Lesson 1 / 25

Why AI ROI Estimates Go Wrong

Common optimism traps.

Five recurring mistakes

AI automation business cases often overstate returns because they assume the AI replaces whole jobs rather than parts of tasks, ignore human review time, treat freed hours as cash, forget running and maintenance costs, and assume instant, full adoption. They also present a single precise number for something highly uncertain. The result is projects that "succeed" technically but disappoint financially, and eroded trust in the next proposal. This course builds estimates that avoid these traps.

Why estimates fail and how to frame them

Most AI business cases fail on assumptions, not arithmetic; good framing comes first.

Figure 1.1 — Failure patterns, vocabulary and framing.

Traps and corrections

Use it to review any AI business case.

trap                                       correction
"the AI does the job"                       decompose tasks; automate shares of steps
ignores review / exception handling        add review minutes and error handling
hours freed = money saved                  model realisation (attrition, redeployment)
only counts API fees                       full cost of ownership incl. people and upkeep
instant 100% adoption                      adoption S-curve and training time
one precise number                         ranges, sensitivity, Monte Carlo

Ask "what would make this wrong?"

For every key assumption, write down what evidence would show it is too optimistic.

Quick check: Which assumption most often inflates AI ROI estimates?

That adoption takes time
That costs exist
That freed staff hours automatically become cash savings
That reviews are needed

Answer

That freed staff hours automatically become cash savings — Freed time only saves money if it is realised.