Collekt Logo
Case StudiesBNPL lending

Collections Built as One Continuous Case, Not a Chain of Handoffs

How Collekt kept materially more of a leading BNPL lender's book current by carrying context from pre-due through day 30 — cutting the flow into collections against the lender's own baseline.

By Collekt Team, Customer SuccessLeading BNPL lender

Impact at a Glance

27%

less of the base flowed into collections under the model

Cohort 2 · vs the lender's own baseline

31%

less of the base flowed into collections under the model

Cohort 1 · vs the lender's own baseline

The industry works a delinquent account in disconnected stages: a pre-due reminder, a post-due push, then agencies, each run by a different team starting from scratch.

Collekt works one case, one context, from before the due date through day 30. On a leading BNPL lender's own portfolio, that continuity kept materially more of the book current, and cut the flow into collections against the lender's own baseline.

1 in 3
fewer accounts reach NPA

Less flow into collections carries almost straight through to the balance sheet. Fewer accounts on the DPD clock means fewer declared non-performing, and less provisioned against them.

One curve, not three buckets. The due date is a marker, not a wall.

Pre-due, early post-due and agency collection are usually three separate playbooks with a handoff at each gate, so the case is re-learned every time it changes hands.

Collekt treats it as a single resolution curve. The intervention that worked the account before the due date still knows it after. Prevention on the left of the line, context-carried recovery on the right, same model throughout.

10080600DUE DATEpreventioncontext-carried recovery-7+30teamCollekt
Collekt — one model, context carried across the whole curveInternal team — re-learns the case at every gate

A double-digit lead over the lender's own team

The cohorts are not directly comparable and the model kept improving underneath. What matters is that the lead survived a change of base.

Different customers, different composition, same result: through selective, real-time intervention across 100+ dynamic segments, the model kept materially more of the portfolio current before it ever entered collections.

Cohort 1

at the due date
Collekt model76%
Internal team65%
31%less of the base into collections · +11 pts current at the due date

Cohort 2

at the due date
Collekt model81%
Internal team74%
27%less of the base into collections · +7 pts current at the due date

The same reduction lands twice: once on capital, once on conduct

A case kept current never starts the journey. It cannot age into the buckets, cannot be declared non-performing, cannot sit in aggressive contact.

One number, two consequences.

For the CFO

Fewer accounts on the NPA path

~38%

fewer accounts declared non-performing on the modelled cohort — roughly 1 in 3

Resolution collapses once a case ages past the first bucket, so the overdue pool effectively sets the NPA feed. A smaller spill means proportionally fewer accounts declared non-performing, and lower provisioning against them, before a rupee of recovery cost is spent. Measured at agency entry; the resolution collapse after the first bucket is why the balance-sheet effect is at least as large as the reduction in inflow.

For the CEO

Fewer customers in aggressive contact

-27% / -31%

fewer accounts entering intensity and agency channels

Conduct risk lives in high-intensity contact and third-party agency activity. Keeping a quarter to a third more of the base out of that channel entirely is a lower regulatory surface and a healthier book, not just a cheaper one.

Post-due recovery is improving month on month, on fresh allocations

This is not the same cases worked harder. Each month is a new allocation.

The 10-to-30 day cure rate rose three months running because the model learns the post-due problem from the entire pre-due and early-due funnel, not from the thin residual most collectors ever see. More of the base means a better model, even on cases it has not met before.

52
54.6
57.55
Month 1Month 2Month 3
10-to-30 day cure rate

+5.55 points in three months.

A brute-force machine does not learn. A model trained on the full funnel does. The rising line is capability, not timing, and not the same accounts re-counted.

Three separate monthly allocations · same 10-to-30 measurement · Collekt actuals

What compounds next

Holding the case, unbroken, from day −7 through day 30.

Today the post-due bucket is still re-allocated as a separate stage. Next, each account stays with the same model and its full history across the entire window, so per-case context carries end to end.

A model already trained on the whole funnel, now also remembering every individual case, compounds the two effects.

This is a stated direction, not a result yet.

The thesis

Aggressive collection gets there in the end, at maximum cost and risk. Precise, continuous intervention gets there sooner, cheaper, and with fewer customers ever in the machine.


Collekt — Prevention Case Study. Same-book comparison · Model vs the lender's internal team.

Frequently asked questions

What was the core problem Collekt addressed?
The lender’s delinquency workflow was split across pre-due reminders, post-due pushes, and agency collection. Each stage behaved like a separate handoff. Collekt treated the account as one continuous case from before the due date through day 30, carrying context across the whole resolution curve.
What impact did Collekt have at the due date?
Across two different cohorts, Collekt kept materially more of the book current at the due date. Cohort 1 reached 76% current versus 65% for the internal team, while Cohort 2 reached 81% versus 74%. That translated to 31% and 27% less of the base flowing into collections.
Why does reduced flow into collections matter for the balance sheet?
A case kept current never starts the delinquency journey. It cannot age into higher buckets, cannot be declared non-performing, and does not require the same provisioning. The case study models roughly 1 in 3 fewer accounts reaching NPA from the lower collection inflow.
How did post-due performance change?
On fresh monthly allocations, the 10-to-30 day cure rate improved from 52.0% in month 1 to 54.6% in month 2 and 57.55% in month 3 — a +5.55 point gain over three months.
What is the next step in this model?
The stated direction is to hold each case unbroken from day −7 through day 30, so the same model and its full per-case history carry end to end rather than re-allocating post-due buckets as separate stages.
Collekt

Collekt Limited is a company registered in England and Wales (Company No. 15771475).
Registered Office: Coppersun Suite, Cardinal Point, Park Road, Rickmansworth, WD3 1RE.
VAT Registration No. 471479858.

Collekt Limited develops and licenses AI-driven technology infrastructure designed to support high standards of operational and regulatory performance within credit and recovery ecosystems.
Collekt Limited does not act as a lender and does not provide consumer credit, debt advice, or regulated debt collection services in the United Kingdom.
Operational debt recovery services in India may be delivered by Collekt Tech LLP (LLPIN: ACK-2677), a group entity registered in Mumbai, India, in accordance with applicable local laws.
hello@collekt.ai