Case study · Reconciliation Engineering

Multi-source reconciliation engine 90–95% auto-match across MIS, banks, and PhonePe.

We engineered a productized matching + classification engine for an NBFC's loan reconciliation pain, five matching dimensions, weighted scoring, fuzzy name match, UTR-based dedup, and a clean exception queue.

NBFC Reco Tool v2.0Client

Challenge

The starting state.

There was no common identifier across MIS, accounting, multi-bank statements, and PhonePe, Loan ID existed in two of those, missing in two.

Customer names did not match: 'Mr. Ravi Kumar' in MIS became 'R KUMAR' in bank narration.

T+1 posting and batch uploads created date mismatches that humans patched manually.

Our approach

How we engineered it.

Weighted matching engine

Loan ID 40 / Name fuzzy 20 / Amount 20 / Date 10 / UTR 10, configurable per use case, with confidence scores per match.

5-class classifier

Collection, Disbursement, Top-Up, Renewal, Adjustment, rule-based + AI for fuzzy / edge cases.

Multi-source ingestion

MIS (Collection / Disbursement / Closure), Day Book, multiple bank statements, PhonePe, with standardization layer for name normalization, date format, amount sign, UTR extraction from narration.

Exception management

Auto-flagging with audit trail per match, only genuine exceptions reach humans, in a clean queue, not a spreadsheet.

What we built

Architecture and the systems it talks to.

Six-stage pipeline: Ingestion → Standardization → Matching engine → Classification → Exception management → Reports

Configurable matching weights and tolerance per use case (no code changes for tuning)

Audit trail per match: score + criteria used + source row references

Daily reco · monthly summary · exception report · transaction tag report, all Excel-exportable

Productized matching primitives reusable across NBFC loan reconciliation, cash collection, and multi-bank workflows

Outcomes

Quantified impact.

90–95%

Auto-match accuracy

Across MIS, accounting, multi-bank statements, and PhonePe.

−80%

Manual reconciliation effort

Field-team time redirected from spreadsheet stitching to dispute resolution.

100%

UTR-based dedup

Same UTR in PhonePe and bank statement no longer double-counted.

5 dims

Matching dimensions

Loan ID / Name fuzzy / Amount / Date / UTR, configurable weights.

5 classes

Transaction classifier

Collection · Disbursement · Top-Up · Renewal · Adjustment.

Audit-ready

Per-match evidence

Score + criteria + source row references retained for inspection.

Audit trail per match

Configurable tolerances

Source row lineage

In the words of the team

“The reconciliation engine has significantly improved our operational efficiency and accuracy. Post go-live, the solution reliably handled large transaction volumes and various matching scenarios with minimal exceptions. The reduction in manual effort, reconciliation errors, and turnaround time has generated substantial business value.”

· IT Head, Unnatti Finserv

Solutions and case studies you should read next.

Let's talk

Ready to start your project?

30-minute call with our team. Bring your project context and we'll map a clear path forward — no decks, no demos.

Get a free reconciliation audit