What does a Dynamics 365 ERP agent actually cost to run?

Your ERP AI budget has three meters, not one. Only the first bills the agent itself — the other two sit beside it, in the same business case, and they are where the surprises come from. Written for whoever has to defend the number.

8 minute read.

Meter one: Copilot Credits

Copilot Credits are the consumption currency for Copilot Studio agents. The published rate card is specific, and the units matter more than the numbers.

Classic answer — 1 Copilot Credit
Generative answer — 2 Copilot Credits
**Agent action (triggers, deep reasoning, topic transitions; Computer-Using Agents bill here too) — 5 Copilot Credits**
Tenant graph grounding for messages — 10 Copilot Credits
Agent flow actions, per 100 actions — 13 Copilot Credits
Text and generative AI tools (basic), per 10 responses (or 0.1 per 1K tokens) — 1 Copilot Credit
Text and generative AI tools (standard), per 10 responses (or 1.5 per 1K tokens) — 15 Copilot Credits
Text and generative AI tools (premium), per 10 responses (or 10 per 1K tokens) — 100 Copilot Credits
Content processing tools, per page — 8 Copilot Credits

Source: billing rates and management, Microsoft Learn — page dated 11 June 2026, rates checked 31 July 2026. Microsoft scopes the card: these rates "apply to all language models that Copilot Studio provides" and "exclude bring-your-own-model configurations, including Azure Foundry models, which are billed separately."

Microsoft's own worked example is the clearest thing in the documentation because it is small. An internal-facing agent is triggered autonomously whenever a new order arrives and makes four action calls in generative orchestration mode. Microsoft states the result as an estimated cost per day — [(4x5)] = 20 Copilot Credits.

Read the unit carefully. Unlike Microsoft's customer-support and sales-performance examples, which multiply by 900 customers and 100 users respectively, the order-processing example applies no volume multiplier at all (billing examples, Microsoft Learn, page dated 11 June 2026, checked 31 July 2026). It is the shape of the arithmetic, not a per-order rate you can scale.

The trap in that table is the premium row. Microsoft: "When an agent uses a reasoning-capable language model, Copilot Studio bills by using two billing meters: feature rate and text and generative AI tools (premium)."

The unit switches when it does. For reasoning, the premium meter is billed "per 1000 tokens, at 10 Copilot credits" — not per ten responses. So a reasoning model's cost tracks token volume, which is the input most business cases never estimate. Model choice is a cost decision before it is a quality one.

The other line worth reading twice is content processing tools, charged per page at 8 Copilot Credits. Any document-heavy ERP process meters on page count, and page counts in a manufacturing business are large.

Do this one yourself, with two inputs you already hold. Take last month's supplier-document count from your AP mailbox or document store, multiply by average pages per document, multiply by 8. Six thousand single-page documents would give 48,000 Copilot Credits before a single orchestration call is made. That is your arithmetic on Microsoft's rate, not a benchmark and not anyone's bill.

The asymmetry nobody costs in

Where the agent runs changes what the ERP tool calls cost.

Orchestration cost — Built in Copilot Studio: Billed as an Agent Action on the Copilot Studio rate card, which includes execution of the MCP server · Built on another agent client (Microsoft Foundry, non-Microsoft): "Billing from the agent client at the client's token consumption rates"
Tool execution cost — Built in Copilot Studio: "Included in the fixed orchestration rate" · Built on another agent client (Microsoft Foundry, non-Microsoft): "Billed at 0.1 Copilot Credits per tool call" — Microsoft also states it as one Copilot Credit per 10 tool calls

Source: Dynamics 365 ERP MCP server, Microsoft Learn — page dated 1 July 2026, checked 31 July 2026.

A tenth of a credit sounds like nothing until you remember that a single useful ERP task is rarely one call. Find the entity type, get the metadata, query, open the form, find the controls, set them, save — a routine sequence is many calls, and outside Copilot Studio every one of them meters individually.

There is an exemption worth knowing. Microsoft: "The following premium licenses are exempt from the tool execution billing for agents built on clients other than Copilot Studio" — Dynamics 365 Finance Premium and Dynamics 365 Supply Chain Management Premium. "Agents built on Copilot Studio continue to bill at the fixed rate per tool call for users with these premium licenses." That is an unusual shape, and it means the licensing you already hold can change which host is cheaper for you specifically.

Meter two: invoice-capture transactions

Invoice capture [GA] runs on its own meter entirely, and it is the one that catches accounts payable business cases.

Included allowance — Dynamics 365 Finance customers "are entitled to 100 invoice capture transactions per tenant per month"
Overflow — Extra Electronic Invoicing stock keeping units (SKUs) "at 300 US dollars (USD) for 1,000 transactions per tenant per month"
Behaviour — "Available on a monthly, use-it-or-lose-it basis, and customers must purchase for peak capacity"
Counting rule — "Only the captured invoices are going to be considered as valid transactions. In case the file has been filtered by the file filter setting it won't be counted"

Source: invoice capture solution overview, Microsoft Learn — page dated 25 August 2025, price checked 31 July 2026. This is the oldest published figure in the article; re-check it on the day it goes into a business case.

The included allowance is a trial allowance, not a working one. Any real accounts-payable volume is in the overflow SKU from month one, and because capacity is use-it-or-lose-it you size it against your busiest month rather than your typical one.

Meter three: AI Builder credits — beside the agent, not under it, and dated

End of sale, capacity add-ons — Date: 1 November 2025 · What Microsoft says happens: Customers without an active AI Builder capacity add-on can no longer purchase one
End of life, capacity add-ons — Date: 1 November 2026 · What Microsoft says happens: Existing add-on customers "can still use their credits linked to their active AI Builder capacity add-ons, but can't renew them, or purchase new ones"
End of seeded credits — Date: 1 November 2026 · What Microsoft says happens: Seeded AI Builder credits "will be removed" from premium licences — Power Apps Premium, Power Automate Premium and Dynamics 365 among them
Contracts already running — Date: 1 November 2026 onward · What Microsoft says happens: Add-on credits run until those licences expire, and contracts that started before 1 November 2025 are honoured for their full term

Source: end of AI Builder credits, Microsoft Learn — page dated 14 May 2026, checked 31 July 2026.

Start with what does not change. Microsoft is explicit: "The progressive end of AI Builder credits doesn't impact AI Builder features; they continue to be supported in all contexts, using Copilot Credits." The entitlement ends; the capability does not.

There is no conversion between the two currencies. Microsoft states there is "no automatic conversion of AI Builder credits to Copilot Studio Credits" and that "rates are different for each scenario" — so there is no ratio to plan against.

Context decides which currency drains. In Power Apps and Power Automate, consumption takes AI Builder credits first, then Copilot Credits, and fails only if both are exhausted. Agents are different: "agents and agent flows only consume Copilot Credits", so an ERP agent never touches this meter (AI Builder licensing overview, Microsoft Learn, page dated 14 January 2026, checked 31 July 2026).

This still reaches accounts payable, by the flow route rather than the agent route. Invoice document analysis is an AI Builder capability — Microsoft's rate table prices "Receipt, invoice, identity document analysis" per page — so what moves on 1 November 2026 is which currency pays for it. The invoice-capture transaction entitlement above is a separate Dynamics 365 Finance entitlement and is not affected by this date.

One operational detail worth carrying. Microsoft: "AI Builder credit consumption is computed regularly. Ideally, this is done on a daily basis, but it can be delayed up to five (5) or more days." During the lag the last computation still holds, so consumption continues and a block can land abruptly once the calculation catches up. Microsoft adds that such overage "only affects the current environment, and is never billed" — but the stop is still a stop.

The brakes that exist

Everything above is an open meter unless you close it. Two controls close it directly:

Per-agent monthly consumption limits. Microsoft: "You can set monthly consumption limits for individual agents in the Power Platform admin center. Go to Licensing > Copilot Studio > Manage Agents to cap credit usage before enforcement is triggered" (billing rates and management, Microsoft Learn, page dated 11 June 2026, checked 31 July 2026).
Exception limits on the Account Reconciliation Agent [PRP], set daily or monthly, with staged usage alerts as you approach the limit and a notification when a run is skipped because the limit was hit (Account Reconciliation Agent, Microsoft Learn, page dated 16 June 2026, checked 31 July 2026).

That second one is worth noting for a different reason. Microsoft's note on the page reads: "Microsoft is enhancing the Account Reconciliation Agent to give customers more configuration flexibility and predictable credit consumption. Until these improvements are released, the Microsoft team must activate this agent." Read that as early adopters finding the burn hard to forecast.

It is not the only first-party agent whose own configuration shapes spend. Of the Procurement Agent, Microsoft writes that "the agent management configurations control this cost" — which is the subject of the next section.

Three more limits sit under the platform rather than under you. A tenant-level enforcement threshold, reached a quarter above your prepaid capacity, disables custom agents: an ongoing conversation is not interrupted, subsequent invocations are rejected, and the tenant admin gets an email. Environments with their own allocation keep running, and pay-as-you-go environments are exempt because their overage bills to the linked Azure subscription.

Agent flows behave differently again — Microsoft: "When prepaid capacity is exhausted, agent flow enforcement blocks new runs rather than disabling the agent." The parent agent keeps answering. Allocating capacity per environment is the third lever, and Power Automate cloud flows sit outside Copilot Studio billing and enforcement entirely.

Any production agent without a per-agent cap is an uncapped meter. That is the sentence to take to your finance business partner.

The cost trap Microsoft states plainly

The Procurement Agent [PRP] has a configuration that quietly multiplies spend, and Microsoft documents it verbatim: "if you configure supplier communications to run on incoming emails from all vendors in a shared mailbox, and you configure impact analysis to run on vendor emails, impact analysis consumes credits when any vendor sends an email about purchase order changes" (impact analysis overview, Microsoft Learn, page dated 24 April 2026, checked 31 July 2026).

Microsoft also gives the shape of the charge: "The fixed cost applies each time the agent runs, and the variable cost depends on the number of line changes communicated." A noisy supplier base drives both halves — more runs and more lines per run.

The configuration exists to be narrowed. Microsoft describes the feature as "enabling Copilot to read emails from some or all vendors". Scope by vendor, not by mailbox, unless you have modelled the volume.

The counter-argument

"This is all included — our users have Microsoft 365 Copilot licences."

Sometimes true, and the conditions matter. Business-to-employee use of Copilot Studio agents is included in the Microsoft 365 Copilot user licence when the user is licensed and the agent operates under that authenticated user's identity. In that case, Microsoft states, "the tool calls to the Dynamics 365 ERP MCP server don't incur additional credit consumption" (Dynamics 365 ERP MCP server, Microsoft Learn, page dated 1 July 2026, checked 31 July 2026). That is a genuinely good deal and it covers a real class of scenario.

It does not cover the class most ERP optimization work falls into. An autonomous agent running on a schedule has no interactive user to operate as. A Computer-Using Agent bills at the agent-action rate and is explicitly not included. And the inclusion is subject to fair-use limits Microsoft reserves the right to change. If your business case rests on the inclusion, write down which scenarios qualify and which do not, because the answer differs agent by agent.

From credits to currency

Every rate above is denominated in Copilot Credits, not money, and that conversion is the step most business cases skip. Microsoft publishes the basis it uses for its own dollar column: "Based on pay-as-you-go billing, 1 Copilot Credit = $0.01" (AI Builder licensing overview, Microsoft Learn, page dated 14 January 2026, checked 31 July 2026).

Treat that as the pay-as-you-go basis Microsoft states, not as your contracted rate. Prepaid packs, prepurchase plans and Enterprise Agreement terms price differently, and Microsoft directs you to the Copilot Studio Licensing Guide and the Copilot Credit Guide for current commercial terms rather than publishing a pack price on the rate card.

For the forecast, Microsoft ships a calculator. The Microsoft agent usage estimator covers Copilot Studio custom agents plus the Account Reconciliation Agent and the Supplier Communications Agent, and returns total monthly credits and cost (agent usage estimator, Microsoft Learn, page dated 7 April 2026, checked 31 July 2026).

Microsoft attaches its own limit: "This estimator provides informational estimates only and makes no guarantees of final costs." It also tells you to carry a buffer above the estimate and to compare estimated against actual consumption each month in the Power Platform admin center. That is the honest answer to what an ERP agent costs — a number you measure, not one you look up.

What to do this week

Take your highest-volume candidate scenario and count the tool calls in one complete task. Not the prompts — the calls. That count, times your volume, is your first honest estimate.
Check whether any agent you run today has a per-agent monthly cap set. If not, you have an uncapped meter and setting the cap takes minutes.
Find out whether anything you depend on runs on AI Builder — invoice capture is the usual answer — and put 1 November 2026 in the plan with a named owner.

Where we would draw the line

We would not put an agent into production without a per-agent monthly cap, regardless of how modest the pilot looked, because the failure mode is a bill rather than an outage and nobody notices a bill until it arrives. We would not select a reasoning model for an ERP agent without modelling the premium meter first — that decision can dominate everything else on the rate card. And we would not build a business case on the Microsoft 365 Copilot inclusion without listing, scenario by scenario, which agents run under a licensed user's identity and which do not.

About Cognilium Cognilium is the AI optimization layer for Dynamics 365 — complementary apps that optimize the pricing, inventory, warehouse and planning decisions your ERP manages but can't optimize. Built on Power Platform, Dataverse and Azure. https://cognilium.ai · https://www.linkedin.com/company/37180269/

Worked example. Modelled from public documentation and typical operations.

If you are sizing an ERP agent business case now, we will walk the three meters against your own volumes on a call. https://cognilium.ai

What does a Dynamics 365 ERP agent actually cost to run?