All products
Audit & Assurance
CA Audit
AboutContactSupportSign InBook a Demo →
EIR & ECL Engine · Ind AS 109 · for NBFCs and HFCs

Your ECL and your interest income, discounted at the same rate. Provably.

Every NBFC books two Ind AS 109 numbers that are supposed to share one rate: interest income on the effective interest rate, and the impairment allowance discounted at that same rate. In practice the EIR comes from a finance spreadsheet and the ECL from a risk model, and they have never agreed. FinIFRS computes both in one engine, from one loan tape, under a methodology that is a versioned, board-approvable document — and stamps every run so a three-year-old provision reproduces to the rupee.

✓ One engine: EIR → staging → ECL → journal✓ 336 golden test cases✓ 17 disclosure notes + IRACP annex✓ Reproducibility triple on every run✓ Methodology as data, maker-checker✓ Works beside any LMS
5
PD methods, with Vasicek / scalar through-the-cycle → point-in-time conversion
17
Ind AS 107 / 109 disclosure notes produced by every run, plus the IRACP annex
3
Stamps on every run — data fingerprint, methodology version, engine version
94/100
Data-quality scorecard on the demo HFC book, computed before a figure is trusted
The problem

Five things quietly wrong in most ECL implementations.

Each is a material misstatement a spreadsheet produces without anyone noticing — and each is what an RBI inspection or a Key Audit Matter eventually finds.

PD × LGD × EAD on averages

ECL is a sum over time of monthly products. On the worked example, month one is understated ~20% (₹46 vs ₹57.49) when the downside moves both PD and LGD.

EAD held flat, life run to maturity

A 240-month mortgage on a 14% CPR book has an ~84-month behavioural life. Flat exposure to contractual maturity charges for defaults on loans that will have left.

Discounted at the contract rate

¶5.5.17(b) says the loan's own EIR. Using the contract rate makes the allowance unreconcilable to interest income — the auditor's first question.

What others can't do

What a stand-alone ECL model cannot do.

The differentiators are structural: they come from computing EIR and ECL in the same engine, from the same cash-flow vector.

ONLY HERE · 01

The rate reconciles by construction

The ECL discounts at the EIR the income was recognised at — because the same code produced both. There is nothing to reconcile.

ONLY HERE · 02

Hard EIR cases, solved

Six-to-ten construction tranches, pre-EMI moratoria, floating resets that flex the tenor, behavioural life, and a fee taxonomy your auditor signs.

ONLY HERE · 03

LGD priced in time

Collateral LGD with SARFAESI realisation lag and a distress haircut. A ₹50 L loan on an ₹80 L flat is not 0% LGD once 3.5 years of workout are priced.

ONLY HERE · 04

A methodology you own

Segmentation, SICR rules, PD / LGD methods, scenarios and overlays are a versioned pack your model committee edits and approves — not a consultant's macro.

ONLY HERE · 05

“Show me March 2027”

Every run is immutable and stamped with data fingerprint, methodology and engine version. The number reproduces exactly, years later.

ONLY HERE · 06

Model vs floor, on one slide

Management overlays and the RBI prudential floor applied after the model, each shown separately, so governance sees what the model said and what policy added.

What is inside

The pipeline, one reporting date at a time.

validate → index → EIR → lifetime PD → stage → ECL → overlays → prudential floor → Stage 3 net interest → aggregate → disclose → post.

INGData contract v1Eight ingestion files; one contract drives the template, the parser and the validator, with row-level exceptions.
DQData-quality scorecardCross-file rules scored before any figure is trusted; minimum-viable-ingest tool for pilot scoping.
EIREIR solverNewton-Raphson with bracketing bisection; tranches, moratoria, resets, prepayment and fee taxonomy; rate at inception, rolled on actual cash.
STGStaging (SICR)Rules engine with probation and cure, individual and collective assessment.
PDProbability of defaultFive methods including Markov transition; lifetime marginal PD term structure; TTC → PIT.
LGDLoss given defaultCollateral, workout, cure-adjusted and benchmark approaches.
EADExposure at defaultAmortising exposure path with CCF on undrawn commitments.
MACMacro scenariosScenario resolution and probability weighting; sensitivity quantified, not described.
OVROverlays & floorManagement overlays with rationale and approver; RBI prudential floor and IRACP comparison annex.
DISDisclosure notesSeventeen Ind AS 107 / 109 notes, including Stage 2 composition — the note nobody else publishes.
JVProvision journalBalanced voucher with GL → segment → loan drill-down.
AUDAudit & governanceHash-chained module audit trail, maker-checker, locked runs, tenant isolation.
AI-native · 3 live

AI that explains the provision. Never one that sets it.

Every figure is engine output first. AI reads the stored run and its evidence.

Run narrative

Live

The run report states on its own initiative what moved and why — stage migration, overlays, floor impact — from the stored run.

Loan-level explainer

Live

“Why Stage 2?” and “why this carrying amount?” answered from the stored schedule and staging evidence, each figure cited.

Column mapping for the loan tape

Live

Proposes how a legacy tape maps to the data contract, refusing mappings that would mis-state income. A person approves.

Early-warning signals

Roadmap

Account-level signals from bureau refresh, bounce history and collateral value, feeding SICR as evidence rather than as a model score.

The boundary

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.

Highly configurable

Your methodology, as data.

A change in methodology is a configuration change with a maker-checker approval — not a six-week vendor release.

Segmentation tree by product, geography, ticket size, collateral or any tape field.
SICR rules — DPD backstop, rating downgrade, watch-list, restructure flags — with probation periods.
PD method per segment, LGD approach per segment, EAD and CCF assumptions.
Macro scenarios, weights and the variables they shift.
Management overlays with mandatory rationale, amount basis and approver.
Fee taxonomy: which fees enter the EIR and which are recognised otherwise — the mapping your auditor signs.
Run windows, reporting dates and archetype packs for HFC, vehicle, MSME and personal-loan books.
Neck and neck

Side by side.

Against the two ways Indian NBFCs and HFCs produce Ind AS 109 numbers today.

CapabilityConsultant ECL modelEIR spreadsheet + LMSFinIFRS
ECL discounted at the loan's own EIRContract raten/a✓ Same engine
Lifetime ECL as Σ MPD × LGD × EAD × DFAveragesn/a✓ Monthly, scenario-weighted
Behavioural life and prepaymentContractual—✓ Per product
Construction tranches and moratoria in EIRn/aManual✓ Native
Reproduce a past run exactlyRe-engage consultant—✓ Stamped, immutable
Methodology changeEngagement—✓ Config + maker-checker
Seventeen notes and the journalInputs to build them—✓ The output
The questions the room actually asks

Do we need to replace our LMS?

No. The engine takes your loan tape through the data contract. If you later want servicing too, the NBFC Lending Suite runs on the same engine.

Who owns the methodology?

You do. It is a versioned pack your model committee edits and approves, with the history of every change. We ship sensible defaults; none is imposed.

How does it handle the RBI prudential floor?

Overlays and the floor are applied after the model and shown separately, with the IRACP comparison annex RBI requires — so the board sees the model, the judgement and the regulatory top-up as three numbers.

What does a pilot need?

The minimum-viable-ingest tool tells you exactly which fields are required for your book. Typically a loan master, a cash-flow or schedule file, a DPD history and collateral values.

The low-risk way in

One quarter-end of your book. Both numbers, one rate.

We run your last quarter through the engine and hand back EIR, ECL, the notes and the journal — with the reconciliation between interest income and impairment your auditor has been asking for.

Works with
Beyond the product lines

We build the AI-native finance automation you can't buy.

A regulator's bespoke return, a partner settlement, a reconciliation between two systems only you run. We build it to your requirement on the same engines, audit trail and AI boundary our products run on — in weeks, with golden tests, and handed over as configuration your team owns.

01Deterministic where money movesEvery figure is computed by a tested engine — never by a model.
02AI where documents and questions areExtraction, narrative and Q&A over stored facts, with validation.
03Governance from day oneHash-chained audit, maker-checker and tenant isolation, inherited.
04Yours to changeRules as versioned configuration your controllers edit after go-live.