All insights

Data model choices that keep integrations honest

Nov 18, 2026

AML platforms fail quietly when entities, events, and decisions are poorly modelled.

Entities before pipelines

Customers, accounts, transactions, and counterparties need stable identifiers. Without that, screening and scoring invent duplicates, cases fork, and audit trails fracture.

Integration work that starts with “just push the files” usually rediscovers entity resolution the hard way.

Events should carry meaning

Ingest is not just ETL. Events need enough structure for scenarios to fire and for cases to explain themselves later. A transaction without product, channel, counterparty, or jurisdiction context is a weak signal dressed as data.

Define the minimum event shape your detection and investigation actually need — then enforce it.

Decisions as first-class data

Alerts, scores, matches, and case outcomes should be queryable. Integrations that only push raw rows leave ops blind and analytics dishonest. If you cannot ask “what did we decide about this customer last quarter?” your model is incomplete.

Honest boundaries with core systems

Core banking, onboarding, and data warehouses each own truths. An AML workspace should respect those boundaries while still assembling an operational view. Duplicating every golden record forever is not a strategy. Selective, well-keyed sync usually is.

Fatcon’s stance

Fatcon expects a coherent data model so decisions sit on the same facts your institution already trusts. The platform is an ops console for decisioning — not a second shadow core.