Segment

Segment popularized the “instrument once, route everywhere” pattern that defines the CDP category. Its spec — six calls: identify (who), track (what they did), page and screen (where), group (what account), alias (identity mapping) — is the de facto event vocabulary, and competitors ship Segment-compatible SDKs on purpose. All six calls share the same common fields, which is what makes translation to downstream tools mechanical. Sources feed a central pipeline; Connections streams events to a catalog of 700+ destinations; Unify resolves identifiers into persistent profiles; warehouse and reverse-ETL syncs move data in and out of BigQuery or Snowflake. Since the Twilio acquisition the docs live at twilio.com/docs/segment, though the product is still sold and used as Segment.

The data model is deliberately flat: spec calls with properties and traits, validated against a tracking plan in Protocols — which flags violations, can block and quarantine non-conforming events, and can rename events and properties in-flight without code changes. That simplicity is why the destination catalog is so broad, and also why teams with complex entity models eventually push modeling into the warehouse. Note that Protocols and Unify identity resolution are both gated to the Business tier — on Free and Team you get the pipeline, not the governance or the profiles. Unify’s identity graph merges userIds, anonymousIds, emails, device IDs, and custom external IDs into one profile, treats identifier casing as equivalent, and lets you enforce uniqueness on chosen IDs to prevent bad merges.

Pricing on MTUs is the classic pressure point — costs scale with audience size, not with the value you extract. Segment counts unique userIds plus unique anonymousIds that never get attached to a userId, once per user per month across the workspace. The practical gotchas are all about anonymousId churn: cross-domain tracking mints a separate anonymousId per domain, cookie clearing and incognito sessions mint new ones, and calling analytics.reset() does too — each inflating the bill. Object cloud sources (Salesforce, Stripe, Zendesk), Engage computed traits, and reverse-ETL extractions don’t count toward MTUs. There’s also an aggregate throughput ceiling on API calls per MTU, though most workspaces never hit it.

Implementation-wise, the free tier (1,000 visitors/mo, two sources, one warehouse destination) is enough to prove out the pipeline; Team ($120/mo, 10,000 visitors included) removes the source cap. Migrating away is easier than with most vendors precisely because the spec is an open standard — RudderStack and others accept the same call shapes — but Unify profiles and Protocols tracking plans don’t port, so the deeper you go into Business-tier features, the stickier it gets.

Use it when you want the widest destination catalog, a spec your whole org can share, and minimal engineering lift to light up new tools. Look past it when MTU pricing bites at scale, you want the warehouse rather than a vendor pipeline as the source of truth, or you need self-hosting for residency and control.