valve. The credits model makes apples‑to‑apples comparisons possible across very different agent patterns. With a simple $0.01 per credit conversion and a handful of allocation practices, you can prevent overage surprises, keep business‑critical agents online, and grow your portfolio with confidence.

Index

• B2E: Business‑to‑Employee usage inside Microsoft 365 with authenticated users
• Copilot Credits: The unit that measures agent work (answers, actions, grounding, flows, AI tools)
• PAYG: Pay‑as‑you‑go billing at USD $0.01 per Copilot Credit (rates may vary by agreement)
• Tenant Pool: Unallocated prepaid credits available to all environments in a tenant
• Allocation: Prepaid credits assigned to a specific environment
• Enforcement Threshold: Default overage point at 125% of the applicable prepaid capacity