Data Engineering › Collection & Instrumentation

Instrumentation

Adding code that records what users and systems do, so the data exists at all.

Also known as: event instrumentation, analytics instrumentation, tracking implementation, adding tracking, telemetry instrumentation

Instrumentation means adding code that records what users and systems do, so that the data exists to analyze later. If nobody instruments “added to cart”, no pipeline, however well built, can tell you how many people did it.

// client-side event
analytics.track("checkout_started", {
  cart_value: 129.5,
  currency: "USD",
  item_count: 3,
});

# server-side event, emitted when the business action really happens
events.emit("order_completed", user_id=user.id, order_id=order.id, total_cents=order.total_cents)

Decide before you code

• What questions will this data answer? Start from the decisions people need to make.
• What events and properties are needed? Write them in a tracking plan with consistent names.
• Where should the event fire? See client vs server tracking: browsers can be blocked, offline or faked, while servers see what really happened.
• Who is it about? An anonymous visitor ID, a user ID, a device. Plan how they’ll be linked (identity resolution).

Doing it well

• Fire events at the moment of truth: order_completed after payment succeeds, not when the button is clicked.
• Include context: timestamp (with the time zone or in UTC), app version, platform, and a unique event ID for deduplication.
• Send it asynchronously and safely. Tracking must never slow down or break the product. Batch events, retry on failure, and swallow tracking errors.
• Test it. Check that events fire once, with the right properties, in a debug view or in the raw data. Bugs here are silent, and are noticed months later when a chart looks odd.
• Respect privacy and consent (consent at collection). Don’t collect personal data you don’t need, and honor opt-outs.
• Treat it as code: review it, version it and test it.

Observability vs analytics

The same word covers technical instrumentation (metrics, traces, logs for operating a system) and product instrumentation (events about user behavior). The principle is the same. Think about the question you’ll need to answer first, and capture the data to answer it (telemetry pipelines).

Missing instrumentation can’t be backfilled. History you didn’t record is gone, so it pays to think ahead.