Fraud detection on 50 million transactions per month, AML compliant.

Context

Detection, screening and audit could not stay in separate silos.

Fraud detection, AML screening, and regulatory reporting had each grown into a separate stack with its own catalog, its own audit trail, and its own scheduling engine. Cross-team investigations required joining datasets across three vendor consoles, each with its own access model. Audit cycles took weeks because evidence had to be reconstructed from logs scattered across the three silos.

• 50 million card and transfer transactions ingested every month.
• Five operational catalogs across core banking, card issuing, payments and partner feeds.
• 100 fraud analysts running ad-hoc investigations, 20 risk officers running periodic risk runs.
• DORA operational resilience, GDPR data subject rights, ACPR sectoral audit obligations.
• Detection models retrained quarterly with explainability requirements per ACPR guidance.

Solution

One federated query layer, one policy engine, one audit trail.

AKKO federated the five operational catalogs behind a single SQL endpoint without physically copying the data. The Catalog layer aggregated lineage, PII tags and ownership in one place. The AI layer let analysts ask natural-language questions on top, with every prompt and every generated query persisted in the audit trail. The Lab layer supported model retraining with reproducible notebook runs. The Compliance layer turned audit evidence requests into a single artefact pull.

• Query — Federated SQL across five catalogs, one endpoint, predicate pushdown.
• Catalog — Unified PII tags, lineage and ownership across all sources.
• Storage — Hot incidents land as open-format tables; cold history stays in source.
• AI — Natural-language to SQL with scope-first policy, prompt audit log.
• Lab — Reproducible notebook runs for fraud-model retraining, signed runs.
• Governance — Single policy engine for row, column, project and analyst scope.
• Identity — SSO via the bank's identity provider, no shadow accounts.
• Observability — Metrics, traces, decision logs emitted in OCSF format.
• Compliance — Evidence pack generation in minutes for DORA, GDPR and ACPR cycles.

Outcome

Conservative measured improvements.

Outcomes below are conservative estimates derived from our internal benchmark across three early production deployments with comparable workload shapes. They are framed as ranges we are confident in defending, not as marketing peaks.