Copilot Credits vs Billed Charges: Why the Two Figures Differ

Copilot Credits and billed charges are not the same measurement, so they are not supposed to match exactly. Credits are the consumption unit Microsoft meters usage in; billed charges are money on an Azure subscription for a closed billing period. They come from different portals, cover different scopes, settle at different times, and one of them can include usage the other never bills. Microsoft's own guidance is direct about which to trust for reconciliation: "Use the monthly billing record for reconciliation, not the usage dashboards" (Microsoft Learn, updated 21 August 2026).

This guide explains the documented reasons the two figures diverge, separates them from explanations you can only confirm against your own records, and gives you a checklist to work through before you write the difference up. All Microsoft behaviour described here was checked against first-party documentation on 8 September 2026; every claim links to its source.

Quick answer: they answer different questions

• Microsoft 365 admin center → Copilot → Cost Management answers who is consuming, under which policy, and are we near a limit? Everything in this view "is expressed in Copilot Credits, not currency" (S1).
• Azure Cost Management answers what amount is billed or accrued for the linked Azure subscription? It shows cost by billing scope, subscription, resource, meter and invoice period (S1).
• The invoice answers what is payable for this billing period? Its line items are finalized billing summaries (S1).

Small differences between the operational view and the bill are expected. Microsoft states that the admin center "shows daily usage and can include non-billable usage, while the Azure bill reflects finalized, billable consumption at the close of the month" (S1). A gap is not automatically an error.

The comparison that explains most disagreements

Sources for the rows above: unit, purpose, scope, period, status, Azure refresh and capacity packs from S1; admin-center tab refresh intervals from S2. Both pages are Microsoft-maintained and change; the intervals and tab names above are as documented on 8 September 2026.

Documented behaviour that produces a difference

These are documented distinctions to check, not diagnoses of your tenant. Where the source itself conflicts, the uncertainty is stated rather than resolved by guesswork.

1. The unit itself. One view counts credits, the other counts money. The cited sources do not establish a universal multiplier for turning a dashboard credit count into your bill. Microsoft's sizing example for pre-purchase plans introduces a per-credit figure with the word assuming — it is an assumption inside that example, not evidence of a universal rate (S4). Pre-purchased commit units are distinct from metered usage credits: they "pay down qualifying costs in US dollars (USD)". Applicable pricing, discounts, purchasing currency and billing method matter. Use the monetary source record rather than importing the example's assumption into your review.

2. The line item is labelled for a different service. Copilot Cowork, Copilot Studio and Work IQ API consumption "appears under the Microsoft Copilot Studio service rather than separate services for each experience". Microsoft answers the question directly: there is no separate Cowork line item currently, and to attribute cost to a specific service you separate it by Azure subscription or resource group (S1). Finding "Microsoft Copilot Studio" instead of "Cowork" is not, by itself, evidence of a billing error.

3. Timing. The admin center refreshes its Overview tab every 4 hours and its Consumption tab every 2 hours (S2); Azure billed amounts update daily and usage can take up to 24 hours to appear (S1). Two screenshots taken minutes apart can legitimately disagree.

4. Exports are snapshots. "Exported reports are generated as a point-in-time snapshot, while dashboard views continue to retrieve live data. For this reason, exported counts and dashboard counts might temporarily differ" (S2). If your spreadsheet came from an export and your comparison came from the screen, that alone can explain a gap.

5. Some usage is metered but not billed. Where a running task exceeds a per-user limit inside a group spending policy, it can complete without interruption, and that excess "doesn't count toward the policy limit, isn't billed (at Microsoft's sole discretion), and doesn't appear as consumed credits" in the Cost Management dashboards (S2).

6. People moved between groups. Microsoft says credits consumed before a move "remain billed against the previous spending policy". But the same page gives conflicting descriptions of the user view: its group-move section says Total credits used includes all assigned policies during the period, while its later Consumption section says only current-policy usage is displayed at user level (S2). Do not assume either display rule is universal. Record the exact view and capture time, compare policy-level evidence and escalate an unexplained difference rather than treating this documentation inconsistency as a proved cause.

7. Prepaid capacity does not appear on the Azure cost view. "Azure Cost Management doesn't show consumption against Capacity Packs" (S1). If your organisation draws on capacity packs, real consumption exists that the Azure cost view will not show you.

8. Which source is consumed first. If you select both capacity packs and an Azure subscription with P3 prepaid credits, Microsoft documents the order as capacity packs, then P3 prepaid credits, then pay-as-you-go (S2). This is conditional on the configured billing method, eligibility and available balances; pay-as-you-go overage requires pay-as-you-go to be enabled (S1). The absence of a new usage charge does not mean the consumption was free: it may have drawn down previously purchased capacity.

9. Actual versus amortized. These are two metrics over the same data, and Microsoft is explicit that "actual cost and amortized cost views show different total numbers" (S3). Actual shows the purchase as it appears on the bill; amortized distributes a pre-purchase across the plan term based on credit consumption (S1). Comparing one person's amortized export with another's actual export will not reconcile.

That P3-specific explanation does not mean every prepaid purchase has an amortized view. S3 concerns supported reservation and savings-plan benefits, not ordinary Azure Marketplace purchases (with a stated exception for supported VM software usage reservations). It also says Cost Analysis does not support amortized reservation costs for an MS-AZR-0003P pay-as-you-go subscription. Check the supported purchase and billing scope before giving amortized-view instructions. Do not substitute S3's even-per-day VM reservation example for S1's consumption-based P3 description. Actual/Amortized is a cost-view metric, not the tool's Finalized bill/Provisional cost source-status declaration; record both separately in the source reference or notes.

10. A pre-purchase is its own charge. A Copilot Credit pre-purchase plan "is billed as a separate charge and appears as its own line item on your invoice" (S4). A large one-off line in a month with modest usage may be that purchase, not consumption.

11. Not everything Copilot is in this billing model. The Microsoft 365 admin center Cost Management experience covers Cowork and the Work IQ API; Copilot Chat, SharePoint agents and the Copilot Retrieval API, Copilot Studio pay-as-you-go, and non-Copilot pay-as-you-go services are documented and managed elsewhere (S5). Charges you cannot find in one dashboard may be governed by a different one entirely. Per-user Microsoft 365 Copilot licences are fixed subscription licensing, which usage-based billing complements rather than replaces (S5).

Possible explanations that need your own records

The following are common and plausible, but nothing in Microsoft's documentation can confirm them for your tenant. Treat each as a hypothesis until your own records settle it:

• The two figures cover different date ranges — a calendar month against an invoice period that does not align with it.
• The two figures cover different scopes — one subscription, resource group or policy versus all of them.
• One figure is tax inclusive and the other is not, or they are in different currencies.
• A credit note or adjustment landed in one view and not the other.
• Someone changed a spending policy or billing method mid-period.
• The comparison mixes two different Copilot products billed through different routes.
• The figures were captured at different times, and one has since refreshed.

The distinction matters more than it sounds. "Microsoft documents that exports are point-in-time snapshots" is a fact you can cite. "Our finance export was probably taken before month-end close" is an inference about your own process. Keep them visibly apart in whatever you write, because the person reading it later cannot tell which is which from the number alone.

An investigation checklist

Work through these in order before concluding anything:

1. Write down both figures with their exact source, including which portal, which view, which metric, and the date and time you captured them.
2. Align the periods. Confirm both cover the same start and end date, not "August" versus "the August invoice".
3. Align the scope. Same subscription, resource group, billing scope and policy set — or note explicitly that they differ.
4. Check the metric. If the Azure figure is amortized, get the actual view too, and vice versa (S3).
5. Look for the shared service label. Cowork, Copilot Studio and Work IQ consumption can arrive under Microsoft Copilot Studio (S1).
6. Identify one-off purchase lines such as a pre-purchase plan (S4).
7. Ask whether prepaid capacity is in play, and remember that capacity-pack consumption is not visible in Azure Cost Management (S1).
8. Confirm tax treatment and currency on both sides.
9. Investigate unexpected in-month usage now, but keep open-period monetary figures provisional. Use the closed monthly billing record before calling the comparison finalized billing reconciliation.
10. Use the monthly billing record as the monetary figure (S1), and keep the credit view as usage context.
11. Write down what is still unexplained, with an owner and a date. An unexplained remainder recorded honestly is worth more than a forced match.

How to record the result without overstating it

Whatever you conclude, the write-up should show the reader which figure is money, where it came from, when it was captured, and what remains unresolved.

The free Copilot Cost Review is built for exactly that shape. You declare one period, one currency and one financial basis — Finalized bill or Provisional cost — so nobody has to guess whether the figures have settled. Each row has fields for its source reference and source date; missing evidence is flagged. Usage credits go in a separate list and are never converted to money, summed across contexts or used to allocate cost. Rows needing follow-up get a decision of Investigate or Correction needed, a note, an owner and a due date; the report lists them as open actions, separately from the unreviewed decisions count.

If you supply a control total, a matching figure is reported as arithmetic agreement only. It does not mean the bill is verified. The Copilot cost report template walks through the full structure and a worked example.

What this tool does and does not do here

It does not reconcile Microsoft systems. It does not connect to Microsoft 365, Azure, Copilot or any billing API, it does not read your tenant, and it does not import native Microsoft billing exports. It has no way to fetch either of the figures you are comparing — you supply both.

It does not convert credits to money. No rate is applied anywhere, in either direction.

It does not authenticate an invoice. It checks that the rows you entered add up to the control figure you entered.

What it does is hold the comparison you made, with its sources, its dates, its unresolved rows and its limitations, in one dated artifact you can hand to someone else. That is a smaller claim than reconciliation, and it is the honest one. CertPilot's general position on what its evidence can and cannot prove is in the methodology and in what CertPilot is and is not; the same boundary applies here.

For the question of who should be doing this investigation each month and who signs off the result, see the Copilot spend ownership checklist.

Frequently Asked Questions

Why is Copilot Cowork billed under a Microsoft Copilot Studio line?

Because Cowork, Copilot Studio and Work IQ API consumption is billed through a single shared Copilot Credits service labelled Microsoft Copilot Studio, and there is currently no separate Cowork line item. To attribute cost to a specific service, separate it by Azure subscription or resource group (Microsoft Learn, checked 8 September 2026).

How much money is one Copilot Credit?

Do not use a universal multiplier to infer your bill. Microsoft's P3 page describes commit units paying down qualifying costs in US dollars and gives a pricing assumption inside a sizing example (P3 guidance). Its separate dashboard guidance says task credit consumption varies with models, context, runtime and tools (dashboard guidance). Both were checked 8 September 2026. Neither establishes a universal conversion for your purchasing arrangement; use its actual monetary billing record.

Which figure should finance use?

The monthly billing record. Microsoft's guidance is to "use the monthly billing record for reconciliation, not the usage dashboards" (Microsoft Learn, checked 8 September 2026). The credit view is valuable operational context — who is consuming, under which policy — but it is not the payable amount.

Do the Actual and Amortized views disagree?

They can, legitimately, for supported benefit purchases. Microsoft states that "actual cost and amortized cost views show different total numbers", because amortized spreads a purchase across its term while actual shows it as it appears on the bill (Microsoft Learn, checked 8 September 2026). Apply the purchase/subscription qualifications above, compare like with like and label the metric separately from whether your source is finalized or provisional.

Does CertPilot reconcile these two systems for me?

No. CertPilot has no connector to Microsoft 365, Azure or Copilot and cannot read either figure. The Copilot Cost Review is a browser-local worksheet for the figures you supply: it does the arithmetic, names the gaps, records your decisions and prints the result. The investigation is yours.