Identity resolution is most of the data project
The streaming architecture is a week. Deciding that customer 4471 in one system and CUST-4471-B in another are the same customer is the other five months.
Anita Deshmukh, Chief Technology Officer ·
The unglamorous majority
Data platform proposals are written about ingestion, modelling and serving. The effort lands somewhere else: reconciling identifiers across systems that were never designed to agree, and deciding what to do when they cannot.
Resolve at ingest, not at query
Pushing identity resolution into the query layer means every consumer re-implements it, slightly differently. Resolve once at ingest and publish the conformed identity, even when that means the ingest path is more complicated than the diagram suggests.
Plan for unresolvable records
Some records will not match. A platform that has no answer for those quietly drops them, and the first anyone knows is a reconciliation gap six months later. Decide the policy and surface the count on a dashboard someone reads.