Industry Deep Dives

Building Fintech That Reconciles to the Paisa

Ledgers, idempotency and daily reconciliation. The unglamorous engineering that separates fintech products that survive audit from those that do not.

Purushottam Kumar Suman
Purushottam Kumar Suman
Founder & CEO, Drema AI
9 min read
Stock market trading screen displaying financial charts

In most software, a bug degrades an experience. In fintech, a bug moves money that should not have moved, and the consequences are a customer complaint, a regulatory event and a reconciliation exercise that consumes a week. The engineering has to be built around that asymmetry.

01

The ledger is the product

Storing a balance as a number you increment is the mistake everything else follows from. A double-entry ledger of immutable entries, with the balance derived, means every figure has a history and any disagreement is resolvable by inspection. Correcting an error becomes a compensating entry rather than an untraceable update.

Never store a balance you can edit. Store the entries and derive it.

02

Idempotency everywhere money moves

Networks time out, users double-tap and queues redeliver. Every money-moving endpoint takes an idempotency key, stores the response against it and returns that response on retry. This one pattern prevents the most expensive class of defect in the sector, and it is cheap to implement at the start and painful to retrofit.

03

Reconcile daily, automatically

Your ledger and the provider's settlement file will diverge — timing differences, failed callbacks, partial refunds. The question is whether you discover that daily by automation or monthly by a customer complaint. Automated reconciliation with an exceptions queue is core infrastructure, not back-office tooling.

Immutable entriesBalances derived, never edited
Idempotency keysOn every money-moving endpoint
Daily reconciliationWith a human exceptions queue
Audit trailWho, what, when — unmodifiable
04

Webhooks are unreliable by nature

Payment callbacks arrive late, twice, out of order, or not at all. Treat them as hints rather than truth: verify signatures, handle duplicates through idempotency, and run a polling reconciliation for anything not confirmed within a window. Systems that trust webhooks as the sole source of truth fail in ways that are hard to detect.

05

Keep card data out of scope

Provider tokenisation means raw card details never touch your systems, which reduces PCI scope from an organisational programme to a much narrower question. This is an architectural decision made once at the beginning, and reversing it later is extremely expensive.

06

Design reporting from the ledger

Regulatory reports assembled by hand before a deadline are error-prone and unrepeatable. Generating them from the ledger means the report is reproducible, traceable to entries and consistent between periods — and it turns a recurring week of work into a query.

Daily
Automated reconciliation
Every
Money endpoint takes an idempotency key
Derived
Balances, never stored and edited
Purushottam Kumar Suman
Written by
Purushottam Kumar Suman
Founder & CEO, Drema AI

Founder and CEO of Drema AI. Builds AI systems, SaaS platforms and industry software — and writes about what actually survives production.

CTA Background

Got a problem like this one?

Bring it to a call with a founder.You will get a straight answer, including when the answer is no.

View Our Work
AI-First Engineering
Secure & Scalable
Built to Deliver Impact