Skip to content

Copilot Credit Estimator

Still being worked on — use with caution

This estimator is still under active development. Numbers, defaults, and logic may change, so treat the results as directional rather than final and double-check anything you rely on for planning or budgeting.

Estimate monthly Copilot Credits (formerly "messages") for Copilot Studio agents or Microsoft 365 Copilot (Cowork) — use the What are you estimating? switch, then pick a mode to match where you are. For Studio: describe the agent in plain words, build the credit profile by hand, batch-size a portfolio from Excel, or upload a finished agent. For Cowork: forecast a population (licensed users × active-usage × credits per user) quickly, per-cohort, or straight from your M365 admin-center data. Everything runs in your browser; nothing is uploaded. New to credit billing? See How Copilot Credits are billed for the official rates and licensing rules. Want the return, not just the cost? Try the ROI Estimator.

How Copilot Credits are billed — rates & licensing

Rates are sourced from the Microsoft Copilot Studio Billing rates and management docs. Each agent turn may combine multiple features (e.g. a generative answer with tenant graph grounding = 2 + 10 = 12 credits).

Key licensing rule: When an agent runs on a Microsoft 365 surface — Microsoft 365 Copilot Chat, Microsoft Teams, or SharePoint, authenticated users with an M365 Copilot license accrue zero credits — only unlicensed users generate credit consumption. When deployed to any external channel (custom website / web widget, external or custom app, standalone, etc.), all users are charged credits regardless of M365 Copilot license status. Use the Deployment type toggle in the Detailed mode to model the correct scenario. This zero-rating applies to the standard and Copilot chat harnesses only — the GitHub Copilot harness is never covered. Pick the Harness in either mode and the estimator shows gross consumption vs. net billable side by side.

Zero-rating exceptions

A few official cases where a Microsoft 365 Copilot license does not zero-rate usage (per the billing-rate footnotes):

  • GitHub Copilot harness agents are never covered by a Microsoft 365 Copilot license — every interaction, plus building and testing the agent, bills Copilot Credits regardless of channel. Only the standard and Copilot chat harnesses are zero-rated. Use the Harness selector to model it. Microsoft bills this harness per task by complexity and publishes only credit ranges — Light 100–300 · Medium 300–500 · Heavy >500 — that bundle model tokens, tools, and the harness itself (there is no per-action rate card for this harness, so the estimator hides the per-action grid and prices a task at an editable credits-per-task anchor seeded from the tier). The anchors sit toward the high end of each band to lean slightly conservative, and Heavy is open-ended (editable upward, no cap). A one-time build & test cost is added on top.
  • Computer-Using Agent (CUA) actions are not included in the Microsoft 365 Copilot license — they bill at the agent-action rate (5 credits) even for licensed users.
  • Agent flow actions are "no charge" for licensed users only when the flow uses the "When an agent calls the flow" trigger. Agent flows on any other trigger consume credits at the standard rate.
  • Generative answers are zero-rated on Microsoft 365 surfaces / in Agent Builder only when they run without tenant-graph grounding — tenant-graph grounding always meters (10 credits/message).

Benchmarked against Microsoft's official tools. This engine's rate card and per-turn math are calibrated to match the public Copilot Studio agent usage estimator and the Learn billing doc — all base rates (classic 1, generative 2, agent action 5, tenant-graph 10/msg, flow 0.13/action, AI 0.1/1.5/10, voice 10/35/75) align, as do the doc's worked examples (a tenant-graph-grounded turn totals ~12 once the generative answer is added). Two nuances it now follows: an autonomous trigger is billed as one agent action (5) — not a flat surcharge — with the actions it invokes billed separately; and when a reasoning-capable model is detected in a solution package, a premium 10 credits / 1K tokens meter is added on top of the feature rate. Reasoning surcharges are otherwise assumed off (standard models).

??? note "Microsoft 365 Copilot (Cowork) — usage-based billing" Microsoft 365 Copilot Chat meters agent / Cowork usage as Copilot Credits through the same Copilot Studio pay-as-you-go meter (learn.microsoft.com). On Microsoft 365 surfaces, users with an M365 Copilot license are covered by that license; unlicensed users on metered agents consume credits (pay-as-you-go). You can pay-as-you-go at the per-credit rate, buy message packs, or commit with a pre-purchase.

   Unlike Studio, you don't inventory components for Cowork — you forecast a **population**: *licensed users × monthly active-usage % × avg credits per active user*. Ground it in your tenant's real numbers with the admin center's **[Copilot Chat usage report](https://learn.microsoft.com/microsoft-365/admin/activity-reports/microsoft-copilot-usage)** (active users, prompts/user) and **[Copilot Credits report](https://learn.microsoft.com/en-us/microsoft-365/admin/activity-reports/microsoft-365-copilot-credits)** (measured credits/user); the **Import from M365** mode reads either CSV export. The active-usage % and credits/user *defaults* in this estimator are neutral planning anchors — **not** official Microsoft figures — so adjust them to your data.
New here? How to use this estimator

First, pick what you're estimating — Copilot Studio agents or Microsoft 365 Copilot (Cowork) — using the switch, then choose how you want to estimate. The two products use different methods (Studio inventories what you built; Cowork forecasts a population), so each has its own modes. Every mode runs locally in your browser; nothing is uploaded.

Mode Best when… What you provide What you get
Quick You're early or unsure and just want a ballpark. A plain-English description of the agent (or an example chip). A T-shirt size, a Studio build outline, and a credit/cost range.
Quick + Import You're sizing many agents — a whole portfolio — at once. An Excel workbook with one row per scenario. Per-scenario sizes and credits, plus a portfolio roll-up.
Detailed You know the building blocks but haven't built yet. Org scope, deployment type, and the features each interaction uses. Credits per month and per user, ready for finance or IT.
Solution package The agent is already built. A Copilot Studio solution export (.zip). A component inventory, a T-shirt size, and a credit estimate.

Estimating Microsoft 365 Copilot (Cowork) instead? Flip the switch to Microsoft 365 Copilot (Cowork) — it uses a population-based method (model licensed users × active-usage % × credits per active user), with its own modes:

Cowork mode Best when… What you provide What you get
Quick You want a fast org-wide ballpark. Licensed users, an active-usage %, and avg credits/user. Monthly credits, spend, annual, and a message-pack / pre-purchase plan.
Detailed You want to model cohorts (roles, business units). One row per cohort, tuned individually. A roll-up plus a 6-month adoption forecast; open any Quick estimate here to refine.
Import from M365 You have real admin-center data. The Copilot Credits or Chat usage CSV export (or the four dashboard totals). Editable cohorts seeded from your measured credits/user, with power-user outliers flagged.
  1. Select the Quick card (the default).
  2. Type a plain-English description — what it does, who uses it, how often, and where it runs — or click an example chip.
  3. Click Build my estimate →.
  4. Answer the short guided follow-ups.
  5. Review the size, build outline, and credit/cost range.
  6. Optionally open it in the Detailed estimator to refine.
  1. Select the Quick + Import card.
  2. Click ↓ Download Excel template (.xlsx).
  3. Fill the Scenarios sheet — one row per agent or use-case (the Examples sheet is prefilled to copy from).
  4. Drop the completed workbook back on the page.
  5. Review each scenario's size and credits, plus the portfolio roll-up.
  1. Select the Detailed card.
  2. Follow the six in-panel steps: set your org scope, choose the deployment type, set the interaction frequency, fill the per-interaction feature rows, add an optional escalation path, then read the results.
  1. Select the Solution package card.
  2. In make.powerapps.com, add your Copilot Studio agent — plus any Power Automate flows and connection references — to a solution and Export it as an unmanaged .zip.
  3. Drop the .zip on the page.
  4. Review the component inventory, size, and credit estimate.
  5. See the panel's "What can I upload?" note for the A / B / C upload options.
What are you estimating?

Describe what you want the agent to do
Try an example:

We read your description to pre-fill a short guided assessment, then estimate the Studio build effort and a credit/cost profile — a directional starting point, not a real LLM analysis. Answer the questions, then open it in the Detailed estimator. Once your estimate is ready, you can also download a ready-to-import Copilot Studio starter agent (.zip) generated from your description.

Upload a built agent or flow package

Drop a Dataverse solution export (.zip) — a Copilot Studio agent, a Power Automate cloud flow, or both — or a curated agent build-spec bundle. We inventory topics, knowledge, actions, flows (with AI Builder prompts, loops and connectors), and multi-agent orchestration, then size it and estimate credits. The file is parsed entirely in your browser — nothing is uploaded to any server.

Drop your solution / package .zip here
or click to choose a file
What can I upload?
A. Dataverse solution export (most accurate):
1. Go to make.powerapps.com → Solutions.
2. Create a solution (or open an existing one) and add your Copilot Studio agent and/or Power Automate flows — plus their connection references.
3. Choose Export solution → Unmanaged or Managed → download the .zip.
4. Drop that .zip above. We read the agent/topic YAML and workflow JSON to inventory components.

B. Power Automate flow only: export just the flow's solution — we parse its trigger, actions, AI Builder prompts, loops and connectors, and treat it as an autonomous (per-run) workload.

C. Agent build-spec bundle: a folder of build docs (.md) + knowledge files (.docx/.pdf) zipped together — we infer knowledge sources, actions, orchestration and guardrails from the spec.
1 · Download the template

Grab the workbook, fill in one row per scenario on the Scenarios sheet (the Examples sheet is prefilled to copy from), then bring it back here. Every scenario gets a T-shirt size, a credit/cost estimate and a build read-out, plus a portfolio roll-up. Everything runs in your browser — nothing is uploaded.

Tip — let Copilot fill the sheet for you. You don't have to fill the Scenarios sheet by hand. Ask Microsoft 365 Copilot to populate it from a plain-English list of your agents — one row each — then re-upload. The example prompt is copy-paste-ready. Pick Copilot Chat (default, concise) or Cowork (a step-by-step analyst brief) below, hit Copy prompt, then just replace the My agents list at the bottom with your own and send it to Copilot; it uses the prefilled Examples sheet as the pattern, infers each column, and leaves unknowns blank. 
You're helping me fill in the 'Scenarios' sheet of this Copilot Credit Estimator workbook, to paste back into the sheet. Use the prefilled 'Examples' sheet as the pattern and add one row per agent from my list below. Infer each column from my descriptions and leave a cell blank if it isn't implied — don't invent enum or number values. For the Harness column, use 'GitHub Copilot' for reasoning-heavy, multistep/agentic agents and 'Standard' for rule-based Q&A / topic agents.
My agents (one per line — what it does, who uses it, how often, channel, knowledge, actions, and standard vs GitHub Copilot harness):
1)
2)
Copilot's inferences are a starting point — sanity-check the enum/number columns before importing.
2 · Import your filled-in workbook
Drop your filled-in .xlsx or .csv here
or click to choose a file
What do the columns mean?
Describe your agents — one per line

Paste a plain-English description of each agent on its own line and get a ready-to-import Copilot Studio starter (.zip) for every one, bundled together with a portfolio roll-up. This mirrors Quick + Import, but the output is generated agents, not just sizing. Everything runs in your browser — nothing is uploaded. These are directional starters to extend, not production-ready agents.

Want more control? Use a header row.
Without a header, each line is treated as one agent description. To set options per agent, make the first line a header naming any of: name, description, experience (new/classic), archetype (interactive/autonomous), knowledge, workIQ (yes/no), skills (;-separated), systems (;-separated). Separate columns with a comma or tab; wrap any value containing a comma in "quotes". description is the only required column. New experience is the recommended default.
Try an example set:

How to use this estimator

  1. Set your org scope — enter the number of users you're modelling and the proportion with an M365 Copilot license.
  2. Choose deployment type — Microsoft 365 (Teams · Copilot Chat · SharePoint) means licensed users accrue zero credits; Standalone / other channel charges all users regardless.
  3. Set interaction frequency — estimate how many times a typical user interacts with the agent per month.
  4. Build your normal path — for each row, set Uses / interaction to how many times that feature fires in a single interaction. Rows default to 0 — only count features your agent actually uses.
  5. Add an escalation path (optional) — set an escalation rate and add extra steps that only fire when an interaction escalates (e.g. a query that can't be self-served triggers an additional lookup or handoff). The rate controls what percentage of interactions incur these extra costs.
  6. Check the results — Credits / month (org) and Credits / user / month are the numbers to share with finance or IT for budget planning.
Agent type

Interactive agents are driven by people — a user sends a message or makes a call. Credits scale with users × interactions / month, and M365 Copilot–licensed users can accrue zero credits on Microsoft 365 surfaces (Teams · Copilot Chat · SharePoint).

Quick presets
Quick presets
Deployment type
Harness (Copilot Studio engine)

Standard & Copilot chat are covered by an M365 Copilot license in M365 channels. The GitHub Copilot harness is never covered — every interaction bills credits, so net billable = gross.

On Microsoft 365 surfaces (Teams · Copilot Chat · SharePoint), M365 Copilot licensed users incur zero credits. Only unlicensed users generate credit consumption. Use the % with M365 Copilot license slider to set the licensed proportion.

Organization
Employees, contractors, or team members you're modelling
%
Microsoft 365 surfaces (Teams · Copilot Chat · SharePoint): licensed users accrue zero credits — only unlicensed users are billed. Pilots typically 10–20 %; full rollouts 60–100 %
How many times does a typical active user interact with this agent each month
How many trigger events the agent processes each month — inbound emails, uploaded documents, queue items, records, etc. Every event is billed (no per-user licensing discount).
%
% of interactions that require additional handling beyond the normal path
Edit rows or add your own process steps

📞 Voice rows are per minute. In the Uses / interaction column, enter the average voice minutes per conversation — it's multiplied by your interactions × users to get the monthly total (credits = rate × minutes). The per-minute voice rate already includes the classic / generative answers and agent actions that occur during the call, so don't also fill in those rows for a voice agent. Tenant-graph grounding, content processing, AI tools, and agent flows are billed separately on top.

Agent feature / interaction type Uses / interaction Credits / use Credits / interaction
Escalation path — 0% of interactions trigger these additional steps
Effective credits / interaction —
Estimated monthly consumption
—
Unlicensed users (billed)
—
Total interactions / month
—
Credits / month (org)
—
Credits / user / month
Budget check (optional)
Enter a monthly credit budget above to see headroom or overage against your estimate.
How do you want to estimate?

Cowork is estimated top-down: model a population — licensed users × how many are active × how many credits each active user burns a month. These are directional planning numbers; tune every input. Everything runs in your browser.

Model your population
People you'd turn Cowork on for.
Share of licensed users active in a month.
The primary driver. Planning default 5,000 — adjust to your data.
Optional — set low/high ends for active usage and credits/user to bracket the estimate.
Global assumptions (price, discount, license floor, budget)
Pay-as-you-go list rate.
Flat $ / licensed user / month, added to spend.
0 = no cap.

Credit & price mechanics follow public Microsoft Learn (PAYG $0.01/credit; $200 / 25,000-credit pack). Active-usage % and credits/user are neutral planning anchors, not official Microsoft figures.

Estimated monthly consumption
—
Active users
—
Credits / month
—
Cowork spend / month
—
Cost / active user / mo

Cohorts

Split the population into groups (by role, region, or business unit) and tune each. Avg credits/user/mo is the primary driver.

Intensity helper optional — estimate a credits/user number from a prompt mix
—
Illustrative weights (125 / 500 / 2,500). Type this into a cohort.
Global assumptions (price, discount, license floor, budget)

Credit & price mechanics follow public Microsoft Learn. Active-usage % and credits/user are neutral planning anchors, not official Microsoft figures.

Optional — brackets every cohort's active usage and credits/user by ±buffer to show a roll-up range.
Applied to both drivers, every cohort.
Roll-up
—
Active users
—
Credits / month
—
Cowork spend / month
—
Cost / active user / mo

6-month adoption forecast
Adoption ramp, month over month.

Start from your tenant's real numbers. In the Microsoft 365 admin center → Reports → Usage → Microsoft 365 Copilot, export the Credits report (measured credits per user) or the Copilot Chat usage report (active users & prompts). Drop the CSV below — or paste the four dashboard totals, or a screenshot. Everything is parsed in your browser; nothing is uploaded.

1 · Bring in your data

Drop a CSV export (Credits or Chat usage) or a dashboard screenshot here, or .

CSV gives the richest result — per-user rows become measured credits/user and power-user outlier flags.

…or paste the four dashboard numbers
2 · Licensed users in scope
Admin reports only cover active/metered users. Enter the licensed population so we can compute active-usage %.
3 · What we found

Import a file or paste totals above to see a summary.

What to do with this number

  • Credit budget planning — share the monthly credit estimate with your IT/finance team alongside the licensed user count to validate your M365 Copilot SKU allocation.
  • Adoption benchmarking — as real usage data comes in from the admin center, compare actuals to this estimate to see whether adoption is ahead or behind plan.
  • Scenario planning — run the estimator at 3 adoption-rate levels (conservative / target / optimistic) to bracket your credit spend.
Was this page helpful?

Found an issue or want to suggest an improvement? Open a GitHub issue  ·  Star on GitHub