Skip to content
SundAIFree call20 min
Menu
Browse all guides

How to budget AI tokens per resolved customer case

Tokens measure model use. A case budget also captures retries, paid tools and the time someone spends checking the answer. Compare workflows on the same unit of finished work.

Follow a single case from start to finish

Imagine a support assistant that classifies a message, looks up a policy and drafts a reply. Those steps can trigger several model calls before a customer receives an answer. Give the case an internal reference and add up all its calls, including retries. Record usage counts and tool charges against that reference. You can build a cost report without copying customer messages into the reporting spreadsheet.

Calculate the cost of the work you delivered

For a simple weekly illustration, suppose 80 cases close and the model and tool charges for handling those cases total €12. The technical cost is €0.15 per closure. If review totals four hours at an assumed €30 an hour, the combined cost is €132, or €1.65 per closure before infrastructure and other overheads. Keep carry-over cases in a separate group until you know their final cost.

Check what happens when you shorten the input

Use the current rates for your chosen model and distinguish input, cached input and output where the provider prices them separately. Try passing only the relevant policy section. Compare the resulting answers on the same sample, including unusual requests. A lower token bill loses its value if staff must spend longer correcting an answer or the customer has to ask again.

Sources