Operations
Cost per call
What a single AI interaction costs you — the number that decides whether a feature is viable at scale.
Cost per call is driven by the model chosen and the tokens in and out. The spread between the cheapest and most capable models is enormous — often more than 50×.
The engineering discipline that follows is routing: send straightforward work to cheap models, escalate only genuinely difficult cases to expensive ones, and decline below a confidence floor. A well-designed tiered system can cost a fraction of one that sends everything to the best model.
Meter this per feature and per customer from day one. Discovering your unit economics from a monthly invoice is a bad way to learn them.
Related