Your company has 2 million customers. Or 1.4 million. Nobody knows.

Entity resolution is the technology that discovers when different records refer to the same real-world entity — the same person registered three times with different spellings, the same company across five systems with diverging legal names, the same address written ten ways. The result materializes in two layers: the golden record, the single, complete and trustworthy record of each entity — the holy grail that traditional MDM promised for decades — and the graph, the resolved entities connected by their real relationships: owners, addresses, phones, devices and transactions.

The difference from traditional MDM is structural. MDM relies on rigid schemas and deterministic match rules — it pairs only what is identical — and therefore takes years to ingest and match and breaks on every exception. Modern entity resolution ingests sources as they are, with disparate schemas and dirty data, and pairs them with models that reason: they recognize spellings, abbreviations, transliterations, incomplete data and even deliberate manipulation, finding incremental matches even when quality is poor. The golden record comes out in weeks, not years, with more connections found — and each pairing stays explainable: the fields, weights and confidence behind it remain reviewable and auditable.

One architectural principle defines the category: ingest once, serve multiple use cases without duplication. Instead of each initiative — customer view, risk, AML, anti-fraud, analytics — keeping its own copy of the data and its own match logic, identity is resolved once and served to all. Operation is dual: batch mode resolves the entire base; real-time mode resolves the new record on arrival — the registration that never duplicates again, and the query that arrives already linked. It is the argument the CTO buys, because it eliminates the fragmentation that makes each data project more expensive than the last.

Kavuka Entity Resolution is native to the pathologies of Brazilian data: the variations of Brazilian names, the CPF/CNPJ as a strong identity anchor, addresses in the national standard and the federal ownership registry as a source of relationships — a structural advantage over generic engines. And it is proven in production: it is the infrastructure of the Kavuka platform's own link graph, the engine that sees the facilitator, the account factory and indirect exposure in anti-fraud, AML and risk cases. Resolving identity, therefore, is no longer one data project among others — it is the prerequisite for all of them. Without knowing who is who, the customer count lies, aggregate exposure vanishes, the sanctioned party slips through a spelling and the AI initiative runs on sand. With identity resolved, the company finally knows how many customers it truly has, sees real exposure per economic group, closes the variant-spelling hideout and gains the clean base that AI models require.