AI business cases often fail in one of two ways. Some count every possible benefit, including better data, happier customers and future use cases. Others forget half the cost, because the vendor quote covers only software.
A good business case is complete on cost and conservative on benefit, and it fits on one page.
The cost side
For a vision inspection station at the end of an assembly line, the one-time cost might be:
- cameras, lighting and an edge computer: EUR 38,000,
- integration into the line control and a reject station: EUR 27,000,
- labeling of 2,000 images and model training: EUR 25,000,
- internal hours of quality, maintenance and IT: EUR 30,000.
Total: EUR 120,000. Internal hours are the most forgotten line. Without them, the process experts are not released and the pilot stalls.
Running cost per year: license and support EUR 12,000, model care and retraining EUR 6,000. Total EUR 18,000.
The benefit side
Count only benefits that the business owner signs:
- two of six inspectors move to other tasks: 2 × EUR 58,000 = EUR 116,000,
- fewer complaints as escapes fall from 420 to 150 parts per million: EUR 40,000.
Total benefit EUR 156,000 a year. Net of running cost: EUR 138,000.
Payback: 120,000 / 138,000 = 0.87 years, or about 10.4 months. Net over three years: 3 × 138,000 minus 120,000 = EUR 294,000.
What the numbers do not include
Better data for problem solving, a documented defect history and the skills the team builds are real effects. Name them on the page, but do not count them. A business case that holds with hard numbers only survives the first critical question from the controller.
People are part of the case
Moving two inspectors to other tasks is a benefit only if there is somewhere for them to go. In the example, a new assembly cell needs staff anyway. This was agreed with the employee representatives before the pilot started.
If the benefit depends on headcount reduction, say so openly and early. A pilot that people suspect of a hidden agenda gets neither honest feedback nor acceptance.
Use it twice
The business case justifies the pilot, and it becomes the yardstick after go-live. Track the actual benefit monthly against the planned EUR 11,500 per month. Six months after go-live the run rate in the example was EUR 135,000 a year, close to plan.
A business case that is never checked again is a sales document. A business case that is tracked becomes a management tool.