12 Mar 2026 · Practice

Instrumentation without noise

Most AI-assisted product performance analytics failures begin months earlier — when teams ship trackers for curiosity instead of decisions.

Rows of illuminated server racks

Event dictionaries expand the way attic boxes do: quietly, then all at once. By the time an assistant is asked to summarise “what changed this week,” it is summarising trivia alongside signal.

Start from the decision, not the SDK

Before adding a property, write the product question it answers and the review forum where that answer will be used. If you cannot name both, archive the idea. In Product Signal Intelligence we call this the “question before tracker” gate.

Freeze seasons beat endless polish

United Kingdom product teams often ship continuously; schemas need intentional freezes. A two-week freeze after a major release gives analysts and models a stable surface and reduces false drift alerts.

What to delete

Unused events with no consumer in ninety days, duplicate page-view variants, and client-side “debug” properties left in production are the usual candidates. Deletion is a product decision — treat it with the same care as a feature sunset.

← Back to journal