When a fashion game comes back as 51% female, most media planners would not blink. But NumberEight says that number is off by 31 points.
The 31-point gap
NumberEight, a London-based audience intelligence company, published a report on September 23, 2026 comparing traditional demographic targeting methods with ID-less approaches. In mobile game Fashion Battle, an IP-based model returned 51% female and 49% male. NumberEight’s own model, which does not use IP addresses, measured the audience at 82% female.
The same flattening showed across four other mobile games. Nitro Nation, a racing title expected to skew male, came back exactly 50/50 under IP data and 74% male under the ID-less model. Cooking Fever, Golf Clash and Mini Football all showed near 50/50 IP splits while the ID-less model picked up clear skews.
Why IP data keeps landing at 50/50
The report argues that an IP address usually represents a household or shared connection, not the individual consumer. IPv4 sharing, VPN adoption and restricted browser user-agent data make it harder to connect an address to one person. As a result, IP-based age and gender data tends to mirror the general population rather than a specific app or show.
A CTV dataset reinforced the point. Across 7,013 shows and more than 3 million unique users, household-level IDs returned 53% to 55% female for every genre—including horror, war and science fiction, which published benchmarks put at 54% to 55% male. Age splits were nearly identical across genres, with the 18-24 bracket around 3% and the 55+ group around 48% to 50% regardless of content type.
What this means for your targeting
This is not just a methodology debate. It affects budgets, creative and campaign measurement. If your demographic data never changes when the audience should, you may be buying reach against the wrong signal.
- Audit whether your provider’s age and gender splits vary by app, genre or show.
- Validate segments against known audience skews before trusting a media plan.
- Test ID-less or contextual models alongside IP-based targeting, then compare.
- Treat self-reported and deterministic data as clues, not ground truth.
The bigger shift
NumberEight framed ID-less methods as using contextual signals, cohort grouping, on-device AI and private aggregation instead of persistent identifiers. The report concedes that deterministic IDs can be more precise for a single individual in some cases, but says they carry their own failures around misreporting and fragmented identity graphs.
Emma Raz, Commercial Director at NumberEight, said the key issue is not whether you can match an identifier but whether the data attached to it tells you something meaningful. CEO Abhishek Sen added that AI-led omnichannel models can focus on signals rather than identity as a proxy.
Source: PPC Land
