Tool · For people who own an AI feature
Sprint tool
Estimate what an AI feature will cost to run, how much work it needs to maintain, and what that leaves for delivery. Every assumption is printed next to the result it affects.
1 · Cost to run quick mode
The expected bill is the average times the users. Model API calls only in this mode.
More: usage spread, revenue, margin, cost limit
Typical and heavy users give scenarios next to the expected bill. Interactions per month.
Revenue per user is ARPU. Current margin is the margin before this AI cost. The cost limit is the spending appetite you agreed for this feature.
2 · Work to keep it alive
Human time only. Automated run time is not counted. Leave empty if there is none.
Guidance: what practitioners report
Test cases are eval cases: inputs with an expected result, used to check the feature after model or prompt changes. Practitioner heuristics, with their own units: 10 to 20 outputs reviewed a day; about 30 minutes a week after setup. The hours in the result come from your entries, not from these.
3 · This sprint
Other obligations outside this stream, hours per sprint
0 means you assume no interruptions. The result will say so.
Optional: what past sprints show
Count items, not points. Comparable in size if possible.
4 · One thing you are not sure will work optional
Give it a time-box and a checkpoint
Cost to run
Enter a cost per interaction.
Work to keep it alive
Enter the maintenance inputs.
Plannable hours
What past sprints show
Enter items finished per past sprint.
Checkpoint
Enter the thing under exploration and the days to give it.