The problem with scattered identifiers

A modern shopper interacts with a store across several channels: they open the site on a laptop,
browse products in the app, click through from an email and complete the purchase in a mobile
browser. Without identity resolution, every one of those touchpoints is a separate user in your
systems.

The typical picture before stitching:

  • Cookie ID on desktop → anonymous profile A
  • Authenticated user_id → profile B
  • Mobile device ID → profile C
  • Email click → profile D

Three real people look like twelve users. Personalization runs on partial data, and analytics
overstates unique visitors.

Deterministic vs probabilistic matching

Type Mechanism Accuracy Coverage
Deterministic Matching email, phone, user_id High (~100%) Signed-in users only
Probabilistic Device fingerprint, IP, behaviour Medium (70–90%) Reaches anonymous users

Most CDPs combine both: they look for a deterministic match first and fall back to a probabilistic
model when there is none.

The moment that matters: stitching at sign-in

The highest-value moment is the point of authentication. The system sees that the current cookie ID
belongs to a registered user_id, and stitches the entire preceding anonymous history onto the known
profile.

Visitor views 15 products  →  anonymous cookie id: abc123
Visitor signs in           →  user_id: 99876
CDP: cookie id abc123 = user_id 99876 → merge
Recommendations now account for all 15 views

Important: if the stitch does not happen at the moment of sign-in, the anonymous history is
lost as soon as the cookie expires or the shopper switches browsers.

Impact on personalization

Once profiles are properly merged, personalization improves along several axes:

  • Recommendations: they account for the full history, not just the current session
  • Segmentation: the user lands in the correct RFM segment, with every purchase counted
  • Frequency capping: the system knows the user has already seen a pop-up and does not repeat it
  • Attribution: the purchase is credited to the channel of the first visit rather than the last touch

Challenges and limits

  • Shared devices: when a household shares one browser, stitching different family members produces a polluted profile
  • Privacy regulation: under GDPR and CCPA/CPRA, linking identifiers across channels needs a lawful basis, and in many cases explicit consent captured through a consent management platform
  • Degradation without first-party data: when the share of registered users is low, deterministic matching only covers a small slice of the audience