Analytics event taxonomy: a tracking plan your data will survive
Events and parameters, naming rules that prevent the signup/sign_up/user_registered mess, a plan people actually review, and the privacy column that makes audits mechanical.
Every team has the same analytics regret: two years in, the dashboard holds signup, sign_up and user_registered, and nobody trusts any of them. The fix is not a better tool; it is a tracking plan - a written taxonomy of events and properties that the team treats like an API contract. This guide covers the conventions that keep event data usable, grounded in how Firebase Analytics structures events.
Events and properties: the only two nouns
Firebase's model generalises cleanly: an event is something that happened (login, purchase), and a parameter is detail about that occurrence (item id, price, method). After configuring the SDK you log events with logEvent(), and the SDK defines recommended events for common app types - retail and ecommerce, travel, games - with prescribed parameter names. Using the recommended events where they fit buys you reporting features for free and keeps your schema aligned with what every new hire's first tutorial uses.
Reserve custom events for what is genuinely specific to your product. A good rule: if a recommended event describes the action, use it; if you are reaching for a custom event called screen_view_2, stop and model the screen properly instead.
Naming rules that prevent the mess
- Object first, past tense or snake_case imperative - pick one grammar.
purchase_completedorcomplete_purchase, not both. Write the choice in the plan. - Never encode values in names.
checkout_step_1,checkout_step_2is a smell: use onecheckout_stepevent with astepparameter. Event names are dimensions you query; parameter values are data. - Same fact, same property name everywhere. If the paywall event calls it
plan, the purchase event must not call itsku_name. Consistent property names are what make cross-event funnels possible later. - Lowercase snake_case, no units in names.
duration_seconds, notdurationSecondsorduration (s).
The tracking plan document
The plan itself is a boring spreadsheet or table that an engineer, a product manager and a data person can all read. One row per event: name, when it fires (the exact user action, not the code location), every property with its type and allowed values, an owner, and the decision it feeds. If an event feeds no decision, cut it - every event you keep is a maintenance and privacy obligation.
Two practices keep the plan honest. First, treat changes like schema migrations: renaming an event breaks every chart built on the old name, so prefer adding over renaming, and record deprecations in the plan rather than deleting rows. Second, review new events the way you review API changes - one named approver who checks the naming rules before the pull request merges.
Privacy is a taxonomy problem too
What you must not log is easier to enforce at the property level than by auditing dashboards later. No email addresses, names, free-text fields or precise locations in event parameters, ever - identifiers belong in consented user-ID mechanisms, not sprinkled across events. The plan is where this rule lives: a column marking each property as anonymous, pseudonymous or prohibited makes the review mechanical instead of aspirational.
Start with the decisions
List the five questions the business actually asks weekly - activation, retention, paywall conversion, feature adoption, churn signals - and define only the events those questions need. Twenty well-named events with owners beat two hundred mystery events, and the taxonomy stays small enough to enforce.
Sources
Prepared with AI assistance and reviewed before publication. The events-and-parameters model and recommended events are described from Google's Firebase documentation, checked on 19 November 2026; the naming conventions and governance practices are editorial recommendations.