Glossary entry · Governance
Layer 2 of AI value management: whether AI is producing work that can be trusted, used, and improved, judged by useful and discarded output, review burden, model fit and quality risk rather than call volume.
A productivity claim built on output nobody uses is not a claim. This layer is where leaders decide to route models, tune prompts, strengthen evaluation, set human review thresholds and stop low-quality use cases.
Starting measures in the KPI library include Token Efficiency Ratio (tokens in accepted output against all tokens consumed), Coding-Agent Cost per Merged Change and Model Routing Efficiency. The library is thinner here than at other layers, so most organisations add workflow-specific acceptance and rework measures. Owners are usually AI operations, engineering and risk.
Output quality is the second of the five layers of AI value management. Organisations that measure portfolio strategy before establishing output quality end up making investment decisions on adoption stories rather than outcome evidence.
The quality floor is the practical test at this layer: the standard an output must meet to count at all. The verification cost of checking output belongs in the cost of the use case, not outside it.
Put it to work
Browser tool. Twenty questions across the five layers of AI value management, scored against concrete behaviours, ending in a maturity level and the two layers to work on first.
Excel template. A starter scorecard that takes KPIs from the library and maps them to the five layers of AI value management, with baseline, target and owner for each.
Browser tool. Compare up to four model setups on what each successful outcome costs, including the people who fix failures. Runs in your browser; nothing is sent anywhere.
