The UA Data Problem: Why 300M Behavioral Device IDs Beat Platform Guesses

If you run mobile UA, you've lived the same frustration for years: you pour money into Meta, TikTok, or Google, and the audience settings feel like a blunt object. "People interested in casino games." "Puzzle game enthusiasts." "Mobile gamers 25–44."

Those labels aren't based on what players actually do. They're inferences — scraped from clicks, video watch-time, and friends-of-friends graphs. On a good day, they tell you someone tapped a gaming creative once. They never tell you whether that person installed, played past day 1, or spent a single cent.

That gap — between platform guesses and behavioral reality — is the UA data problem. It's why your CPIs are trending up and your D7 ROAS keeps plateauing even when your creative gets better.

The platform model: infer from engagement, not spend

Meta and Google don't know what happens inside other apps. They see impressions on their own surfaces. If you clicked a match-3 ad in a Reel, Meta may decide you are "interested in puzzle games." That's it. That's the signal.

This is fine for lifestyle-brand advertising, where interest graphs roughly correlate with purchase intent. It's terrible for mobile gaming, where the distribution of value is extreme — the top 5% of players generate most of the revenue, and they look nothing like the median user who clicked a gaming ad last week.

The outcome: you run a $50K campaign, pay for a million impressions, get installs from thousands of users who clicked an ad but never spent in any app. Then you wonder why the cohort's D30 ARPPU is flat.

The behavioral model: score devices by what they actually do

Behavioral audience data inverts the assumption. Instead of asking "what category did this person click on?" it asks "what apps does this device actually open, use, and spend in?"

Onmo Signal runs on 300M+ device IDs across 22 countries, scored on 4B+ behavioral events from 170,000+ apps. The scoring isn't inference — it's observed app usage, app category activity, time-of-day patterns, spend signals, and cross-app intensity.

The output is a catalog of 273 pre-built audience segments:

• 122M behaviorally-active puzzle game players in the US
• 98M casual game players scored by in-app activity
• 37M social casino players, scored by spend patterns
• 25M card game players, 18M board game players
• 48M Saudi device IDs scored by gaming behavior
• Cross-cutting segments: Top In-App Spenders, Premium Hardware, Power Engagers, Mega App Users

None of these are built from "clicked an ad once." They are built from observed behavior over time.

Why the scale number actually matters

300M device IDs sounds like a marketing line. It isn't. Scale matters for two specific reasons in gaming UA.

Lookalike seed quality. Meta's lookalike engine is only as good as the seed you give it. Seed it with 5K of your own paying players and the top 1% lookalike is narrow; seed it with 500K behaviorally-identified high-spenders and the same lookalike is a different audience entirely. Match rates climb, density holds, and the algorithm has enough signal to project beyond your first-party ceiling.

Sparse-segment depth. High-ARPU categories — social casino, card, mid-core RPG — are structurally narrow. If your data source has only 2M behaviorally-scored casino players, you run out of depth inside one big campaign. Signal has 37M, and when you intersect with geography (US only, or US + Saudi), you still have working room instead of burning through your audience in 72 hours.

Why you can't get this from the platforms

Meta, TikTok, and Google can't show you the in-app behavior of the device they're about to serve. They don't have SDKs inside the games their audience is playing. They model interest from their own surfaces.

The platforms that do sit in-app — MMPs, mediators, ad networks — either:

• Keep the data locked inside their own buying surface (you can't upload it to Meta),
• Require minimum spend commitments ($50K–$100K/month) before you get audience access, or
• Build custom cohorts on request, with 2-week turnaround, instead of a self-serve catalog.

That's the structural gap. Behavioral data exists, but it's usually paywalled behind scale or trapped in the wrong buying surface.

What self-serve behavioral data changes

Signal is self-serve at signal.onmoai.com. You sign up, browse 273 pre-built segments, filter by category and geography, and upload to Meta, TikTok, Snap, or Google in one click. You seed lookalikes or direct-target inside your existing buying surface — no SDK integration, no minimum spend, no 2-week custom-build wait.

Currently free. No upload limits. Paywall expected within a couple of months.

The decision point

You will either keep paying Meta to guess at "people interested in puzzle games," or you'll start feeding your campaigns audiences scored by observed behavior — spend, session length, cross-app intensity, device tier.

The creative bar in mobile gaming UA has never been higher, and creative ROI has a ceiling set by the quality of who sees the ad. If the data underneath is interest-graph guesses, better creative just costs more per wasted impression.

Try Signal at signal.onmoai.com while it's free.