SpendAssureSplit spending by how it is priced
AI spending usually combines several pricing models, and each moves for a different reason. One trend line across all of them hides that: a seat price rise, a new team adopting an API and a cloud commitment all look like the same growth.
| Pricing model | What drives it | Where the figures come from |
|---|---|---|
| Seats | Seat count, price per seat and contract dates | Order forms, invoices, licence reports |
| Usage | Tokens, credits or requests, times the rate | Usage exports, cloud cost exports, invoices |
| Commitments | The committed amount and how it is drawn down | Cloud agreements and consumption reports |
| Card subscriptions | Individual plans bought outside procurement | Expense reports and card statements |
Forecast each part from its own driver
Seats follow contracts. Keep the contracted quantity until a decision to change it is recorded, and apply the change from the date the contract allows. Planned hiring or a rollout to another department is an assumption: state it.
Usage follows adoption. A run-rate over recent weeks, with a trend fitted on recent complete months, is a reasonable base. Damp the trend so that one fast quarter does not compound for a whole year. New products have little history, so show how much of the forecast rests on them.
Commitments follow their schedule. A consumption commitment is not spend in itself, but a shortfall at the end of the term can be. Forecast usage against the commitment, not instead of it.
The FinOps Foundation describes the same quantity times rate approach for AI costs, with a weekly or monthly rolling forecast. Source: FinOps Foundation, How to Forecast AI Services Costs in Cloud.
Add known price changes from their effective dates
Several AI vendors changed prices or pricing models in 2026, and a forecast built only on last year’s invoices will miss those changes. Record each change with its effective date and the contracts it affects, and apply it from that date. The changes below are as the vendors documented them on 26 October 2026.
GitHub Copilot moved Business and Enterprise to usage-based billing on 1 June 2026. Seats still cost $19 and $39 per user per month, and each seat now includes the same value in monthly AI Credits, pooled across the organisation or enterprise. Unused credits do not carry over to the next month, additional usage beyond the pool is enabled by default and code completions do not consume credits. Forecast the seats as a fixed line and usage beyond the pool as a usage line. Source: GitHub announcement, 27 April 2026 and usage-based billing.
Claude Enterprise is priced as a seat fee plus usage. Anthropic lists $20 per seat per month, billed annually with a 20-seat minimum, and bills all usage separately at API rates. Organisations on older Enterprise seat types move to this model at their next renewal, so that renewal changes how the cost is calculated, not only its level. Source: Claude pricing and Enterprise billing.
ChatGPT Enterprise has no public list price; it is agreed with OpenAI’s sales team. OpenAI’s pricing documentation covers credit-based billing and Enterprise agreements with usage-based billing in US dollars, and notes that some Enterprise customers stay on a legacy rate card until they move to token-based pricing. The Services Agreement says price changes on OpenAI’s pricing page take effect fourteen days after they are posted. Source: ChatGPT pricing documentation and OpenAI Services Agreement, section 6.6.
Microsoft 365 Copilot lists at $30 per user per month in the US and €26 on Microsoft’s euro-area sites, excluding VAT, on an annual subscription. Microsoft’s price update for Microsoft 365 suites from 1 July 2026 excluded standalone Copilot. Some newer agent features are billed on usage through Copilot Credits, on top of the user licence. Microsoft’s cost management API for those credits does not include prices, invoices or forecasts, so the rates have to come from your agreement or price list. Source: Microsoft 365 pricing update and Copilot Credits.
Convert currencies at a stated planning rate
Usage rates for AI models are usually published in US dollars, while many budgets are approved in euros or another local currency. Keep each amount in the currency of its source and convert it for reporting at a rate you name.
Fix a planning rate when the budget is approved and use it for both the budget and the forecast. A variance against budget then shows a change in spending, not in the exchange rate. The currency effect can be reported on its own line: invoices at the rate actually charged, compared with the same invoices at the planning rate.
The European Central Bank publishes euro reference rates on working days at around 16:00 CET. The ECB says they are published for information purposes only and discourages their use for transactions. They work as a documented planning reference; what you owe is set by each invoice. Source: ECB euro reference rates.
Show ranges and scenarios, not false precision
Low, base and high assumption sets show how sensitive the forecast is to usage growth. They are not a statistical confidence interval unless the method actually produces one, and the report should say which it is.
A scenario answers a conditional question: what if we remove 300 seats at the anniversary, or accept a renewal quote? A scenario should not change the approved budget or the base forecast until a decision is recorded.
Check the method before you trust it
Test the method on your own history. For each of the last six complete months, make the forecast you would have made at the end of the previous month and compare it with what was actually billed. Compare the error with a simple baseline, such as repeating the previous month. If the method does not beat the baseline, use the baseline.
Quote accuracy with its period and method, for example weighted absolute error over six months, one month ahead. A percentage without that context is not a promise about future bills.
What a forecast report should show
- The approved budget, its version and who approved it
- Spend to date, split into invoiced and estimated amounts
- The forecast to period end, with its version and assumptions
- The variance to budget, with its main drivers named
- Data that is missing or late, and which figures it affects
Where SpendAssure fits
SpendAssure keeps the budget, the forecast and scenarios apart, lists the assumptions behind each forecast version and measures accuracy on the organisation’s own history. See budgets and forecasts in the product.