Why identity matters
A customer who browses on their phone, comes back on a laptop two days later, and finally buys on a tablet, looks like three different users to a basic tracking system. The conversion attaches to the tablet session. The phone and laptop touchpoints are stranded, wrong channel attribution, wrong audience targeting, wrong recommendation engine input.
The identity graph fixes this by stitching all three sessions to one person. Now the conversion has the full journey behind it: the discovery ad on the phone, the comparison shopping on the laptop, the purchase on the tablet. Every downstream system, attribution, retargeting, lifecycle email, operates on the full picture instead of fragments.
How identity stitching works
Two complementary techniques:
- Deterministic matching. Same login across devices. Same email entered in a form. Same hashed phone number. High confidence, low coverage.
- Probabilistic matching. Same IP at similar times, similar device fingerprint, overlapping behavioral patterns. Lower confidence per match, but covers most users who don’t log in.
The best graphs combine both. A purchase from a logged-in user becomes a deterministic anchor that promotes probabilistic neighbours into the same identity cluster.
What identity unlocks
Almost everything downstream:
- Cross-device attribution. Mobile-to-desktop journeys stop looking like two separate users.
- Audience deduplication. Your “10K monthly active users” is actually 10K humans, not 15K device-sessions.
- Lifecycle targeting. Email triggered by browsing behavior across all the customer’s devices.
- AI agents on customer data. An AI assistant answering “what did this customer do?” needs one identity, not a smear of device IDs.
What breaks identity graphs
The same forces that break tracking:
- ITP / ETP shorten cookie lifespans, breaking long anonymous sessions
- Ad blockers prevent identity tags from firing
- Privacy enforcement prevents persistent fingerprinting in many jurisdictions
- Cross-domain limitations make stitching between owned properties harder
Modern identity graphs compensate with server-side first-party identifiers, hashed-email matching, and graceful degradation when individual signals are missing.
Common mistakes
- Conflating users on shared devices. Family iPad accidentally merges three customers into one identity. Use behavioral signals to detect the split.
- Treating identity as a one-time job. Identity assignments shift as new signals arrive, your graph needs to be a live system, not a batch export.
- Forgetting the privacy dimension. Strong identity graphs need strong consent records to match. Build them together.
FAQ about Identity Graph
What is an identity graph?
An identity graph is a unified record of every device, session, and identifier that maps to a single customer. It stitches anonymous cookies to logged-in IDs over time so each user appears as one person, not many.
How does identity resolution work?
Two complementary techniques: deterministic matching (same login across devices, same email entered) and probabilistic matching (shared IP, similar device fingerprint, overlapping behavior).
Why does my analytics show too many users?
Usually because identity resolution is weak. Each device looks like a separate user. Cross-session anonymous activity does not stitch to known customers. A proper identity graph collapses these into single profiles.