Users on a Network

The problem

In 2014, the marketing industry could see either who a user was in the real world (name, physical address, mailing lists) or what they did online (IP addresses, cookies, device IDs) — but almost never both at the same time in the same record, and stitching those two universes together usually required a deterministic login event. The patent targets the gap: how do you relate a physical address to an IP address when the two are never observed together in a single online exchange?

The mechanism

The system relates a physical address to an IP address indirectly by way of intermediate identifiers — cookies, device IDs, email hashes, account handles — that bridge the two. The method observes the IP address co-occurring with an intermediate identifier in one exchange, then independently observes the physical address co-occurring with that same intermediate identifier in a different exchange. By chaining across intermediates, the system can relate the physical address and the IP address even though they were never directly observed together. This is, in effect, graph-based identity resolution using transitive co-occurrence as the edge weight.

What it proved

Identity resolution is a graph problem, not a record-matching problem. You don't need two fields in the same row — you need enough overlapping edges across enough exchanges to make the join statistically trustworthy. The patent taught the industry that the raw material for identity is the sequence of partial observations, and that anonymity-preserving intermediaries can still carry enough signal to close the loop.

If it were built today

In 2026 the inputs have changed dramatically and the mechanism has grown up. Third-party cookies survived against expectation (Google reversed course in 2024-2025 and Chrome still allows them with a "Privacy Choice" prompt), but cookie-based identity is clearly a legacy rail; the new intermediate identifiers are hashed emails (UID 2.0, ID5, LiveRamp RampID), first-party retail-media IDs, Apple's Private Relay endpoints, Amazon Sign-In tokens, and agent-session tokens from ChatGPT/Gemini checkouts. What was clever in 2014 — the graph-chaining across intermediates — is now commoditized inside identity graphs. What is newly clever in 2026 is building this mechanism on cryptographic primitives instead of cleartext: run the edge-resolution inside confidential compute (AWS Nitro, Azure Confidential, Google TEE) so neither party sees the other's raw identifiers, emit zero-knowledge proofs that a correlation exists without revealing which intermediate carried it, and use private set intersection so two parties can discover "we share a user" without disclosing anyone who isn't in both sets. Federated learning pushes the edge-weight calculation to the edge devices themselves, so the central graph never sees raw addresses at all. The patent's anti-surveillance spirit — preserve anonymity while enabling useful correlation — maps almost perfectly to the 2026 privacy-engineering stack. It just runs on better math now.