Business Metrics
Measuring what the system exists to do — frequently the fastest and most reliable outage detector available.
Definition
Business metrics count outcomes rather than mechanisms: orders placed per minute, signups completed, messages delivered, payments settled, videos started. They are observability instruments even though they are usually owned by product.
Why they detect incidents better than technical metrics
Technical metrics measure what you thought to measure. Business metrics measure whether the system is doing its job, and are therefore immune to the most embarrassing failure mode in operations: every dashboard green while the product is broken.
That happens constantly, and the causes are mundane — a broken client-side deployment, a third-party script failing, an expired certificate on one path, a feature flag misconfiguration, a payment provider silently declining. In every case error rates are normal, latency is fine, and orders have gone to zero.
A single alert on "orders per minute has dropped more than X% against the same time last week" catches an enormous range of failures that no technical alert covers, including all the ones nobody imagined.
How to alert on them well
- Compare against the same period historically, not against a static threshold. Traffic has daily and weekly shape, and a fixed number is either noisy at night or blind in the afternoon.
- Account for seasonality and known events — a marketing campaign, a public holiday, a sale.
- Alert on rate of change as well as level, so a sudden drop is caught before the absolute number looks unusual.
- Segment by dimension — platform, region, client version, payment method. A total that looks normal can hide iOS being completely broken.
- Set the sensitivity to catch a partial failure, not only a total one. A 15% drop is a real incident and rarely alerted on.
Failure scenarios
- Nobody owns them for alerting, because product owns the dashboard and engineering owns the alerts, so nothing pages.
- Static thresholds on a metric with strong daily shape.
- Only totals, so a segment failing completely is hidden by the rest.
- A metric that lags — a daily aggregate cannot detect an incident.
- Treated as a substitute for technical monitoring. They tell you something is wrong, not what.
Interview question
"Every technical dashboard is green and revenue has dropped 40%. What are the likely causes and what would have detected it sooner?"