An identity graph that multiplies activable audiences

The challenge

A media group could only activate its first-party audiences on logged-in web traffic. The same readers also came back on mobile, in apps and without logging in, but with other identifiers, so most of the audience could not be reached and most of the inventory could not be sold on these segments.

What we built

An identity graph that links all the known identifiers of the same person: first-party IDs, cookies, mobile advertising IDs, hashed e-mails and ad platform identifiers. From any list of identifiers, the matcher finds all the other identifiers of the same people, ready to be targeted, or excluded, in Google Ad Manager or in a data clean room.

• Daily and automatic. The graph is rebuilt every day from the data platform and the ad server’s log files. Each step only starts if the previous one finished successfully that day.
• Consent first. Hashed e-mails that are not opted in are filtered out before they enter the graph, and removed from it when consent changes.
• Built for volume. A graph database builds and maintains the links; flattened relational tables serve the matching, so lists of millions of identifiers are matched quickly.
• Driven by configuration. Identifier sources are mapped in an admin interface, without code changes.

The result

• 6.2 million people reconciled
• ~5 identifiers per person (median)
• ×3 to ×4 activable audience
• +40% to ×2 sellable impressions

Enriched audiences reached mobile web and in-app inventory that the raw audience could not reach at all. The same held for audiences built with advertisers in a data clean room. The eCPM distribution kept the same shape after enrichment: more inventory to sell, at comparable quality.

What comes next

The same graph can detect in-market users, score high-value readers, rebuild households for targeting and frequency capping, filter invalid traffic and measure cross-device attribution.