Cheap models win most automations

This summer we shipped UX for 10+ AI products at Humbleteam. Biggest lesson — for routing, tagging, and summarising, the smallest and cheapest ChatGPT tier worked as well as the premium models while cutting run costs by multiples.

Users never asked which model we used. Nobody cared about ChatGPT vs Claude. They cared that the result was clear, fast, and reliable.

Here is a simple side-by-side we like to demo:

Prompt: “Summarise this customer email into 3 bullets — issue, severity, next step.”

Cheapest nano model: Issue: card top-ups fail on weekends for Czech users. Severity: high — payments blocked. Next step: roll back PSP routing change and notify affected users.

Expensive model: Issue: weekend card top—ups failing for CZ merchants. Severity: high — revenue at risk due to blocked payments. Next step: revert PSP update and send update.

Different price, same decision :)

For many backend automations, pick the smallest model first. Upgrade only when you can prove the lift.