Mixpanel

Mixpanel stores events with arbitrary properties alongside mutable user profiles, and everything in the UI — Insights, Funnels, Flows, Retention — is a query over that one model. Events are point-in-time actions with properties describing the interaction; user properties describe the person, and both feed cohorts and filters. Its differentiator has always been speed of ad-hoc segmentation: sliced-and-diced breakdowns return without pre-aggregation, which keeps analysis exploratory rather than dashboard-bound. Lookup tables enrich event properties with metadata, and Group Analytics (an add-on) adds a B2B account-level entity on top of the user model.

The implementation decision that bites people is identity management. Mixpanel stitches anonymous and known activity via a distinct_id, and there are two incompatible ID-merge APIs: Simplified ID Merge (the default for new organizations since April 2024), which merges on explicit $user_id / $device_id properties with no limit on device IDs per user, and Original ID Merge, which uses $identify / $merge / $create_alias events and hard-caps a user’s identity cluster at 500 IDs. You cannot switch APIs once a project contains data, so pick deliberately — and if events arrive via Segment, RudderStack, or mParticle, their configuration has to match the API version your project uses.

Warehouse Connectors are the other big architectural choice: instead of (or alongside) SDK tracking, Mixpanel ingests events, user profiles, group profiles, and lookup tables directly from Snowflake, BigQuery, Databricks, Redshift, or Postgres. Mirror mode keeps Mixpanel fully in sync with warehouse changes — including updates and deletes — on everything except Postgres, which supports append-style syncs. That makes a “model in dbt, serve in Mixpanel” setup genuinely workable, with the warehouse remaining the source of truth. Session Replay ties recorded sessions to the same event stream, so you can watch the sessions behind a funnel drop-off; the free plan includes a 10K-replay monthly quota and paid plans start at 20K.

Pricing is event-based: 1M events per month are free (the free plan caps saved reports at 5), Growth is self-serve at $0.28 per 1K events past the free allowance with volume discounts, and Enterprise is sales-quoted with unlimited events. Mixpanel’s own docs push you to plan your tracked events before instrumenting anything, and that advice is earned — a governed, intentional event schema is what keeps the fast segmentation useful. Implementation details live at docs.mixpanel.com.

Use it when product and growth teams want fast, self-serve behavioral analysis without waiting on a data team, or want to point it at events already modeled in the warehouse. Look past it when you need built-in feature flagging and experimentation in the same tool, or full control over where event data physically lives.