Financial Services
Entity resolution for financial crime investigations
A Tier 1 North American bank
The bank could not reliably tell when two customer records were the same person. Retail, commercial and wealth each held their own version of a client, and the financial crime team was reconciling them by hand during investigations.
Why it mattered
KYC and AML obligations do not bend to how a bank organises its systems. An investigator who cannot see a whole customer cannot assess that customer, and the gap shows up in exactly the situation where it is least affordable: a regulator asking how a decision was reached.
The Pawa approach
Started with the matching problem rather than the platform. Profiled the real data to find why records failed to match — name conventions across languages, addresses captured to different standards, identifiers held in fields never meant to carry them. Match rules were then tuned against known cases the investigations team already understood, so the results could be judged by people who knew the right answer.
What was built
- Profiling pass across the contributing source systems to quantify the real match failure modes
- Deterministic rules for identifier matches, probabilistic scoring for the remainder
- A mastered customer record with survivorship rules agreed per attribute, not per system
- Stewardship queue for the scored middle band, so ambiguous matches go to a person rather than a threshold
- Lineage retained from the mastered record back to every contributing source
Delivered capability
Investigators could pull a single customer view across lines of business, and could show which source each attribute came from.
Outcome
Entity resolution moved from a manual reconciliation step inside investigations to a maintained capability the financial crime team could rely on. The lineage back to source meant a match could be explained rather than asserted.
This engagement was delivered by our principal during fifteen years at Informatica, before PaWa Data Solutions. It is professional background, not this firm's client work. The client is not identified, and no quantitative outcome is claimed.
Related solutions
Not sure which one you need?
That is what the Data Health Check is for. Two to three weeks, and you leave knowing what to fix and in what order — whether or not you work with us.
Book a Data Health Check