Use case #0003

Churn root cause: what Customer Insights AI finds when repayment rates drop

When a lending institution's repayment rate drops from 94.2% to 91.8% in a single quarter, two responses are possible. The first is a collections escalation: increase call frequency, add agents, and push harder. The second is a root cause investigation: why did the rate drop, and is the cause something that collections can address or something that an operational, product, or macroeconomic change has created? The Customer Insights Agent AI runs the second response automatically when repayment rate deviation is detected — correlating the timing of the drop with operational changes, external events, borrower segment shifts, and feedback signals, and surfacing the cause rather than the symptom.

When a lending institution's repayment rate drops from 94.2% to 91.8% in a single quarter, two responses are possible. The first is a collections escalation: increase call frequency, add agents, and push harder. The second is a root cause investigation: why did the rate drop, and is the cause something that collections can address or something that an operational, product, or macroeconomic change has created? The Customer Insights Agent AI runs the second response automatically when repayment rate deviation is detected — correlating the timing of the drop with operational changes, external events, borrower segment shifts, and feedback signals, and surfacing the cause rather than the symptom.

Why collections escalation is the wrong first response to a repayment rate drop

A repayment rate drop has many possible causes. If the cause is economic — a sector-specific income shock, a localised employment disruption, a monsoon failure affecting agricultural income — then more collections calls will not improve the repayment rate. The borrowers who are not paying cannot pay. More calls produce more grievances, more FPC complaints, and no more collections. If the cause is operational — a processing error that caused NACH mandates to be submitted with wrong account numbers, an interest rate reset that changed EMI amounts without adequate notice, a system error that misapplied partial payments — then the collections team is calling borrowers who think they have already paid. If the cause is a product mix shift — the institution onboarded a new segment of borrowers in the prior quarter whose risk profile differs from the institution's historical portfolio — then the collections effort is addressing the symptom of an underwriting policy question. Each of these causes requires a different response. Only the root cause analysis can distinguish between them.

"Before you add collections agents, find out whether the problem is a collections problem. A repayment rate drop is an effect. The root cause investigation is the diagnostic that tells you what produced it."

The root cause investigation: Q3 2025 repayment rate drop 94.2% → 91.8%

Repayment Rate Drop — Root Cause Analysis · Q3 2025 · Karnataka NBFC
Rate: 94.2% (Q2) → 91.8% (Q3) · Drop: −2.4pp · Triggered investigation Nov 1, 2025
Q2 repayment rate94.2%
Q3 repayment rate91.8% (−2.4pp)
Accounts missed EMI (incremental)+1,162 accounts
First-time missers (no DPD history)841 (72.4% of incremental)
MSME proportion of missers68.4% (vs 42% of portfolio)
Textiles sector concentration44.8% of MSME missers
01
Sector-specific income shock — Karnataka textile MSME segment 841 of the 1,162 incremental misses are first-time misses — borrowers with zero DPD history who never missed before Q3. Of these, 68.4% are MSME borrowers. Of MSME first-time misses, 44.8% are in the textile sector in the Hubballi-Dharwad belt. Cross-referencing with external data: Karnataka textile exports fell 31% in Q3 2025 due to the Bangladesh supply chain disruption removing a key buyer. GST outward supply for textile MSME borrowers in this geography fell 28% quarter-on-quarter in Q3 — visible in AA data 6 weeks before the EMI misses appeared. → This is not a collections problem · It is a portfolio concentration risk that materialised · Collections escalation will not improve payment rates for borrowers with zero income from their primary trade · Recommended: proactive restructuring conversations for 378 affected MSME accounts · No enforcement action for first-time misses in this cohort
72.4% of
incremental
missers
02
NACH mandate error — 184 accounts with incorrect account numbers post-system migration 184 accounts that should have paid via NACH in Q3 did not — and the NACH returns show return code 11 (account number invalid or incorrect). Cross-reference with operations log: the CBS migration completed August 15 produced 184 account records where the NACH mandate account number was overwritten with an internal reference number rather than the borrower's actual bank account number. The borrowers believed they were paying via NACH — the debits were simply not reaching their accounts. → This is an operational error, not a repayment failure · 184 accounts restored to correct NACH mandate · Penal charges waived · Borrowers notified · IT team notified of migration audit requirement
15.8% of
incremental
missers
03
New borrower cohort — Q1 2025 MSME disbursements entering DPD 90+ window 137 accounts reaching DPD 90+ in Q3 are from the cohort disbursed in Q1 2025 (January–March). The Q1 MSME disbursements showed a materially higher early DPD rate than prior cohorts — 9.8% at 90 days versus the historical average of 3.4%. The Q1 cohort was notable for a DSA campaign that sourced 38% more MSME volume than any prior quarter, with several DSAs who have since been identified as having elevated early DPD rates across their sourced portfolios. → This is an underwriting and DSA quality issue from Q1 · Collections will work some of these accounts · The policy question is whether the DSA-channel eligibility criteria need tightening · 6 DSAs with >12% early DPD rate flagged for performance review
11.8% of
incremental
missers
● 3 causes identified · Cause 1: economic (72.4%) — restructuring, not collections · Cause 2: operational error (15.8%) — fix mandate, waive penal · Cause 3: underwriting policy (11.8%) — DSA quality review

The three responses — completely different from collections escalation

Cause 1
Textile sector
income shock
Economic — not a collections problem

Response: proactive restructuring for the 378 affected MSME textile accounts

The Loan Modification AI prepares a revised repayment schedule for each of the 378 accounts in the textiles-affected geography — a 3-month moratorium with capitalised interest, followed by a step-up EMI schedule that matches expected sector recovery. The contact is a support call, not a collections call. The first word the account manager says is "we understand what is happening in the textiles sector" — not "your payment is overdue." For first-time missers in an economic shock, the institution that responds with understanding and a practical solution retains the relationship. The one that responds with a collections escalation loses it permanently.

378 restructurings
No enforcement
Cause 2
NACH mandate
error
Operational — immediate fix required · Borrowers are blameless

Response: mandate correction + penal waiver + borrower apology

The 184 affected accounts are identified and their NACH mandates corrected within 24 hours of the root cause being identified. Every penal charge accrued during the period of the error is waived. Every affected borrower receives an apology message explaining what happened and confirming that their credit record has been corrected. The IT team is given the root cause and the CBS migration audit is expanded to check all 48,000 accounts for similar errors. None of these borrowers should have been in a DPD position — the institution created their DPD through its own error.

184 accounts
Penals waived
Cause 3
DSA quality
issue
Underwriting policy — collections works the accounts, policy changes the future

Response: targeted collections for 137 accounts + DSA quality review for the future

The 137 Q1 MSME accounts with DPD 90+ receive standard collections treatment — these are genuine credit events and the standard collections process applies. The parallel action is the policy response: the 6 DSAs whose Q1 MSME sourcing has produced early DPD rates above 12% are flagged to the DSA AI for performance review and commission hold. Origination criteria for MSME accounts sourced through these DSAs are tightened for Q4. The collections team addresses the symptom; the policy change prevents the next cohort from having the same problem.

137 accounts
+ policy fix
72.4%Of incremental missers explained by textile sector income shock — economic cause requiring restructuring, not collections escalation
184Accounts with NACH mandate error — the institution's operational mistake created their DPD · Penals waived · Accounts restored
6DSAs flagged — early DPD rate above 12% on Q1 MSME sourcing · Commission hold applied · Origination criteria tightened for Q4
3Distinct causes, 3 distinct responses — economic, operational, underwriting · A single escalation response would have been wrong for all three

A 2.4pp repayment rate drop that triggers collections escalation before root cause analysis will escalate on the wrong accounts and miss the operational error entirely

The standard institution response to a repayment rate drop is to increase collections intensity across the portfolio. In this case, that response would have put collections calls on 841 first-time misser textile MSME borrowers who cannot pay because their sector's primary buyer disappeared — generating FPC complaints and permanently damaging 378 relationships that could have been restructured and retained. It would have put collections calls on 184 borrowers whose NACH mandates the institution itself corrupted — generating Ombudsman complaints and regulatory scrutiny. And it would have applied generic collections treatment to 137 genuinely high-risk accounts without identifying the DSA sourcing pattern that produced them. The Customer Insights Agent AI runs the root cause investigation before the collections escalation is approved — so the institution knows what it is actually dealing with before it decides what to do about it.

← Back to Customer Insights Agent AI