THE PROBLEM
A WORKING ECL MODEL IS ONLY THE START. PRODUCTION-SCALE IMPLEMENTATION IS THE HARD PART.
NBFCs have been required to compute ECL under Ind AS 109 since 2018-19. The framework is not new. What remains unsolved for most mid-market lenders is the operational layer: transition matrices built from five-plus years of granular DPD history, scenario-weighted PD models running across every active loan, stage classifications updated continuously as DPD moves, and audit trails that can withstand model validation and board-level review.
Institutions that have attempted this on spreadsheets, or by exporting data to a standalone tool, encounter the same set of failures: data preparation consumes most of the effort, scenarios cannot be rerun quickly when macro assumptions change, and the output carries no defensible audit trail. ECL compliance requires a system, not a calculation.
🗂️
DATA PREPARATION BOTTLENECK
Assembling five-plus years of DPD history, collateral records, and outstanding balances from separate systems is the hidden cost of running ECL outside the lending platform. By the time data is prepared, it is already stale.
📋
GOVERNANCE REQUIREMENTS
Ind AS 109 requires adequate documentation of ECL methodology, assumptions, and model results. RBI's model risk management guidelines add independent validation and board oversight requirements. Spreadsheet outputs cannot satisfy these at a governance review.
🔄
SCENARIO ITERATION IS MANUAL
When macroeconomic assumptions change, rerunning scenarios on a spreadsheet or external tool requires rebuilding the computation from fresh exports. Real-time sensitivity to macro shifts is not practical.
🔗
NO LIVE PORTFOLIO CONNECTION
An ECL model running outside the LOS and LMS cannot react to live changes — new disbursements, restructured loans, DPD movements — without a manual refresh cycle that introduces lag and error.
PLATFORM INTEGRATION
ECL RUNS ON YOUR LIVE LOAN DATA — NOT AN EXPORT.
An ECL model reading from a spreadsheet export answers questions about your portfolio as it was last week. OneFin's ECL module reads from the live data layer — the same loan records that drive origination, servicing, and collections.
WITHIN THE ONEFIN PLATFORM
DATA SOURCES
LOS AND LMS RECORDS
Origination history and loan type
Repayment records and DPD status
Outstanding principal balances
Collateral classification
Restructuring and write-off events
→
ECL ENGINE
COMPUTATION
Stage classification (SICR rules)
Transition matrix PD
LGD per loan type
EAD calculation
Scenario weighting (3 paths)
Probability-weighted ECL
→
OUTPUT
PROVISIONING AND AUDIT
ECL by loan, stage, and segment
Stage transition log
Scenario input assumptions
Model parameter record
Full computation audit trail
Staging updates as DPD moves. New disbursements enter the computation automatically. Restructured loans flow through the same data path without a separate import step.
HOW IT WORKS
FOUR STEPS FROM LOAN DATA TO AUDITABLE PROVISION.
The computation follows the IFRS 9 / Ind AS 109 structure: stage classification, component estimation, scenario overlay, and probability-weighted output.
STAGE 1
Performing
0–30 DPD
12-Month ECL
STAGE 2
Significant Credit Risk
31–89 DPD · SICR trigger
Lifetime ECL
STAGE 3
Default
90+ DPD
Lifetime ECL (credit-impaired)
1
CLASSIFY STAGE
Loans are assigned to Stage 1, Stage 2 or Stage 3 based on credit quality, DPD movement and configured significant-credit-risk triggers.
DPD RULES · SICR
2
COMPUTE PD
Probability of default is derived from bucket-based migration history. A transition matrix tracks the probability of a loan moving across DPD states from the LMS history.
HISTORICAL LOAN DATA
3
APPLY LGD AND EAD
LGD captures the portion lost after default. EAD is the lender's exposure at the point of default, shown in the source as unpaid principal plus remaining principal.
RBI BACKSTOPS
4
WEIGHT SCENARIOS
Three scenarios — Optimistic, Baseline, Pessimistic — adjust PD via macroeconomic variables. Final ECL is the probability-weighted average.
IFRS 9 · IND AS 109
WORKED EXAMPLE
SCENARIO-WEIGHTED ECL FOR A SECURED RETAIL PORTFOLIO.
Illustrative computation for a Stage 1 secured retail segment. Three economic scenarios produce different PD rates; the probability-weighted average determines the final provision.
GOVERNANCE AND COMPLIANCE
BUILT FOR THE GOVERNANCE REQUIREMENTS THAT APPLY TO NBFCS UNDER IND AS 109.
ECL obligations for NBFCs flow from Ind AS 109, IFRS 7 disclosure requirements, and RBI's model risk management guidelines — which have been in effect since the Ind AS adoption cycle. Each requirement below maps to a platform capability, or a confirmed open question.
IND AS 109 REQUIREMENT
HISTORICAL DATA
ECL estimates must use reasonable and supportable information about past events, current conditions, and forecasts. RBI's model risk guidance for NBFCs recommends at least 5 years of granular default and recovery data for model construction.
PLATFORM
ECL computation draws from the full DPD history in the LMS data layer. No data export or external warehouse required — historical records are available within the same system.
IND AS 109 REQUIREMENT
SICR AND STAGING TRIGGERS
Ind AS 109 (Para B5.5.22) presumes SICR when contractual payments are 30 or more days past due. Institutions may rebut this presumption with documented evidence of no significant increase in credit risk.
PLATFORM
Stage classification is rule-based and configurable per Ind AS 109 guidance. SICR thresholds are set within the platform. Stage transitions — including any rebuttal documentation — are logged automatically for every loan.
IND AS 109 REQUIREMENT
MACROECONOMIC SCENARIOS
Ind AS 109 (Para 5.5.17) requires ECL to reflect an unbiased and probability-weighted amount determined by evaluating a range of possible outcomes — incorporating forward-looking macroeconomic information.
PLATFORM
Three scenarios — Optimistic, Baseline, Pessimistic — with configurable probability weights. Scenario parameters, macro assumptions, and weights are recorded as part of every ECL run and available for review.
IFRS 7 / IND AS 107 REQUIREMENT
AUDIT TRACEABILITY AND DISCLOSURES
IFRS 7 (adopted in India as Ind AS 107) requires entities to disclose the basis of ECL calculations, key assumptions, and a reconciliation of opening and closing loss allowances — by stage and asset class.
PLATFORM
Every ECL run logs its inputs - DPD data, staging assignments, macro assumptions, model parameters with computed outputs.
RBI GOVERNANCE REQUIREMENT
BOARD OVERSIGHT
RBI's governance framework for NBFCs requires board-level oversight of significant risk management methodologies. RBI's draft Model Risk Management Guidelines (2026) explicitly extend this to AI/ML-based credit risk models.
PLATFORM
Staging rules, scenario parameters, and model configuration are auditable within the platform.
IND AS 109 REQUIREMENT
INDEPENDENT MODEL VALIDATION
RBI's draft Model Risk Management Guidelines (2026) require NBFCs using AI/ML models — including ECL models — to conduct independent model validation before deployment and periodic revalidation thereafter.
PLATFORM
Model parameters, transition matrix inputs, and computed outputs are accessible for review.
KEY CAPABILITIES
WHAT THE MODULE DOES — AND HOW.
🔢
TRANSITION MATRIX PD
Probability of Default is derived from your portfolio's own DPD migration history. A bucket-based transition matrix tracks the observed probability of loans moving from each DPD state toward 90+ DPD default — using the same data already in the LMS. The matrix reflects your actual book, not external benchmarks or regulatory floors.
⚖️
RICH SCENARIO-BASED ECL
Final ECL is the probability-weighted average of Optimistic, Baseline, and Pessimistic outcomes — baseline carrying the highest weight. Stage 1 loans use 12-month ECL. Stage 2 and Stage 3 loans use lifetime ECL. The approach is consistent with IFRS 9 / Ind AS 109 requirements for scenario-weighted measurement.
🛡️
RBI-ALIGNED LGD FLOORS
LGD is applied per loan type following RBI's regulatory backstops: 65% for secured exposures, 70% for unsecured, and 45% for collateral-backed loans. Rates are configurable per product and segment within these guardrails — no need to maintain a separate LGD table outside the platform.
📋
FULL AUDIT TRAIL
Every ECL run is logged end-to-end: input DPD data, staging assignments, macro assumptions, scenario weights, model parameters, and computed ECL values by loan and segment. Traceability is built into the computation — not appended afterward. Available for model governance review, independent validation, and RBI disclosure without a separate export process.