A DPD 45 borrower with a propensity score of 82 will pay if contacted today. The same DPD 45 borrower with a score of 31 has already decided not to — and sending a collections agent to call them today is a cost the institution pays to confirm what the model already knows. Propensity scoring separates these two borrowers before the call is made, so that every contact goes to the account most likely to produce a payment today.
Why propensity scoring transforms collections economics
Traditional collections prioritisation is DPD-ordered: the oldest accounts get the most attention, the newest get the least. This logic has an intuitive appeal — older accounts are at greater risk of permanent default — but it ignores the variable that actually determines whether a contact produces a payment: the borrower's current willingness and ability to pay. A DPD 75 borrower who has just received a salary payment and has been engaging with every contact is more collectible today than a DPD 35 borrower who has changed their number and gone silent. DPD ordering sends the agent to the wrong account.
Propensity scoring reorders the queue by collectibility — the probability that a contact with this account today will result in a payment or a meaningful commitment. It does this by combining historical behavioural signals, real-time financial signals, engagement pattern analysis, and contextual signals into a single score updated daily. The accounts at the top of the propensity-ordered queue are the ones where a contact today has the highest probability of producing a payment today — regardless of their DPD age.
"DPD tells you how old the problem is. Propensity tells you whether there is a solution available today. Collections agents should be calling the second list, not the first."
The 40 signals — organised by category and predictive weight
Category A — Payment Behaviour Signals
Model weight: 32% · Highest predictive category
01
EMI payment regularity — prior 12 monthsRatio of on-time payments to total EMIs due before delinquency onset. Strong predictor of willingness vs inability to pay.
High
02
Partial payment history in current delinquencyHas the borrower made any partial payments since going delinquent? Even ₹500 signals intent to pay.
High
03
Promise-to-pay fulfilment rateOf all PTPs made in this delinquency cycle, what percentage were kept? Broken PTPs reduce score; kept PTPs increase it significantly.
High
04
Days since last payment (any amount)The more recent the last payment, the higher the propensity. A borrower who paid ₹2,000 yesterday is more collectible than one who last paid 60 days ago.
High
05
Payment method at originationAuto-debit (NACH/ECS) borrowers who hit a technical NACH failure have higher willingness than those who migrated to manual payment after delinquency.
Medium
06
Historical delinquency curesHas this borrower had a prior delinquency episode on any loan and self-cured? Self-cure history is a strong positive propensity signal.
High
07
Days from first DPD to first partial paymentBorrowers who make a partial payment within 14 days of delinquency onset cure at 3× the rate of those who do not.
High
08
Payment channel used in current cycleUPI self-initiated vs. cash collected by agent vs. RTGS from a third party — each indicates different borrower engagement level.
Medium
Category B — Engagement and Communication Signals
Model weight: 24% · Real-time signal
09
Call pick-up rate — last 30 daysThe percentage of outbound calls answered in the last 30 days. Declining pick-up rate is the single fastest-moving propensity indicator.
High
10
Average call duration when answeredShort calls (under 30 seconds) indicate evasion; calls over 2 minutes indicate genuine engagement. Duration predicts subsequent payment probability.
High
11
WhatsApp message read rateHas the borrower read (blue-tick) recent WhatsApp messages without responding? Read-without-response scores differently from unread — the borrower is aware but disengaging.
Medium
12
Payment link click-through rateBorrowers who click payment links but do not complete payment have significantly higher next-contact propensity than those who do not click at all.
High
13
Inbound contact initiated by borrowerAny inbound call, message, or portal login initiated by the borrower is the strongest single engagement signal — they are seeking resolution.
High
14
Communication channel consistencyBorrowers who engage on the same channel repeatedly have stable communication patterns. Those who switch channels frequently are often evasion-adapting.
Medium
15
Time of day answer patternBorrowers with consistent answer windows (e.g., always answers between 12–2 PM) have higher propensity — predictability correlates with cooperation.
Medium
16
Verbal sentiment in last callNLP analysis of last call transcript — tone, cooperative language, specific date mentions, and hardship disclosure all change the propensity vector.
High
Category C — Financial Capacity Signals
Model weight: 22% · Ability-to-pay component
17
Current bank balance trend (AA data)Rising balance in last 30 days indicates recovering financial capacity. An account balance that just crossed ₹50,000 is a payment timing signal.
High
18
Recent salary credit detectionA salary credit in the last 5 days creates a high-propensity contact window of 3–7 days before discretionary spending absorbs the credit.
High
19
UPI inflow volume — current month vs prior 3 monthsRecovering inflow volume signals income recovery — a key indicator that the ability-to-pay component of the propensity score is improving.
High
20
NACH or ECS bounce — last 30 daysA recent NACH bounce may indicate the borrower attempted payment but had insufficient balance — a willingness signal even in a failed payment.
Medium
21
Existing obligations at other institutionsBureau credit pull shows how many other EMIs are active. A borrower servicing two other active EMIs has demonstrated current capacity — the delinquency is selective, not systemic.
Medium
22
Employer EPFO contribution statusConfirmed EPFO contributions in the last month confirm active employment — correlates with salary-driven payment capacity within 30 days.
Medium
23
GST turnover trend (SE borrowers)For self-employed borrowers, a recovering GST turnover trend is the most reliable leading indicator of near-term payment capacity.
Medium
24
Insurance premium payment regularityA borrower who has continued paying insurance premiums through a delinquency period is prioritising financial commitments — high willingness signal.
Medium
Category D — Contextual and Life Event Signals
Model weight: 14% · Timing intelligence
25
Day of month (salary calendar)Days 1–7 of the month are highest propensity for salaried borrowers. Scores recalculate on day 1 to reflect the salary window opening.
High
26
Festival season and bonus cycle proximityBorrowers in sectors with known Diwali or year-end bonuses have elevated propensity in November and February — contextual calendar signal.
Medium
27
Recent change of employer (EPFO)A recent employer change with a confirmed new employer signals income recovery in progress — elevated propensity for contact 30+ days post-change.
Medium
28
Property market conditions (secured loans)For LAP and home loan NPAs, rising local property values increase both settlement willingness (lower loss perception) and the borrower's ability to refinance.
Low
29
Days since last legal notice servedPropensity typically spikes in the 7 days after a Section 13(2) notice is served — legal awareness creates short-term urgency to engage.
High
30
Tax refund or GST credit seasonFor SE borrowers, March–April GST credit receipts are known liquidity events. Propensity scores for this segment are weighted upward in Q1.
Low
Category E — Behavioural and Psychological Signals
Model weight: 8% · Intent inference
31
Application portal login recencyA borrower who has logged into the loan app in the last 7 days is engaged — they are checking their balance or status, which correlates with payment intent.
Medium
32
Statement download or account queryBorrowers who have downloaded their statement or made an account balance query are likely preparing for a payment or resolution conversation.
Medium
33
Number of collection agents spoken toBorrowers who have been handled by 3+ different agents in a delinquency cycle have lower propensity — agent continuity matters.
Low
34
Complaint filed vs. not filedA borrower who has filed an RBI complaint about collection conduct is in a different engagement mode — propensity for payment drops; propensity for formal resolution rises.
Medium
35
Willingness to discuss vs. hang upThe transcript of the last call is analysed for the borrower's engagement depth — did they stay on the call to discuss, or terminate quickly?
Medium
36
Prior settlement attemptA borrower who previously engaged in settlement discussions — even if unsuccessful — has demonstrated resolution intent. This is positive for propensity even in a failed prior negotiation.
Medium
Category F — Demographic and Socioeconomic Signals
Model weight: Contextual — used to set baseline, not override behaviour signals
37
Loan tenure vs. remaining tenorBorrowers early in their loan tenor (first 20% of tenure) have higher propensity — the asset relationship is newer and the loss of collateral is more psychologically salient.
Low
38
Loan-to-value ratio at current valuationBorrowers with positive equity (LTV below 80%) have more to lose from default and auction — higher willingness to pay to protect the asset.
Medium
39
First-time borrower flagFirst-time borrowers, particularly in Tier 2 cities, have higher reputational sensitivity to delinquency — their willingness-to-pay is higher holding financial capacity constant.
Low
40
Bureau score trajectory since delinquencyA borrower whose bureau score has dropped sharply since delinquency onset — and who has queried their score — is aware of the bureau impact and motivated to repair.
Medium
A propensity score output — what the model produces per account
Propensity Score — Account HL-2024-7741 · Suresh Kumar Mehta
DPD 58 · Updated daily · Nov 14, 2025 07:00
78
Propensity Score
Scale 0–100
Scale 0–100
High
Collectibility
Contact today
Contact today
Voice
Recommended
Channel
Channel
Top signal drivers — what moved the score to 78
Salary credit detected
+18
Payment link clicked
+14
Partial payments made
+12
Call pick-up rate
+8
Multi-institution delinquency
−10
PTP fulfilment rate
−4
Score updated daily at 07:00 · 40 signals weighted · Recommended action: Voice call today between 11 AM and 1 PM (borrower's confirmed answer window)
40Signals across 6 categories — payment behaviour, engagement, capacity, context, behavioural, demographic
DailyScore recalculation cadence — salary credits, link clicks, and call outcomes update score same day
32%Highest model weight category — payment behaviour signals, particularly recent payment history
+18ptsSalary credit detection — single highest positive score driver, triggering a 3–7 day contact window
The score is not a prediction of default — it is a prediction of collectibility today
Propensity scoring is not a credit risk model. It does not predict whether a borrower will default eventually. It predicts whether contacting this borrower today will produce a payment or a commitment — and it does this daily, for every account in the DPD book, so that the collections team always knows which calls are worth making. The difference between a 78 and a 31 is not the borrower's character — it is the state of the contact opportunity today. Tomorrow's scores will be different, because the salary window will have changed, a PTP will have been kept or broken, a call will have been answered or avoided. The model follows the opportunity.