Identity Stitching

Identity stitching associates observations recorded under different identifiers with a chosen analytical identity. For example, a trusted login can connect earlier browser activity to an account. The resulting profile depends on the linking policy; it is not proof of a unique human.

If browser A logs into account U1 and later U2, mapping all A history to the last account misattributes activity. On a shared device one browser serves several accounts, so a rule that ties each browser to a single account is wrong there. Define reset rules, such as replacing the anonymous identifier at logout so the next visitor on that browser is not linked to the previous account, and time-bounded link rules, such as linking a browser to an account only for a set period after the login. Retain link evidence, and leave ambiguous observations unresolved when the policy cannot decide.

Report whether a metric counts installations, accounts or stitched profiles. Linking anonymous viewers with logged-in purchasers changes the denominator as well as overlap: when browser visits are merged into the accounts that later bought, a conversion rate of purchasers per visitor divides by fewer visitors than the same data counted by installation. Record the policy version and whether links learned later are applied retrospectively. Preserve permitted source identifiers and propagate retention or deletion decisions to derived mappings.

References:

• Segment identity practices