Business KPIs
The measures the business runs on — which architects should understand, influence and instrument.
Definition
Key performance indicators are the small set of measures an organisation manages against: conversion rate, average order value, churn, activation, gross margin, cost to serve.
Why architects need them
To justify work in the organisation's currency. "Reduce p99 latency by 200 ms" is an engineering statement. "Reduce checkout latency by 200 ms, which historical data suggests moves conversion by roughly X%" is a business case. The same work, and only the second gets funded.
To choose between options. When two designs are technically comparable, the one that moves a KPI is better, and knowing which KPI matters most this quarter is a genuine input to design.
To detect incidents. Business metrics are frequently the fastest and most reliable outage detector available, because they are immune to the failure mode where every technical dashboard is green and the product is broken.
To avoid building the wrong thing well. An architecture optimised for a metric nobody manages against is effort spent for nothing.
The chain to construct
Connect the technical measure to the business one explicitly:
Latency → bounce rate → conversion → revenue Deployment frequency → time to market → competitive position Availability → transactions completed → revenue and trust Unit cost → gross margin → viability at scale
The links are estimates, and estimates stated openly are far better than the alternative, which is technical work justified only by technical arguments and therefore perpetually deprioritised.
The instrumenting obligation
If a KPI matters, it must be measurable in near real time and segmented — by platform, region, cohort and client version. A monthly figure cannot detect an incident, and an unsegmented total hides an entire platform being broken.
Owning that instrumentation is legitimately an architectural responsibility, because it depends on event design and data flow decisions made long before anyone asks for the dashboard.
Failure scenarios
- Engineering metrics with no business connection, so investment is argued on faith.
- KPIs nobody instrumented, available monthly from a finance system.
- Optimising a proxy metric that diverges from the outcome it stood for.
- Business metrics on a dashboard nobody alerts on — the ownership gap where product owns the dashboard and engineering owns the alerts.
Interview question
"How would you justify a six-week reliability investment to a commercial stakeholder?"