# What happens when you don't use your AI commitment

Many AI contracts trade a discount for a commitment: you promise to spend a set amount over a year or term, and the vendor prices your usage lower. Spend less, and you pay the difference anyway.

## How a shortfall happens

- **The commitment was sized on a forecast**, often a launch that slipped or a migration that didn't happen.
- **Usage moved**: to a cheaper model, another provider, or caching that cut tokens.
- **The commitment counts only some spend**: one provider's API, but not the same models through a cloud marketplace.

## See it coming

Carry your recent daily spend to the end of the commitment period:

> projected = spent so far + last 28 days' daily average × days left

If the projection is short of the commitment by 10% or more with a quarter to go, it's time to act. [Agreements](/docs/agreements) does this per term, year or month, and warns owners, admins and finance once per period when it happens.

## What to do about it

1. **Move eligible spend onto it.** Workloads on another provider, or the same models bought through a different channel, may count if you route them through the committed account.
2. **Bring planned work forward.** Evaluation runs, backfills and batch jobs you'd run next quarter.
3. **Negotiate early.** Vendors would rather extend a term or roll part of a shortfall into a renewal than lose the account. Ask before the period ends, not after.
4. **Size the next one on actuals.** Use your effective rate and real pace, not a forecast. See [What to bring to an AI vendor renewal](/guides/negotiate-ai-contracts).

## The opposite problem

Passing a commitment can be expensive too, when overage is billed at a premium. The same projection shows when you'll cross it, and whether the next tier would have been cheaper.
