Honest automation ROI is realized value minus fully-loaded cost, measured against a baseline you recorded before you started, and 'hours saved' only counts if you can say where those hours went. Most reported automation ROI fails one of those three tests, which is why leadership has stopped believing the numbers.
The full cost side
| Cost | Where it hides | Typical share of total |
|---|---|---|
| Build / setup | Visible, usually budgeted | 20-35% |
| Licenses & inference | Visible, grows with volume | 15-25% |
| Human review time | Uncounted 'quick checks' on outputs | 15-30% |
| Maintenance & fixes | Prompt drift, integration breaks, model changes | 15-25% |
| Failure cleanup | Rework when outputs were wrong | 5-15% |
The honest value side
- Hard savings: costs that left the P&L, a cancelled tool, an unfilled backfill, reduced overtime. Provable.
- Capacity: hours redirected to named work. Real if you can name the work; fiction if you can't.
- Quality: fewer errors, faster cycle time, better customer outcomes, measure the metric, not the sentiment.
- The test for 'hours saved': multiply by fully-loaded rate only if headcount, overtime or backlog actually changed.
A method that survives scrutiny
- 1Record the baseline before building: volume, hours, error rate, cycle time.
- 2Measure the same numbers after, same definitions, same period length.
- 3Subtract fully-loaded costs, including review and maintenance time.
- 4Classify the result: hard savings vs capacity vs quality, no blending.
- 5Report the range, not the best case, and update quarterly as costs continue.
