Microsoft Copilot articles
Copilot Studio licensing cost for a small business: credits, packs and what an agent actually burns
Copilot Studio licensing cost is not a per-seat number, which is why the usual budgeting instinct fails on it. Microsoft sells Copilot Studio as a tenant-wide capacity: $200 per pack per month for 25,000 Copilot Credits, read from Microsoft’s pricing page on 15 September 2026, or a pay-as-you-go meter billed through an Azure subscription. What decides your invoice is not how many people use the agent but how the agent was designed — because a scripted answer costs 1 credit and the same question answered with tenant-wide grounding costs 12. Twelve times the bill, from one toggle.
A tenant capacity, not a seat
Three facts reframe the whole budget. First, the maker license is free: an administrator assigns it from the Microsoft 365 admin center, and Microsoft notes you need a tenant prepaid Copilot Credit pack subscription in place before that user license can be granted. Second, consumption is pooled across the entire tenant rather than tracked per person, so ten agents in four environments all draw on one balance. Third, since 1 September 2025 the billing unit is the Copilot Credit rather than the message — a rename with the same quantity per pack and the same metered rate, but every pre-2025 article you find still counts in messages.
There are four ways credits reach your tenant, and they are not alternatives so much as layers. Microsoft’s Copilot Studio licensing and billing documentation (opens in new tab) is the authoritative reference for all four, and it is worth reading before anyone raises a purchase order.
| How you buy | Commitment | Published rate | Best for |
|---|---|---|---|
| Prepaid Copilot Credit pack | Monthly subscription, tenant-wide | $200 per pack per month, 25,000 credits ($0.008 per credit) | Steady, measured volume that reliably fills the pack |
| Pay-as-you-go meter | None — billed to an Azure subscription in arrears | About $0.01 per credit; Microsoft states packs save up to 20% against it | Pilots, seasonal demand, and overage protection |
| Credit prepurchase plan (CCCUs) | One year, prepaid, bought in the Azure portal | Commit units usable across eligible Microsoft products | Known annual volume across more than Copilot Studio |
| Microsoft 365 Copilot license | Per user, $30 per user per month billed annually | Core employee-facing agent use is zero-rated | Internal agents for people who already hold the seat |
Figures above are Microsoft’s published list prices, read on 15 September 2026, and Microsoft’s to change. The one rate Microsoft does not print on its pricing page is the metered per-credit figure; the page states only that up-front purchase saves up to 20%, which puts the meter at $0.01 against the pack’s $0.008, and that is the rate independent licensing guides publish. Treat it as a close estimate and confirm it on your own Azure invoice after the first month.
The rate card that actually sets your bill
Every Copilot Studio budget that goes wrong goes wrong here. Credits are consumed per interaction type, and the types differ by two orders of magnitude. The table below is Microsoft’s published rate card, with the dollar cost of each interaction worked out at both the pack rate and the metered rate so you can see the shape of the spend rather than an abstract credit count.
| Agent feature | Copilot Credits | At the pack rate ($0.008) | At the meter (~$0.01) |
|---|---|---|---|
| Classic (scripted) answer | 1 | $0.008 | $0.010 |
| Generative answer | 2 | $0.016 | $0.020 |
| Agent action | 5 | $0.040 | $0.050 |
| Content processing tools, per page | 8 | $0.064 | $0.080 |
| Tenant graph grounding, per message | 10 | $0.080 | $0.100 |
| Agent flow actions, per 100 actions | 13 | $0.104 | $0.130 |
| Text and generative AI tools (basic), per 10 responses | 1 | $0.008 | $0.010 |
| Text and generative AI tools (standard), per 10 responses | 15 | $0.120 | $0.150 |
| Text and generative AI tools (premium), per 10 responses | 100 | $0.800 | $1.000 |
Credit counts are Microsoft’s, from the Copilot Credits billing rates documentation (opens in new tab); the dollar columns are that rate card multiplied out, not a Microsoft figure. Two rows deserve a highlighter. Tenant graph grounding adds 10 credits on top of the answer, so Microsoft’s own worked example prices a single grounded response at 12 credits. And the premium text and generative AI tools row is how a reasoning model bills: Microsoft charges the feature rate for the operation plus 10 credits per 1,000 tokens of reasoning. Switching an agent to a reasoning model does not nudge the bill, it multiplies it.
Architecture decides the invoice, headcount does not
Take a 60-person company running one internal helpdesk agent. Forty conversations a day, five turns each, every answer generated from an attached SharePoint knowledge source at 2 credits a turn. That is 400 credits a day, roughly 8,800 across a 22-day working month — about $88 on the meter, comfortably inside a single $200 pack. Now switch on tenant graph grounding so answers draw across the whole tenant. Each turn becomes 12 credits: 2,400 a day, 52,800 a month, about $528 metered. Same people, same questions, six times the bill.
Illustrative scenario — not a client. Volumes are assumed; the credit rates are Microsoft’s.
That second number also exposes the packs-versus-meter trap. 52,800 credits needs three packs, because packs are bought whole: $600, against $528 on the meter. The general rule falls out of the arithmetic — a pack is $200 for 25,000 credits, so it only beats the meter once you consume more than 20,000 of them. Fill at least 80% of every pack you buy, or the meter is cheaper. And since Microsoft enforces capacity monthly with no carry-over of unused credits, a pack bought for a month that under-delivers is simply gone.
The practical sequence, then, is meter first and packs later. Run the pilot on pay-as-you-go, take one month of real consumption out of the Power Platform admin center, and only then buy packs sized to the floor of that demand rather than its ceiling. If you are still at the stage of choosing what the agent should do at all, our walkthrough of building your first Copilot Studio agent covers the design choices — knowledge sources, topics, actions — that this page prices.
The Microsoft 365 Copilot line that zero-rates internal agents
This is the largest single lever on a small-business bill, and it is easy to miss. Where a user holds a Microsoft 365 Copilot license and interacts with an agent as their authenticated self, Microsoft zero-rates the core activity: classic answers, generative answers and Microsoft Graph tenant grounding used in Microsoft 365 Copilot, Teams or SharePoint do not count against your packs or your meter. Agent actions and agent flow actions are included too, though for flows the inclusion applies only to runs started by the “When an agent calls the flow” trigger. Fair-use limits apply, and Microsoft reserves the right to change them.
The consequence is a clean split. An employee-facing agent, deployed in Teams, for a population that already holds Microsoft 365 Copilot seats at $30 per user per month, may consume close to nothing from your credit balance. A customer-facing agent on your public website, talking to unlicensed and anonymous people, consumes every credit at list rate — and that is the agent whose volume you must model honestly before launch. Whether those $30 seats earn their keep in the first place is a separate decision with a separate answer; we work it through in our assessment of whether Microsoft Copilot is worth the cost rather than re-arguing it here.
One more boundary worth knowing: Power Automate cloud flows are licensed under Power Automate and are neither billed as Copilot Credits nor subject to Copilot Studio enforcement. Only agent flows managed inside Copilot Studio draw on this balance. Mixing the two up is how a Copilot Studio forecast ends up double-counting automation you already pay for.
What happens when the credits run out
Microsoft’s enforcement is specific, and knowing it changes how much headroom you buy. On prepaid capacity, enforcement triggers when a tenant reaches 125% of its purchased credits: custom agents are disabled, an in-flight conversation finishes, and subsequent attempts are rejected with a message telling your users the agent has reached its usage limit. An email goes to the tenant administrator and a notice appears in the Power Platform admin center. Agent flows are stricter still — new runs are blocked as soon as prepaid capacity is exhausted, though the parent agent keeps answering.
Three controls exist against that outcome. You can allocate credits per environment so a runaway development agent cannot drain the balance serving production. You can set a monthly consumption cap per agent in the Power Platform admin center, which is the cheapest insurance on this list and takes minutes. Or you can link the environment to pay-as-you-go, which removes enforcement entirely because overage simply bills to Azure — useful protection, and also the fastest way to a surprise invoice if nobody is watching the meter. For a business that has never metered a Power Platform workload before, the same governance questions are the ones our Microsoft 365 Copilot rollout consulting works through before anything is switched on tenant-wide.
Where this sits in the rest of your Microsoft bill
Copilot Studio is an additive line, not a substitution. It does not replace Microsoft 365 seats, it does not replace Power Automate licensing, and it does not reduce anything you already pay. So it belongs in the same review as the rest of the estate rather than in a separate AI budget: if there are inactive seats, duplicated third-party tools or leavers still licensed, that money is easier to find than it is to justify, and Microsoft 365 license optimization is the exercise that finds it. Copilot Studio capacity is billed to you by Microsoft, never through us; Microsoft Made Easy is not a reseller or CSP and does not sell, procure or discount Microsoft licenses.
When the question turns from what an agent costs to run into what it costs to build and govern, that is scoped work rather than a licensing decision, and our Copilot Studio consulting services cover the forecast, the build and the handover as a fixed fee agreed before anything starts. What we quote is that fee. What Microsoft charges is the capacity above — and the honest version of this page is that most small businesses should spend a metered month finding out which of the two is actually the bigger number.
FAQ
Questions that come up before the first pack is bought
Copilot Studio is sold tenant-wide, not per seat. Microsoft's published list price is $200 per pack per month for 25,000 Copilot Credits, read from Microsoft on 15 September 2026. Pay-as-you-go through an Azure subscription bills only the credits you use. Your bill follows how the agent is built, not headcount.
A Copilot Credit is Microsoft's billing unit for agent work. Per Microsoft's published rates, a classic scripted answer costs 1 credit, a generative answer 2, an agent action 5, and tenant graph grounding 10 on top. One grounded response can therefore cost 12 credits where a scripted one costs 1.
It depends how full your packs are. Microsoft prices a pack at $0.008 per credit against roughly $0.01 metered, but unused credits expire monthly. Packs only win once you consistently consume about 80 percent of every pack you buy. Below that line, the meter is the cheaper instrument.
Largely no, for employee-facing agents. Microsoft states that classic answers, generative answers and Microsoft Graph tenant grounding used in Microsoft 365 Copilot, Teams or SharePoint by a Microsoft 365 Copilot licensed user are zero-rated against your packs or meter, subject to fair use limits.
On prepaid capacity, Microsoft triggers enforcement at 125 percent of purchased credits and disables custom agents until capacity is increased or resets. Agent flows are blocked earlier, as soon as prepaid capacity is exhausted. Pay-as-you-go environments avoid enforcement because the overage bills to the linked Azure subscription.
No. Microsoft enforces purchased capacity monthly and unused Copilot Credits do not carry forward. That is why buying packs ahead of measured demand wastes money: a half-used pack is a half-wasted $200. Measure a month of real conversations on the meter first, then size packs to match.
Related reading
Price the capacity, then scope the agent
Your First Copilot Studio Agent
The design choices this page prices — use case, knowledge sources, topics and actions — walked through step by step.
Copilot Studio Consulting
When the forecast is done and the agent needs building: scoped, grounded in your own data, governed, on a fixed fee.
Is Microsoft Copilot Worth It?
The other half of the Copilot bill — the per-seat licence, the break-even math, and who actually recovers the fee.
Book a free consultation
Get more from the Microsoft tools you already pay for
Tell us what you’re trying to fix — a report, an approval process, an intranet, a Copilot rollout. We scope it as a fixed-fee project, you approve, and a senior engineer delivers in 2–4 weeks.
- Fixed-fee scope agreed before any work starts
- Senior, US-based Microsoft engineers — no handoffs
- Typical 2–4 week delivery
- 30 days of post-delivery hypercare included
The fastest way to reach us is the form — tell us the task and we’ll reply within one business day.