Most NBFCs run an origination system, a loan management system, an EIR spreadsheet, an ECL model and a GL interface — five systems telling five versions of the same loan. FinIFRS Lending is one book: the application you underwrite is the loan you service, the schedule you service is the cash flow the EIR is solved on, the EIR is the rate the ECL is discounted at, and every rupee reaches the ledger through a balanced journal. Consumer durable, gold and vehicle loans ship configured; the next product is a template and a maker-checker approval away.
An auditor asks it every year, and in most NBFCs the answer lives in three places that were never built to agree: the LMS knows what the customer owes, a spreadsheet knows the EIR, and a consultant's model knows the provision. Each is internally right. Together they cannot be reconciled — which is exactly the property a regulator and an auditor test for.
Knows the dues and the DPD. Has never heard of amortised cost, and books the dealer's subvention as day-one income.
Recomputed at quarter end from an export, on cash flows that no longer match what the LMS actually collected.
Discounted at the contract rate, staged on a different DPD cut, and reconciled to nothing the ledger holds.
These are not features bolted on. They fall out of having one engine and one book — which is why a vendor with an LMS and a separate accounting add-on cannot match them without rebuilding.
The impairment is discounted at the rate the interest income was recognised at, because one engine produced both. Your auditor's reconciliation is a query, not a project.
A loan is booked against the exact effective-dated version of every policy. Changing the product tomorrow never re-prices a loan booked yesterday — and the explainer can prove which version applied.
Underwriting returns PASS, REFER or FAIL. A REFER names the waivable rule, the authority allowed to waive it, and the arithmetic — ‘FOIR 58.3% against a 50% ceiling’ — not a silent reject.
Ind AS 109 ECL and RBI IRACP computed side by side on the same classification, with the 13-Mar-2020 impairment-reserve rule applied where the prudential floor is higher.
Rail vs book reconciliation flags SETTLED_NOT_POSTED — a paid EMI that never reached the loan — before it becomes a collection call, a penal charge and a bureau report.
Insufficient funds re-presents after salary day; a technical decline re-presents tonight; a mandate cancellation stops presenting. Codes drive action, not an error log.
The lending pipeline runs left to right across the product. Every stage below is live in the product today — none is a slide.
Each step consumes the stored output of the one before it — never a re-keyed export.
Pick a starter product, adjust the six policy families, submit. A checker approves; the product goes live effective-dated.
KYC, bureau and income verified; the credit policy returns PASS, REFER or FAIL with its reasons and the eligible amount.
The KFS is generated with APR, fee split and contingent charges; the validity window is enforced before execution.
Schedule and EIR solved at booking from the lender's own cash flows, subvention and fees included.
NACH presentation, receipts, suspense, part-payments and foreclosures, appropriated by policy.
EOD accrues on the EIR, ages DPD, classifies NPA and posts balanced journals shard by shard.
Provisions, bureau file and returns built from the one measured book, each with its own cross-check.
The engine computes every figure first and stores it. A model is allowed to put stored facts into sentences, rank what the engine already ranked, or propose a mapping a human approves — nothing else.
“Why is this loan carried at ₹21,701?” — answered from the stored EIR schedule, accrual register and bound policy versions, each figure cited.
Computed observations — four days from NPA, eligible for upgrade, mandate failing — ranked into next actions. Penal charges are never called interest.
Ask the book a question. The model picks a reviewed read-only query and returns the SQL that ran beside the answer.
Shock the benchmark, revalue every pledge, re-rank LTVs against each cap; the model drafts the outreach order.
Rail vs book breaks classified and explained, with SETTLED_NOT_POSTED escalated first.
Proposes how `PRIN_OS`, `INT_SUSP` and `NPA_DT` map to the loader — and refuses a mapping that would book unearned income. Never applies one itself.
Borrower-facing reminders drafted from the dues engine, in the borrower's language.
Across every FinIFRS product the rule is the same: a deterministic, tested engine computes and stores every figure first. AI reads documents, drafts sentences and answers questions over what is already stored — it never computes, posts or approves an amount.
Every rule a credit or finance team argues about is a validated, versioned policy — parsed against the engine's own schema on write and again on read, so a stored policy can only describe something the engine can compute.
Categories, not vendor names — the comparison is about architecture, and architecture is what you cannot configure your way out of.
| Capability | Legacy core LMS + add-ons | Spreadsheets + consultant | FinIFRS |
|---|---|---|---|
| EIR solved on the lender's own cash flows at booking | Add-on, from exports | Quarter-end workbook | ✓ Native, stored per loan |
| Subvention and fees earned through the EIR | Often day-one income | Manual adjustment | ✓ B5.4.1 by construction |
| ECL discounted at the loan's own EIR | Separate model | Contract rate | ✓ Same engine, provable |
| Ind AS 109 and RBI IRACP side by side | IRACP only | Two workbooks | ✓ One classification, both provisions |
| RBI KFS with APR computation sheet | Template document | — | ✓ Generated and validated |
| CIC file validated before submission | Rejection report next week | — | ✓ Reasons found in advance |
| New product live | Vendor change request | — | ✓ Template + maker-checker |
| AI that cannot invent a figure | — | — | ✓ Engine computes, model narrates |
No. Take the EIR & ECL engine on its own and feed it your LMS's loan tape — the same measurement, provisions and notes, without migrating servicing. Many lenders start there and move servicing later.
Consumer durable (including zero-cost EMI with dealer subvention), gold loans with vault custody and LTV monitoring, and vehicle / e-bike loans ship as templates. Other term-loan products are configured from the same six policy families.
EOD runs per tenant, per business date, per shard, with a constant number of database round-trips per shard rather than one per loan. Every stage is idempotent, so a crash mid-run is re-run, not repaired.
No. Every AI feature reads figures the engine has already computed and stored. The migration agent proposes; a person approves. The portfolio assistant can only run reviewed read-only queries, and shows the SQL it ran.
All money is held as integer paise end to end. There is no floating-point rounding to drift between the schedule, the ledger and the provision.
A four-week pilot on a closed month of your loan tape: we book it, solve the EIR, run ECL beside IRACP and hand back the reconciliation your current stack can't produce.