Every department has an AI idea, and each one sounds urgent. Without a common scale, the loudest idea wins, or the one with the most convincing vendor.
A use case portfolio replaces that with one list, one scale and one decision meeting.
Two scores
Each idea gets two scores from 1 to 5.
Value reflects the size of the loss and the strategic fit. A score of 1 means a loss below EUR 20,000 a year with no link to plant objectives. A score of 5 means a loss above EUR 300,000 or a customer critical problem.
Feasibility reflects data readiness, process stability and integration effort. A score of 1 means no data, an unstable process and a complex integration. A score of 5 means ready data, a stable process and a simple connection to the line.
Priority is value times feasibility.
A worked portfolio
| ID | Use case | Value | Feasibility | Priority | Decision |
|---|---|---|---|---|---|
| UC-01 | Vision inspection, Line 3 | 4 | 4 | 16 | Pilot now |
| UC-03 | Demand forecast, 180 parts | 3 | 4 | 12 | Next |
| UC-04 | Maintenance assistant | 3 | 4 | 12 | Next |
| UC-07 | Shift report summaries | 2 | 5 | 10 | Quick win |
| UC-02 | Predictive maintenance | 4 | 2 | 8 | Prepare data |
| UC-05 | Scrap prediction | 4 | 2 | 8 | Prepare data |
| UC-08 | Energy peak forecast | 2 | 2 | 4 | Later |
A simple decision rule
- Value and feasibility both 3 or more: do, in the order of priority.
- Value 3 or more, feasibility below 3: prepare data. The idea is good, the basis is missing.
- Value below 3, feasibility 4 or more: quick win, if it is cheap.
- Everything else: later.
The rule is a suggestion, not an automatism. The sponsor decides, but now with a reason that everybody can see.
Prepare is a real decision
Predictive maintenance on four molding machines has high value: 310 hours of unplanned downtime a year. But the machines store no sensor data and failures are recorded as free text. A pilot now would fail on data, not on the idea.
The right decision is to prepare: failure codes in the maintenance system, sensors on two machines, three months of history. That costs EUR 14,000 and turns a 2 into a 3 or 4 on feasibility.
Three at a time
A plant of medium size can run one pilot and prepare two more use cases at the same time. The same few process experts, data specialists and automation engineers are needed in all of them. Ten parallel initiatives mean ten slow ones.
Review the portfolio monthly. Ideas move between columns as data improves or losses change, and stopped pilots stay on the list with the reason, so nobody starts them again by accident.