Ask employees and the answer is clear. Ask the controller and it goes quiet. The gap between felt benefit and measurable result is the biggest open question about AI in companies.
The gap in numbers
McKinsey surveyed 1,719 participantsU between 4 May and 8 June 2026. 80 %U report personal productivity gains from AI. An "AI high performer" is anyone who attributes at least 5 % of EBIT to AI and calls the effect significant. That share is flat at around 6 %U. 32 %U of respondents prefer to build their own solutions rather than buy. 39 %U expect AI to reduce headcount.
In Germany, budgets are growing at the same time. According to Bitkom, 45 % of companiesU plan to spend more on AI in 2026. A third see their own existence at risk without AI.
The Danish study by Humlum and Vestergaard offers one explanation for the gap: so far, the time saved hardly shows up in wages or hours. It gets absorbed into the working day.
The calculation
Monthly net effect:
measured time saved × full-cost hourly rate × number of people − (licences + rollout and training spread over months)
An example with clearly stated assumptions:
| Item | Assumption | Per month |
|---|---|---|
| Team | 20 people | |
| Time saved | 1.12 hours a week (2.8 % of 40 hours), 22.4 hours a week across the team | about 97 hours |
| Full-cost rate | 50 euros per hour | about 4,850 euros |
| Licences | about 25 euros per person | −500 euros |
| Rollout and training | 3,000 euros one-off, spread over 12 months | −250 euros |
| Net | about 4,100 euros |
That amount is only potential. It becomes a result once the time gained flows into measurable work: more cases handled, shorter lead times, less overtime or lower external costs.
The workflow
- 01StartBaseline: time, volume, quality before the start
- 02PersonPilot with a clear task and group
- 03SystemSame measurement after four to six weeks
- 04SystemNet effect after licence, rollout and review costs
- 05CheckPositive and reliable?no: switch task or stop
- 06ResultDecide what the time gained will be used for
Limits and risks
- Self-reports: Estimated time savings are systematically too high. Measure with a stopwatch, ticket data or time tracking.
- Hidden costs: Review effort, error correction and workflow maintenance belong in the calculation.
- The research: Many ROI studies are funded by vendors. Rely on your own measurements.
How to start
- Choose a task with measurable output, such as tickets, proposals or reports.
- Record time and volume for two weeks without AI.
- Repeat the measurement after four to six weeks with AI.
- Decide in advance what the time gained will be used for, and check after three months.