Identity Resolution at Scale: Lessons from 200M+ Member Programs
By SocialHub.AI Team
Real-time identity resolution across 15+ data sources. Lessons from managing profiles at massive scale.
Identity is the foundation everything else stands on
Before an AI agent can decide anything, it has to know who it's deciding for. Identity resolution — collapsing a person's many touchpoints into one durable profile — is the unglamorous foundation of the entire retention loop. Get it wrong and every downstream decision inherits the error: the wrong customer gets the wrong message, points land on the wrong account, and the personalization that should build trust quietly erodes it.
At small scale, identity feels easy. At the scale of 200M+ members across more than fifteen data sources, it becomes one of the hardest engineering problems in the stack — and the one most worth getting right.
Lesson one: resolve in real time, not overnight
Batch identity resolution — reconciling profiles in a nightly job — was acceptable when marketing ran on a calendar. It is not acceptable when agents make decisions on live intent. A customer who buys in-store at lunch and opens your app that evening should be one person to your system immediately, not tomorrow morning.
Real-time resolution is what lets the loop close at the speed of behavior. When in-store POS events and online activity merge into a single profile as they happen, the Decide node is working with reality instead of a stale snapshot. The lag between action and recognition is where relevance goes to die.
Lesson two: more sources is a feature and a liability
Pulling from fifteen-plus sources gives you a richer profile, but each new source multiplies the ways identities can collide or fragment. The same person appears as a guest checkout, a loyalty member, an app login, and a service ticket — and any two of those can be wrongly merged or wrongly split. The discipline that matters is conservative matching: be aggressive about enrichment, cautious about merges, and explicit about confidence.
The cost of a bad merge is asymmetric. Splitting one person into two profiles wastes some relevance; merging two people into one can expose private data across customers. At scale, you design for the failure that's expensive, not just the one that's common.
Lesson three: privacy and identity are the same problem
A signal-loss world makes first-party identity more valuable and more sensitive at once. Apple's App Tracking Transparency is estimated to have cost Meta around $10B in a single year (Meta investor guidance) — a reminder that identity built on third-party signal is fragile, and identity built on consented first-party relationships is durable. Resolution at scale has to treat consent, data residency, and access as first-class inputs, not afterthoughts.
Done right, this is an advantage. The brands that can resolve identity cleanly on their own data don't depend on the platforms that keep changing the rules. Their foundation is theirs.
Why this scale is provable, not theoretical
These lessons aren't abstractions. The McDonald's China program runs on this kind of resolution at 200M+ members — and the business outcome was dramatic: member sales grew from 5% to 41% of the total, and monthly active members rose 129%. None of that is possible without identity that holds together in real time across every channel a customer uses.
Resolution is the Capture node doing its job well enough that everything after it can work. A loop built on shaky identity compounds errors; a loop built on clean identity compounds value.
Build the foundation, then build on it
If your profiles fragment across channels, no amount of clever agent logic will save the experience — the foundation has to come first. The good news is that identity is solvable at scale, and getting it right unlocks every node downstream.
We've done this at the scale of hundreds of millions of members. Book a demo and we'll walk through how your sources would resolve into a single live profile, where your current fragmentation is costing you, and how a pilot would prove the foundation before you build the rest of the loop on it.