Banking · AIEast Africa·Reconciliation · Settlement2022

AI Reconciliation Engine — Interbank Settlement Operations


Situation & Challenge

A commercial bank processing high daily transaction volumes across four core banking systems was experiencing a persistent mismatch rate requiring large analyst teams to resolve manually each day. The legacy rules-based tool could not learn from past resolutions and generated high volumes of false positives that consumed analyst time without producing fixes.

What Coadjutant Built

A four-layer pipeline: format normalisation across heterogeneous CBS schemas → ML classification of mismatch typology across 26 defined categories → root-cause isolation → automated correction entry generation with human-override at each threshold. Trained on seven years of historical resolution data and retrained weekly on live patterns. Full audit trail on every automated action.

Technology & Protocols

Python reconcile engineXGBoost classifierEvent-driven architectureSWIFT MT940 parsingHuman-in-the-loop UIAudit trail generation