RISK INTELLIGENCE
SCORE EVERY LOAN
ACCOUNT FOR DEFAULT
RISK — BEFORE THE
PAYMENT IS MISSED.
OneFin's EWS model assigns a predictive risk score to each active loan account in the LMS. Higher scores indicate lower default risk. Collection teams, underwriting review, and portfolio monitoring all draw from the same signal — updated on live loan data.
THE PROBLEM
DPD MEASURES WHAT ALREADY HAPPENED. EWS MEASURES WHAT IS ABOUT TO.
DPD is a lagging indicator by design — an account enters a watchlist only after a payment is already missed. Among DPD = 0 accounts in a validated portfolio, EWS scores range from below 500 to above 750: a subset carrying materially elevated default risk that payment history alone cannot surface.
For most NBFCs, this invisible risk sits inside the current book. Borrowers with deteriorating bureau scores, falling income consistency, or rising multi-lender obligations continue to be treated as performing — because DPD has not moved. The signal exists; the metric to read it does not.
The same lag compounds across the credit function: collections is reactive, underwriting rules are back-tested on past defaults, and resource allocation follows DPD recency rather than repayment probability.
⏰
COLLECTIONS ACTS AFTER THE FACT
DPD-based queues are triggered by a missed payment. By the time an account enters a collection workflow, the EMI is already in default and the window for early engagement has closed.
🔍
CURRENT ACCOUNTS CAN BE HIGH-RISK
Accounts at DPD = 0 are treated as safe by any DPD-based system. A subset of that population carries elevated default risk that no standard metric surfaces until the first payment is missed.
🎯
COLLECTION EFFORT IS NOT PRIORITISED BY NEXT-PAYMENT RISK
Without a predictive score, teams distribute effort by DPD recency rather than predicted repayment risk. High-effort cases receive the same attention as straightforward resolutions.
📉
UNDERWRITING RULES LAG DEFAULT REALITY
New credit rules are typically back-tested against accounts that have already defaulted. Without a live default probability signal, rule improvement operates on historical outcome data, not current risk profiles.
PLATFORM INTEGRATION
EWS READS FROM THE LIVE LMS DATA LAYER AND WRITES SCORES BACK TO EVERY LOAN ACCOUNT.
The EWS model runs inside OneFin's platform. It draws from the same loan account records that drive collections, servicing, and reporting — not a periodic export. Scores are written back to the LMS per LAN so collection workflows, monitoring dashboards, and risk reports all draw from a single, current signal.
WITHIN THE ONEFIN PLATFORM
DATA SOURCES
LMS LOAN RECORDS
Cashflow history
Repayment behaviour
Allocation patterns
Credit Bureau enrichment
Account Aggregator signals
→
EWS ENGINE
RISK SCORING
Risk signal computation
Default probability modelling
Per-account score: 300–900
Risk band classification
Full active portfolio coverage
→
OUTPUT
LMS SCORE UPDATE
Score written per account
Risk band assigned
Available to collection queues
Available to risk dashboards
Scores feed into downstream workflows — collection queues, monitoring views, and risk reporting — directly from the same LMS data layer. No separate export or manual update step is required.
HOW IT WORKS
FOUR STEPS FROM LOAN DATA TO A PER-ACCOUNT DEFAULT PROBABILITY SCORE.
EWS scores every active loan account in four steps — drawing on live LMS data, enriched with bureau and Account Aggregator signals, and writing results directly back to the loan record.
1
PULL LOAN DATA
LMS loan records — cashflow history, repayment behaviour, and allocation patterns — are enriched with Credit Bureau and Account Aggregator signals for every active account.
LMS · BUREAU · AA
2
COMPUTE RISK SIGNALS
Loan data is converted into risk indicators — cashflow trends, repayment patterns, bureau variables, and income consistency — designed to capture deterioration before any payment is missed.
CASHFLOW · BUREAU · INCOME
3
SCORE EACH ACCOUNT
Every active account receives a score from 300 to 900 — the estimated probability of paying the next EMI. Higher scores mean lower default risk. DPD = 0 accounts are included.
SCORE RANGE: 300–900
4
UPDATE THE LMS
Scores are written back to the LMS per account and immediately available to collection queues, monitoring dashboards, and risk reports — no manual import required.
LMS · COLLECTION QUEUES
MODEL VALIDATION
THE SCORE TRACKS ACTUAL DEFAULT OUTCOMES — ACROSS EVERY 50-POINT BAND.
Validated on a sample portfolio. Accounts with DPD above zero score entirely below 450. The 600 mark is the consistent boundary between at-risk and performing. Every 50-point movement in the score corresponds to a measurable, predictable change in actual default probability.
ROC–AUC
>95%
When a healthy loan and a defaulting loan are randomly selected, the model ranks the healthy loan higher more than 95% of the time.
ACCURACY
94.5%
Overall correctness of predictions across both defaulting and paying accounts in the validated sample.
GINI COEFFICIENT
0.92
Measures how effectively the model separates high-risk from low-risk accounts across the full 300–900 score range.
KEY INSIGHT — RISK BEYOND DPD
Among accounts at current DPD = 0, the model surfaces significant score dispersion — from below 500 to above 750. Accounts scoring below 550 within this DPD = 0 group carry measurably elevated default probability that their payment history does not indicate. The median EWS score for DPD = 0 accounts in the validated sample is 735. Accounts clustering well below this median warrant proactive monitoring regardless of their current DPD status — an intervention window that DPD-only systems cannot provide.
DATA ENRICHMENT
BUREAU AND ACCOUNT AGGREGATOR SIGNALS, EMBEDDED IN THE SCORE.
EWS enriches LMS loan data with 2 external data sources per active account. Credit Bureau signals add cross-institutional credit history that the lender's own DPD data cannot see. Account Aggregator pulls add real-time financial visibility that borrower-reported income figures cannot provide.
CREDIT BUREAU SCRUB
CROSS-INSTITUTIONAL CREDIT SIGNALS
Credit Score
Comprehensive creditworthiness assessment aggregated across all lending relationships the borrower currently holds.
Credit Utilization
Debt-to-available-credit ratio indicating financial stress that may not yet be visible in the lender's own book or DPD history.
Bureau Delinquency History
Past payment behaviour across all institutions — capturing repayment patterns the current lender's DPD history may not reflect.
Account Age Mix
Maturity and diversity of credit accounts across lenders — a proxy for how long the borrower has managed credit obligations.
ACCOUNT AGGREGATOR (AA) PULLS
REAL-TIME FINANCIAL VISIBILITY
Bank Transaction Data
Direct visibility into current cash flows and spending patterns — not self-reported, and not dependent on what the borrower discloses at origination.
Income Consistency
Stability and predictability of income sources — deteriorating income consistency often precedes EMI stress by several months before DPD moves.
Savings Behaviour
Emergency fund adequacy and financial discipline indicators — eroding savings balances signal elevated repayment vulnerability before any payment is missed.
Multi-Institution Obligations
Complete picture of debt obligations across all lenders — directly relevant to FOIR and overall repayment capacity at the time of scoring.
KEY OPERATIONAL APPLICATIONS
THREE WAYS THE EWS SCORE CHANGES HOW THE PORTFOLIO IS MANAGED.
The EWS score is an operational input, not a report. It changes the basis for collection prioritisation, underwriting rule improvement, and portfolio segment evaluation — in each case replacing a backward-looking metric with a forward-looking one.
01
DYNAMIC COLLECTION MANAGEMENT
Build collection queues from EWS bands, not DPD buckets — engaging at-risk accounts before the first payment is missed.
Accounts below 450 flagged for immediate engagement regardless of DPD
450–600 band treated as proactive intervention window
Communication schedules and account assignment calibrated by score
02
UNDERWRITING RULE IMPROVEMENT
Back-test new credit rules against modelled default probabilities, not historical DPD outcomes — making rule improvement continuous.
Score distributions reveal which channels and products carry the most risk
Rule changes simulated against live portfolio score distribution
Improvement loop runs on current risk, not lagged default data
03
SOURCING AND PRODUCT STRATEGY
Monitor risk by channel, geography, and product type in real time — without waiting months for DPD trends to surface deterioration.
Median score drift signals channel-level quality decline early
Product-level distributions inform pricing and limit decisions
Channel mix decisions grounded in current default probability
SEE EWS SCORE YOUR ACTIVE PORTFOLIO.
Bring your loan book structure — product mix, portfolio size, and DPD distribution — and we will show how EWS scores distribute across your accounts, where the risk concentration sits, and how collection prioritisation changes when the queue is built from predictive scores rather than DPD alone.