Analytics Instrumentation: Track Events, Not Page Views

Page views tell you where people went. Events tell you what they did, which is what product decisions require. The difference between an analytics setup that answers questions and one that produces arguments is almost entirely in the naming.

Design the taxonomy first

• Adopt an object-action convention — Order Completed, Report Exported, Invite Sent — and apply it without exception.
• Pick a case convention and stick to it. "signup", "Signup", and "sign_up" will otherwise coexist forever as three different events.
• Start with the twenty events that describe your funnel and your core value moments. Resist tracking everything; volume isn't insight.
• Write it down in a tracking plan that engineering and analytics both work from, and review new events against it.

Properties are where the value is

• Attach the context you'll want to segment by: plan, role, source, device, and the relevant entity ids.
• Use consistent property names across events, so filtering works the same way everywhere.
• Include values for things you'll want to sum — order total, item count — rather than deriving them later from joins.
• Never put personal data or credentials in event properties; they land in third-party systems with broad internal access.

Implementation discipline

• Instrument server-side for anything that must be accurate — purchases, subscriptions, conversions. Client-side events are lost to ad blockers and poor connections.
• Client-side is fine for interaction detail where approximate counts are acceptable.
• Identify users consistently, and handle the transition from anonymous to identified so pre-signup behaviour connects to the account.
• Version your events: if the meaning changes, use a new name rather than silently redefining history.

Keep it trustworthy

• Test tracking in QA like any other feature. Broken instrumentation is invisible until someone needs the number.
• Alert on volume anomalies — an event that stops firing after a release is the most common and least noticed analytics failure.
• Audit annually and delete unused events; a taxonomy nobody prunes becomes unusable.
• Document what each event means in business terms, or two teams will interpret the same number differently.