Analytics Instrumentation
The practice of defining which user actions to track and, critically, which contextual attributes (event properties) to attach to each action. The quality of instrumentation determines whether a product team produces measurements or insights.
Measurements vs insights
A measurement is a raw observation: a data point in a database. It records that something happened.
An insight is a measurement with context: a data point plus the properties that explain why it happened and for whom. Insights enable action; measurements enable entertainment.
Examples from Crystal Widjaja's Gojek playbook:
| Measurement | Insight |
|---|---|
| Map loaded | Map loaded + drivers visible (2) + surge active + city + user had voucher |
| Power users do 4× more bookings | GoFood power users are 2× more likely to use free-shipping discounts on high-GMV baskets |
| User cancelled subscription | User cancelled citing “too much product” — solvable with a pause button, not a win-back campaign |
The insight version enables a specific intervention. The measurement version enables a dashboard.
Instrumentation spec
The instrumentation spec is a formal document mapping each user-facing action to its event name and required property set. It should be written before a feature is built, not after.
Diagnostic. Open the spec; count properties per event row. Events with zero or one property signal a broken analytics culture — the team is tracking that things happen, not why.
Writing a spec. For every event, ask “if I were to do this action, why would I and why would I not?” Then instrument for those reasons as properties.
Analytics maturity stack
Crystal Widjaja’s recommended tool progression by stage:
| Stage | Tool | Notes |
|---|---|---|
| Single warehouse | Google Data Studio | Free; adequate for early stage |
| Multiple databases, SQL team | Metabase | Open-source |
| Mobile event tracking | CleverTap | CRM + events integration |
| Scaled analytics | Amplitude | Funnels, retention at scale |
| Data piping / CDP | Segment | Normalise events across tools |
| Experimentation | Eppo | Auto-generated experiment dashboards |
Tool sophistication should lag product complexity. A six-month integration project that produces no growth learnings is the worst outcome.
Why most analytics efforts fail
Root cause: teams treat metric gathering as entertainment — seeing an interesting number and moving on without changing a decision. Crystal Widjaja’s formulation: real news is information that changes what you do. If it doesn’t change what you do, it was entertainment.
Proximate cause: poor instrumentation. Events without properties cannot produce insights. You can see that something happened but not why — limiting your ability to segment, hypothesise, and test.
Related
- Crystal Widjaja on Growth at Gojek, Analytics Failure, and Scrappy Experimentation — primary source; full case studies from Gojek
- Edwin Chen on AI Data Quality — data quality as the upstream precondition for reliable analytics
- Ronny Kohavi on AB Testing and Experimentation — rigorous experiment design once you have clean, propertied events
- Brian Tolkin on Product-Ops Integration, Experimentation at Scale, and the Kernel of Truth — experimentation at scale; instrumentation quality as prerequisite