How it works

Load your Entra user export(s) and your Copilot credit / Cost Management export, then give each entity the share of the prepaid pool it funded. Prepaid credits sit in one tenant-wide pool with no way to reserve part of it, so whoever consumes earliest in the month gets the discounted rate. Settlement re-bills each entity against its own funded share instead, which takes drawdown order out of the calculation.

1. A

   Microsoft bills the tenant, not the entities

   One shared pool, consumed first, then pay-as-you-go. The order cannot be changed or reserved against, and the invoice never says which entity got the discounted credits.

2. B

   Each entity states what it funded

   Its entitlement: the share of the pool it paid for. A claim on the money, not a reservation of credits. Enter them directly, or propose a split of the pool by usage, by headcount, or evenly.

3. C

   Each bill is rebuilt from that entity’s own entitlement

   bill = min(used, entitlement) × prepaid rate  +  max(0, used − entitlement) × pay-as-you-go rate

   Drawdown order is not an input, so it stops mattering. An entity gets the prepaid rate on the credits it funded whether it consumed them on the 1st or the 28th.

4. D

   Those bills will not add up to the invoice

   An entity that funded more than it used had the difference consumed by someone else at the prepaid rate, while that someone is billed pay-as-you-go for it. The gap is the unused entitlement multiplied by the difference between the two rates.

5. E

   Decide where that difference goes

   Redistribute lends unused entitlement to the entities that went over, at the prepaid rate. Rebate returns it to the entities that funded credits they did not use. Hold centrally keeps it. Redistribute and rebate both reconcile to the invoice exactly; they differ in who benefits.

The report walks these same five steps against your own numbers. There is also a flat internal rate mode, if you would rather charge every entity one blended rate and reconcile centrally.

What you configure, and what this settles

A reasonable question: if spending policies already control Copilot spend, why is a settlement needed at all? Because the two do different jobs.

The one input the platform cannot give you is who funded what share of the pool. That is a commercial agreement between the organisations, and it is what you enter below as entitlements.

One caveat worth knowing: usage above a per-user limit “isn't billed (at Microsoft's sole discretion), and doesn't appear as consumed credits” in Cost Management. Your chargeback still reconciles, because the tenant was not charged for it, but credits-used will understate the work actually done. Source: Managing AI experiences enabled by usage-based billing.

Load your data

Where do I get these files?

1. Entra user export(s) — in the Microsoft Entra admin center, open Identity → Users → All users and choose Download users. That CSV carries the org attributes (Department, Cost Center, Job title, etc.) used to allocate each person to a unit. You can load more than one.
2. Copilot credit / Cost Management export — in the Microsoft 365 admin center, go to Copilot → Cost management → Consumption and choose Export CSV. It just needs one row per user with the credits they consumed.

Column names are auto-detected, so most export layouts work as-is — you can remap the org column inside the report if needed. Every file is parsed locally in your browser; nothing is uploaded.

Scale and limits

Measured on synthetic tenants, not estimated. There is no cap on the number of entities: 35 and 200 both settle and reconcile identically.

• Entities: no limit. Entitlements are matched by name, so the only practical constraint is that the names in your entitlement file match the org column in your export. The report tells you which ones did not match.
• Credits: no practical limit. Values are exact to 9,007,199,254,740,991 credits, far beyond any tenant.
• Workbook Users tab: capped at the top 50,000 users by credits. Beyond that the workbook becomes slow to open and Excel refuses more than 1,048,576 rows outright. The Line items CSV always contains everyone, and the totals on every other tab cover all users regardless.
• Memory: the whole tenant is held in browser memory. Expect roughly 350 MB at 200,000 users and 1.3 GB at 500,000. On a low-memory machine, filter to fewer entities first.
• Rounding: per-entity amounts are allocated so they add up to the invoice to the cent, rather than each being rounded independently.