All book labs
THE COST OF USEFUL AI / LABS 04 AND 05
Which system earns the work?
Fictional planning models. No provider calls or input uploads. Expected counts may be fractional.
Price the routing gate
All tasks pay for the first answer and gate. Escalated tasks also pay for the larger model. Quality is measured separately on each path. Fixed operations, human handling and setup are excluded.
tasks / month
USD / task
USD / task
USD / escalated task
% of eligible tasks
% meeting the outcome rule
% meeting the outcome rule
YOUR SCENARIO
- Variable cost
- $400.00
- Useful answers
- 9,730
- Overall quality
- 97.3%
- Cost per useful answer
- $0.0411
- Escalated tasks
- 3,500
- Failures across both paths
- 270
Try the printed exercise
Compare the three gate presets against a 97% overall-quality requirement. Base is the cheapest qualifying fictional setting at $400; the $200 liberal gate fails the requirement.