AD-22 · AI Engineering / Measurement.
Baseline and measurement design before attributing impact to AI
IMPLEMENTEDSUPPORTED
- The decision
- Establish and preserve a baseline before attributing improvements to AI agents or workflows.
- Why it mattered
- Without a prior baseline, any improvement claim lacks a comparable reference.
- The trade-off
- Measuring first delays early claims and requires data discipline, but makes it possible to distinguish real improvement from perception or normal variation.
- What changed
- Improvement claims require comparison against a prior reference instead of relying only on narrative.
- The outcome
- No generic AI productivity improvement is published; results must be demonstrated per initiative against its baseline.
Reusable principle
Improvement without a baseline is narrative, not evidence.
Related transformation
Engineering Capability SystemSummary of a documented architecture decision. Method, mechanism and supporting evidence are not published.