Contextual targeting built from a brand’s identity

The challenge

A Belgian media group with several news and lifestyle sites wanted to answer advertiser briefs with something better than age and gender targeting, and without depending on user identifiers. The idea: show a brand’s ads next to the articles whose topics match the brand’s world.

What we built

From a brief, or simply a brand name, the tool writes a portrait of the brand: history, values, audiences, territories and competitors. A language model then compares that portrait with the topics, concepts, categories, keywords and companies that the group’s audience data platform already knows about its articles.

• No invented criteria. The model can only choose values returned by the data platform’s API. Every value is checked again against that list before the segment is generated.
• Scored and explained. Each criterion gets a confidence score, and the advertiser receives a short summary explaining every choice.
• Human validation. An operator validates each family of criteria, and the advertiser approves the targeting before anything goes live.
• Activated in the ad server. The result is a contextual segment in the data platform, exported to Google Ad Manager and sold on the group’s premium inventory, in video and display.

Because the segment targets pages and not people, it works without cookies or personal data, and sensitive content is excluded.

The result

For a global energy-drink brand and its media agency, a video campaign ran as a transparent A/B test: the same creatives, formats, placements and timing, with age-based targeting (18–34) as the control and the contextual segment as the test.

• +8.74% completed view rate
• −40.77% cost per completed view
• +66% eSports fans reached
• +29.1% sports enthusiasts reached

These results are specific to that campaign: performance depends on the brand, the content available, the inventory and the creative.

What comes next

A memory per advertiser, scores fed by the performance of past campaigns, and a set of specialised agents in place of a single assistant, up to the creation of the segment in the data platform.