KPIs · AI Engineering & Product

AI Engineering and Product: KPIs

The six measures for shipping learned behaviour, why cost per call is the wrong denominator, and the counter-metrics that catch a feature hitting its numbers by pushing work elsewhere.

Markdown. No sign-up, no email.

Deterministic software is right or wrong against a specification. A learned system is better or worse against a distribution, which means the usual engineering metrics do not transfer. A test suite at 100% green says nothing about whether answers got worse this week.

Every measure below is chosen because it survives that difference.

The six that matter#

MeasureDefinitionTargetHow it gets gamed
Eval pass rate, per categoryHeld-out set results broken down by case typeNo category below its floorReporting one blended score, which hides the category that collapsed
Regressions caught pre-releaseShare of quality drops found by the eval rather than by a customerAbove 90%Shrinking the eval set until nothing fails
Cost per successful outcomeTotal spend divided by outcomes that actually workedFallingUsing cost per call instead, which rewards a cheap answer that fails twice
Escalation rate to a humanShare of cases the system hands overInside a stated bandSuppressing escalations to look autonomous
p95 latency95th percentile end to end, not the meanUnder the product's stated limitQuoting the average, which hides the experience people abandon
Time to roll backMeasured by rehearsal, not estimatedUnder 15 minutesNever rehearsing, so the number is a guess

Cost per successful outcome is the one people get wrong. Cost per call makes a cheap model look efficient while it fails a third of the time and the retry, the escalation and the apology all cost more than the better answer would have. Divide by outcomes, not by requests.

The escalation rate is a band, not a target to minimise. Too low means the guardrails are not firing and the system is answering things it should hand over. Too high means the feature is not carrying its weight. Both directions are failures, and only one of them looks like success on a dashboard.

Two counter-metrics#

Counter-metricCatches
Support contact rate on journeys the feature touchesA feature that hits its latency and cost targets by producing answers people then have to query
Human review time per escalated caseEscalation that technically works but hands over cases so poorly framed that reviewing one costs more than doing it from scratch

What is deliberately not measured#

  • Model accuracy in isolation. A model at 94% inside a process that routes 40% of cases to a person anyway has an unclear effect on the outcome. Measure the product.
  • Tokens consumed. An input, not a result. It belongs in the cost calculation, not on the scorecard.
  • Prompt count or version churn. Activity.
  • Benchmark scores. Public benchmarks measure a distribution that is not ours. They inform model selection and they do not belong in operational reporting.

Quality floors, per category#

A single pass rate hides the thing worth knowing. Each category of the held-out set carries its own floor, and a release is blocked if any one of them drops below it, regardless of the blended number.

CategoryWhy it gets its own floor
Common casesVolume. A small drop here affects the most people
Edge casesWhere quality decays first and quietly
RefusalsThe system must still decline what it should decline. Refusal quality degrades silently when a model is tuned for helpfulness
AdversarialPrompt injection and jailbreaks. Shared with Security

Cadence#

Per releaseEval pass rate by category, cost per outcome, p95 latency
WeeklyDrift against the last accepted release, escalation rate, both counter-metrics
QuarterlyRollback rehearsal, eval-set review for contamination and staleness

The quarterly eval-set review is the one that decides whether any of the rest is trustworthy. An eval set that has leaked into training reports confidence rather than quality, which is worse than having no eval set at all, because a team acts on it.

Back to AI Engineering & Product

Get new material when it is published

Everything here is free and stays free. There is no form in front of any document. If you want to know when new guides and templates go up, leave an email.

Roughly monthly. Unsubscribe in one click. We do not share your address, and we will not call you.