The borrower who closes a loan application mid-way does not, in most cases, close it because they changed their mind. They close it because something specific got in the way — a field they could not answer, a document they did not have to hand, a form that would not load on their phone. Left alone, 70% of these borrowers will not return. Reached within 4 minutes with a specific message about the specific obstacle they encountered, 20–30% of them will. The Drop-Off Agent AI detects abandonment at the moment it happens and sends the intervention before the borrower has moved on.
Where borrowers drop — and what they are doing when they leave
Digital loan application drop-off across Indian NBFCs is not randomly distributed across the application form. It clusters at four points: the income declaration section (where self-employed borrowers face field labels designed for salaried applicants), the document upload section (where file format errors or size limits create failure states), the identity verification section (where Aadhaar-based OTP failures or V-KYC technical issues abort the session), and the consent and declaration page (where legal language creates final-step hesitation). Each clustering point has a distinct drop-off signature — a pattern of behaviour that precedes the exit — that allows the Drop-Off AI to identify what went wrong before the borrower has left the session.
A borrower who has been on the income declaration page for more than 4 minutes without advancing is not reading carefully — they are stuck. A borrower who has attempted an upload three times in 90 seconds is encountering an error, not reconsidering their file choice. A borrower whose Aadhaar OTP screen has loaded and stayed idle for 6 minutes has either not received the OTP or does not understand what is being asked. Each of these signatures triggers a different intervention — not a generic "complete your application" message but a specific response to the specific obstacle the signature indicates.
The digital lending application funnel: where the losses occur
A 28% end-to-end completion rate on a 2,400-application-month pipeline means 1,728 borrowers who started but did not finish. At an average disbursement of ₹35 lakhs, those represent ₹604 crores of potential disbursements that the institution is not capturing. Not because the borrowers do not need the loan. Because they got stuck.
The abandonment signal model: how the Drop-Off AI classifies each departure
The abandonment signal taxonomy: 8 signals and what each predicts
| Signal | Threshold | Abandon Probability | Predicted Problem | Intervention |
|---|---|---|---|---|
| Idle time — income section (SE borrower) | 4+ minutes, no field advance | 87% | Salaried field label confusion — SE borrower does not know what to enter | SE income explanation WhatsApp in 45s |
| Upload failure — same file, 3+ attempts | 3 failures in 90 seconds | 79% | File too large, wrong format, or corrupt upload | File fix instructions + WhatsApp upload alternative |
| Aadhaar OTP screen — idle 6+ minutes | 6 minutes, OTP not entered | 82% | OTP not received (wrong mobile registered with Aadhaar) or confusion | OTP troubleshooting guide + alternative KYC path offered |
| Consent page — scroll without action, 3+ minutes | 3 minutes, no checkbox action | 64% | Legal language anxiety — unsure what they are agreeing to | Plain-language consent summary sent in WhatsApp |
| Tab closed — mid-application, not consent page | Tab close event (JavaScript) | 72% | Interrupted by external event (call, distraction) — intent likely still present | "Save and continue" link + progress summary sent immediately |
| Return visit — same section as previous drop | Second visit, same drop point | 68% | Borrower returned but hit the same obstacle again — previous message not sufficient | Escalate to human assist — offer call-back from ops team |
| Mobile — pinch-zoom on form section | 3+ zoom events, same section | 58% | Form not mobile-optimised — borrower cannot read or fill the field on their phone | Desktop link sent + "can we call you to complete this?" offered |
| V-KYC — video call initiated but disconnected within 2 minutes | Call connected, dropped <2 min | 76% | Poor video quality, technical failure, or borrower not ready (wrong environment or documents) | V-KYC checklist + reschedule offer sent within 60 seconds of drop |
The re-engagement messages: what goes out within 4 minutes
Trigger: SE income label confusion (Signal 1)
Sent within 45 seconds of idle threshold · WhatsAppTrigger: Upload failure — 3 attempts (Signal 2)
Sent within 60 seconds of third failure · WhatsAppTrigger: Tab closed mid-application (Signal 5)
Sent within 30 seconds of tab close event · WhatsAppThe 4 minutes is not the threshold — it is the window
Research on digital application re-engagement consistently shows that the probability of a borrower returning decreases steeply with time. A message sent within 5 minutes of abandonment reaches a borrower who is still in the same mental context — still thinking about their loan, still on their phone, still reachable. A message sent 2 hours later reaches a borrower who has moved on to dinner, a work call, or sleep. The Drop-Off Agent AI's 4-minute detection and 45-second dispatch are not operational achievements — they are the conditions under which re-engagement works. The specific message matters. The timing is what gives it the chance to be read.