← home
RESEARCH · GUIDE

AI spend forecasting for CFOs.

A finance guide · July 26, 2026

By the LLM CFO team

AI spend is hard to forecast because it is tied to product usage, not headcount. But a structured forecast is still possible. The key is to build it from use cases up, not from a single top-line growth rate.

1. Build the baseline

Start with the last three months of actuals, reconciled to the provider invoice. Break it down by team, feature, and model. This baseline is your anchor. Everything else is a change from it.

2. Add use-case growth

Forecast each major use case separately. A customer-facing assistant and an internal coding agent will have very different growth curves. Ask product for usage assumptions: new users, queries per user, and model mix.

3. Plan for price changes

Model prices change. Add a sensitivity line for price increases or decreases on your top two models. This prevents surprises when a provider announces a new pricing tier.

4. Scenario planning

Build three cases: conservative, base, and aggressive. The conservative case assumes no new use cases and stable usage. The aggressive case includes planned launches. The base case sits in between.

5. Reconcile and update

Every month, compare actuals to forecast and explain the variance. Update next month's forecast based on what you learned. Over time, your assumptions get better.

Related

← Back to llmcfo.com