User ID: The Key to Cross-Device Analytics

The mental model

A User ID stitches together a single person's journey across multiple devices and sessions. Without it, a user browsing on their phone and later buying on their laptop appears as two separate people. With a User ID, analytics tools see one continuous customer journey, shifting the unit of analysis from anonymous sessions to actual people.

How it works

You, the developer, are responsible for generating and assigning a unique, persistent, non-personally identifiable ID to each user when they create an account. When a user signs in, you pass this ID to your analytics tool via a special, reserved user_id parameter. The analytics platform then associates all subsequent events with that ID until the user signs out. The logic is specific: send the ID for signed-in users, send a null value when they sign out to disassociate the session, and do not send the parameter at all for anonymous users.

When to use it

This is essential for any application with an authentication system. Three prime examples: first, e-commerce, to track a user who adds an item to a cart on mobile and completes the purchase on a desktop; second, SaaS products, to analyze feature adoption for a single user across web and native mobile apps; third, media services, to understand how a subscriber consumes content on their TV, tablet, and phone.

When not to use it

Avoid this for purely anonymous experiences where you have no stable identifier for a user. Do not try to create a user ID from transient data like a session cookie or IP address, as this defeats the purpose of persistent identity. More importantly, do not use this feature if you cannot guarantee the ID is unique and stable for each user over time.

One canonical example

The most common mistake is treating the User ID as a generic piece of data. A developer might want a report listing actions by each user, so they create a custom dimension in Google Analytics called 'UserID' and pass the ID to it. This is a critical error. Analytics systems are not designed for custom dimensions with millions of unique values (high cardinality). This will cause reports to become heavily sampled, group most data into an '(other)' row, and ultimately corrupt your analytics. The user_id must only be sent in its designated system parameter; the platform uses it for identity stitching behind the scenes, not for direct reporting dimensions.