Engineering, production, governance, cost, communication · insight-0303
Estimate the value of a model with a back-of-envelope calculation before building it
- Evidence type
- practitioner heuristic
- Difficulty
- easy
- Roles
- Data scientist, Data analyst, ML engineer, Applied AI
- Industries
- aviation, cross-industry
Summary: situation, action, reason, boundary
Situation
A team proposes a three-month project to improve a no-show prediction model.
Action
Multiply decision volume by the plausible reduction in costly errors and the cost per error, compare with build and run cost, and state the assumptions; proceed, shrink or drop based on the range.
Reason
A rough upper bound on value is cheap and often decisive; it also defines the success metric.
Boundary
Prioritising ML projects, scoping proposals, budget requests.
Symptom or risk
Without a value estimate, effort goes to projects whose best case is smaller than their cost.
Exceptions
Strategic capability building may justify work beyond direct value; label it as such.
Trade-offs
Rough estimates can be wrong by a factor; use ranges and revisit after a pilot.
Pitfalls
Assuming the model's full accuracy gain converts into value, ignoring adoption and operational constraints.
Example
Hypothetical: 200,000 bookings a year, 1% fewer costly no-show errors, 80 per error: about 160,000 a year at best.