Before an AI pilot begins, the team should have a clearly defined service purpose, metrics with a clear meaning, and data showing how the service currently performs against those metrics. The sequence is important: define the purpose first, then design the metrics, then collect the data that establishes the starting position.
The cited guidance sets out that sequence. It advises defining the purpose of the service before designing metrics, and collecting data that shows how the service performs against them. For an AI pilot, this baseline provides a reference point for later assessment; it does not by itself promise or prove an improvement.
What the baseline should contain
| Baseline element | Readiness check |
|---|---|
| Service purpose | The team can state what the service is for clearly enough to guide metric design. |
| Metric meaning | Each selected metric has a clear meaning rather than an ambiguous label. |
| Current performance | Data shows how the service performs against the selected metrics before the pilot starts. |
How to check it
A pre-pilot review can ask for:
- A clear statement of the service purpose.
- The selected metrics, with the meaning of each one recorded.
- The starting data for those metrics.
If a metric cannot be explained clearly, its meaning is not yet established. If no starting data is available, the team cannot show current performance against it. In either case, the baseline is not yet complete enough to support a later comparison.
What the team must still confirm
No universal target, evaluation period, or pass/fail rule is verified by the cited guidance. The team must confirm those pilot-specific choices, along with how the baseline will be documented and whether the available data can support the intended comparison.
Any later claim of improvement should be assessed against the documented starting position and the measurement basis used for the pilot. Without that record, the team cannot verify the claimed change from the baseline alone.