Many AI features are no longer priced per seat but per use: credits, tokens, conversations or actions. That aligns cost with value — and makes cost harder to predict.
How pricing is shifting
- Salesforce moved Agentforce from about $2 per conversation to Flex Credits in May 2025: $500 per 100,000 credits, with one agent action using 20 credits ($0.10 per action), sold pre-purchased, pre-committed or pay-as-you-go.
- Many martech and creative platforms now include AI "credits" that run out, then bill extra.
- In June 2025 Cursor moved its Pro plan from request limits to usage-based pricing. On July 4 it apologized for unclear communication and offered refunds of unexpected usage charges from the previous three weeks: a surprise bill caused by a pricing change, not by misuse.
Why bills surprise
- An agent loops or retries; a campaign scales to millions of variants.
- Teams discover a new feature and use it heavily.
- Credits are shared across teams with no owner.
- Prices or credit rates change at renewal.
Controls borrowed from cloud cost management ("FinOps")
- Owner per contract and per use case.
- Budgets and caps monthly, with alerts at 50/80/100%.
- Unit metrics: cost per asset, per conversation, per qualified lead — not just total spend.
- Forecasts updated monthly with finance.
- Contract terms: price protection, credit rollover, caps on overage rates, the right to see usage data.
Make it visible
Tag AI usage spend in the budget as its own line, even when it is bundled in a platform invoice. The budget planner's "AI tools and usage" line is there for this.