Data Foundations

Walled gardens and the pixel

Platforms like Meta don’t share granular audience lists with anyone. Instead, they operate as walled gardens: they know who their users are and how they behave, but that intelligence stays inside the platform. The mechanism that bridges a brand’s website and Meta’s audience data is the pixel: a small tracking code placed on the client’s website. (This is exactly why the platforms guard their data so closely: out of the hundreds of millions who follow a star like Virat Kohli, an advertiser would love to single out just the few thousand most likely to actually buy, but that targeting power stays inside the walls, so you reach the audience through the platform, not around it.)

When a visitor arrives from Meta, the pixel records their journey: which pages they visited, what actions they took, whether they booked a test drive, submitted a lead form, or just browsed. This lets Meta classify visitors into categories like window shoppers (people who skim but don’t act) versus serious buyers (people who follow the high-signal path, on a carmaker’s site like Mercedes: build-a-car → submit a lead form → book a test drive). Based on these patterns, Meta can then optimize ad delivery to find more people likely to follow the same high-signal path.

Custom audiences

A custom audience is a seed list uploaded directly to the ad platform: phone numbers, email addresses, or pixel data that the client already has. This data is uploaded to the platform (Meta, Snapchat, TikTok), not to Consumr.AI. Consumr.AI never handles first-party client data directly; clients upload it themselves, which keeps the workflow privacy-safe and compliant.

The platform matches the uploaded list to its own user base and returns the matching audience for targeting.

Lookalike audiences

Once a custom audience exists, the platform can bloat it into a much larger audience of people with similar behaviors. This is a lookalike audience. The platform lets you choose how broad to go, typically in 1% increments of the total platform population. Narrower percentages (1%) are tighter behavioral matches; wider ones trade precision for reach.

A client with 80,000 store visitors can upload that list, and the platform will identify people who behave like those 80,000 across hundreds of millions of users. Consumr.AI can then pull behavior reports from those lookalike audiences to understand the broader cohort.

Audience network, publishers, and retargeting

Beyond the core platform, ads can be served through the audience network: a collection of third-party websites and apps that allow the platform’s ads to appear on their pages. A shoe you looked at on one site follows you across the web because the pixel identified you, added you to a retargeting audience, and the network serves you relevant ads wherever you go.

Product tags extend this further: if you buy a PS5, you can be added to a games audience, and the ads you see progressively narrow, from all games to action games, from action games to specific titles, because each click adds a new behavioral tag.

Interest sources: briefs, Meta, Snapchat, TikTok

Every major platform maintains its own interest taxonomy. When a client writes a brief describing their customers, Consumr.AI’s models translate that description into the matching interest categories on Meta, Snapchat, TikTok, and others. The client doesn’t need to know the platform’s taxonomy; the brief bridges human language and platform inputs.

Coverage reality: Meta as population

Meta’s reach covers roughly 60–70% of the US population across Facebook, Instagram, and WhatsApp. This is not a sample; it is the population for a given cohort. When Consumr.AI pulls data on a defined audience, the resulting cohort represents real behavioral patterns at scale, not an approximation from a small survey.

This coverage figure also creates a denominator mismatch that matters for surveys: the 60–70% figure means that a cohort of, say, 80 million on Meta might represent 150 million in the real world. Handling that gap is the job of extrapolation. See the Respondents & Extrapolation page for how ACS/census data is used to correct for this.

RFM as a seeding method

One common way to build strong custom audiences is by segmenting a client’s existing database using RFM: Recency, Frequency, Monetary value. Clients classify their customers into high/mid/low tiers on each dimension and upload those segments as separate custom audiences. Consumr.AI can then build lookalikes from each RFM tier and compare their behavior reports, revealing, for example, whether a high-monetary-value cohort has meaningfully different interests than a low-frequency one.

RFM is a seeding method, not a segment taxonomy. Full treatment of RFM as a statistical concept lives in the Statistics section.