Amplitude
Amplitude is a product analytics platform: applications send events — "Project Created", "Report Exported", "Plan Upgraded" — with properties, and product teams analyze them to understand activation, engagement, conversion, and retention. Beyond charts, it offers user cohorts for targeting, experiments and feature flags, session replay, and connections to data warehouses.
The platform is only as good as the data sent to it. Most analytics disappointments come from implementation problems: inconsistent event names, missing properties, broken identity between anonymous and signed-in users, duplicate tracking from frontend and backend, and nobody owning the tracking plan. This page covers implementing Amplitude so its answers can be trusted, and interpreting those answers with appropriate caution.
TL;DR
- Model behavior as events with event properties, plus user properties and group (account) properties.
- Get identity right: anonymous device IDs before login, a stable internal user ID after, never an email address.
- Use the SDKs with selective autocapture, and send critical business events from the server.
- Maintain a tracking plan with owners and schemas; generate typed tracking wrappers and block unplanned events.
- Analyze with segmentation, funnels, retention, journeys, and cohorts; run experiments with predefined metrics and enough sample size.
- Respect privacy: consent, opt-out, no PII in events, and deletion requests.
Quick Example
Browser SDK setup, identity after sign-in, and a planned event:
The tracking plan entry that defines the event (and that typed wrappers are generated from):
Core Concepts
Events and Properties
Name events consistently (for example "Object Verb" in title case), describe outcomes rather than UI mechanics ("Invite Sent", not "Blue Button Clicked"), and keep property values enumerated where possible. See Event Tracking.
Identity
Before login, events carry a device ID; after login, setting a user ID links that device's history to the user, and the same user ID on other devices merges their activity. Use an immutable internal identifier — emails change and are personal data. Reset identity on logout on shared devices. Account-level (B2B) analysis uses groups to attribute users to companies.
Where to Track
Client tracking loses events to ad blockers, network failures, and consent choices; important business events should come from the server, with deduplication so the same event is not counted twice.
Governance
A tracking plan defines every event and property with descriptions, types, allowed values, owners, and platforms. Amplitude's data governance features can validate incoming events against the plan, and generated typed wrappers make it a compile error to send an unplanned event or a misspelled property. Review tracking changes in pull requests like code. See Data Contracts and Schema Evolution.
Analysis
Correlation is not causation: users who adopt a feature may already be more engaged. Use experiments to measure causal impact.
Experiments
Amplitude Experiment combines feature flags and A/B tests. Define the primary metric, guardrails, and sample size before launch; avoid stopping early when results look good (peeking); check for sample ratio mismatch; beware novelty effects; and correct for testing many metrics at once. See A/B Testing.
Privacy and Compliance
- Consent: initialize tracking or set opt-out according to the user's choices. See Consent Management.
- No PII in events: avoid names, emails, and free text in properties; use internal IDs.
- Data residency: choose the EU data center where required.
- Deletion: honor deletion requests through Amplitude's user privacy API.
- Access control: limit who can see sensitive properties and export raw data.
Best Practices
Track Outcomes, Not Clicks
A small set of well-defined business events answers more questions than thousands of autocaptured clicks.
Own the Tracking Plan
Assign owners, review changes, version the plan, and remove events nobody uses.
Test Tracking Like Features
Verify events in development and staging with debug tools and automated tests that assert the right events and properties are sent.
Reconcile With the Source of Truth
Compare analytics counts with backend and billing data regularly; large gaps reveal lost or duplicated events.
Common Mistakes
Email as User ID
Inconsistent Event Names
Declaring Victory From a Correlation
FAQ
What is Amplitude used for?
Product analytics: understanding how users move through a product — activation, engagement, conversion, and retention — through event data, plus experimentation, feature flags, cohorts, and session replay.
How does Amplitude identify users?
Events start with a device ID. When you set a user ID after sign-in, Amplitude links that device's events to the user, and events from other devices with the same user ID are merged into one profile.
Should events be tracked on the client or the server?
Both, for different purposes. Track interface interactions on the client and critical business events — purchases, subscription changes — on the server, where they cannot be blocked or lost.
What is a tracking plan?
A documented specification of every analytics event and property, with descriptions, types, allowed values, owners, and platforms, used to keep tracking consistent and to validate incoming data.
How do I avoid mistakes in Amplitude experiments?
Predefine metrics and sample sizes, avoid stopping early, check for sample ratio mismatch, watch guardrail metrics, and be cautious with many simultaneous metrics and novelty effects.
Related Topics
- Product Analytics — Measuring product usage
- Event Tracking — Designing analytics events
- Mixpanel — Another event analytics platform
- A/B Testing — Running trustworthy experiments
- Consent Management — Honoring user privacy choices
- Data Contracts and Schema Evolution — Governing event schemas