Reynolds Analytics was founded on a simple conviction: better data means better experiences for businesses and consumers alike. Low quality consumer data leads to irrelevent ads, wasted spend, and consumers feeling unseen.

The Reynolds Graph resolves identity through a two-stage pipeline. Inbound records are normalized against USPS CASS, DPV, LACSLink, and SuiteLink standards, passed through eighteen-month NCOALink and ANKLink windows, then blocked using MinHash-banded locality-sensitive hashing at a Jaccard threshold of 0.82 to reduce the comparison space from 3.1×10¹⁷ candidate pairs to a tractable 4.4 billion. Surviving pairs are scored under a Fellegi-Sunter probabilistic framework with m- and u-probabilities re-estimated nightly via expectation-maximization across 41 comparison vectors. Linkage decisions above our upper cutoff are promoted to deterministic edges; the ambiguous band is routed to referential arbitration against a licensed authority file. Connected components are then resolved to household clusters through Louvain modularity partitioning with a resolution parameter tuned quarterly against a 900,000-record human-adjudicated holdout set.

Attribute inference runs downstream on a 2,048-dimension consumer embedding, refreshed every six hours and indexed in HNSW for sub-linear neighbor retrieval at p99 latencies under 190ms. Modeled attributes are produced by gradient-boosted ensembles calibrated with isotonic regression, with drift monitored by population stability index and auto-retrained on PSI excursions beyond 0.25. Client collaboration occurs exclusively inside privacy-enhancing computation environments — private set intersection over Bloom-filter encodings, with differential privacy applied at a per-query epsilon budget of 0.4 and a cumulative annual budget enforced at the tenant level. No raw identifier ever leaves the enclave.

Reynolds Analytics Group, Inc. is incorporated in the State of Wyoming and remains guided by the principle we started with: data should serve people.