Early payment default review that re-reads every file and says why each loan failed — fraud, a credit miss or the bank's own slip
Every early default explained within a week, the fix assigned.
See one case, screen by screen ↓A review of every loan that misses one of its first payments, to find out why it failed — fraud, a credit miss, an onboarding slip or a real hardship — and send the fix to the team that owns it. Early Payment Default Review opens a case the day a loan goes 30 days late on any of its first three payments and has agents re-read the whole file.
How to choose AI solutions for lending →
Updated
A loan that misses one of its first payments is the loudest signal a lender gets. Either the borrower was never going to pay, the credit policy let in a borrower it should not have, or the bank broke something between approval and the first payment — autopay never set up, the wrong due date. In a vintage chart the three look the same.
So credit risk counts early defaults by vintage and re-reads a sample of files by hand, weeks later. The rest stay a number. A ring coming in through one partner sub-source looks like a credit problem, the score cut-off gets tightened, good volume is lost and the ring stays open — while the borrowers the bank itself made late carry a 30-day mark nobody corrects.
Estimated split for one monthly vintage, a sampled review by hand against the solution; the sampled-review baseline of about six weeks is an estimate.
Six specialist agents re-read every early default's file and propose its cause; the credit risk analyst confirms.
Lanternfield Bank booked 4,812 personal loans in August. 87 of them are now 30 days late on their first payment — 1.81% against 1.06% for the six vintages before. Here is how the senior credit risk analyst finds out why, screen by screen, in the working solution.
Priya opens the August vintage: 87 early defaults, $1,268,450, causes due Fri 6 Nov. Only what the systems already know can branch the tree — channel and partner — and every leaf reads “Not explained yet”. Fernway Lending Network is 13% of the loans and 38% of the early defaults, 5.14% against its 3.0% contract trigger.
“The channel says where it happened; only the files say why.”
The early default monitor has already assembled each file: application, documents, verification results, the decision record, fraud checks, channel and every payment attempt. Priya presses one button and the agents read them, step by step, with the progress in the tree’s header and the run log in the side rail.
“3 of 6 · one ring of 19 loans on 4 devices”
The tree regrows by underwriting exception. 36 of the 87 went through the instant-decision path IV-W, which accepts stated income with a pay-stub image and no payroll or bank-data check. Waived loans defaulted early at 3.25%; loans with income checked, at 1.38%. The waiver runs through the ring, the first-party fraud and the CP-26-07 credit misses alike.
“Income verification waived — 41% of early defaults this quarter”
Switched to the cause found, the tree splits: fraud 34 loans ($551,300), credit 29 ($389,250), onboarding 16 ($214,700) and hardship 8 ($113,200). Each leaf opens its loans with the reason the agents found in each file.
“Fraud is the thickest branch — 34 of 87, and 19 of them are one ring.”
The early defaults list shows the run log of how the 87 were explained and the patterns found — ring R-0926, rule CP-26-07, autopay never saved on Android app 6.14 — each with its owner. 79 causes are at 90% confidence or more and can be confirmed together; the 8 below are read one by one. The review clock shows 4 days left.
PL-2608-41377 came through Fernway sub-source FW-2207 and was approved automatically: score 702, debt-to-income 31.0%, fraud score 412 with review from 600. The timeline flags what the decision missed — device DV-7F3A on 6 other applications, a pay stub made from the same template as 18 others, an employer phone that is an internet line registered 9 Jul, a bank account opened 19 days before applying. Proposed cause: fraud, ring R-0926, 96% confidence.
“A borrower in trouble usually pays once.”
The link map ties the 19 loans to 4 devices, 3 receiving bank accounts opened in July, 2 employers with one pay-stub template and 11 employer phone numbers that are internet lines. All 19 came from FW-2207. Four September loans carry the same signals and two applications are still open, ready to hold for review. The suspicious activity report clock runs 30 days from detection on 2 Nov.
The link analyst never blocks an account or files a report — it prepares the referral for the BSA team.
CP-26-07 raised the debt-to-income cap for scores 640–679 from 40% to 45% on 14 Jul. August booked 412 loans in that band above the old cap; 21 defaulted early — 5.10%, against 1.62% below 40%. In the what-if, a 42% cap with a $1,800 monthly residual-income floor declines 268 of the 412 and avoids 17 of the 21 early defaults; the 144 loans kept would default at 2.8%.
The credit rule tracer never changes a rule. Every proposal goes through fair-lending review first.
Four fixes, each with its evidence and what it costs: close the ring and the door it came through (Lena Hoffmann), amend CP-26-07 and end the income waiver (Marcus Bell), fix the borrowers the bank made late (Renee Alvarez), and a Fernway partner review (Jordan Ellis). Nothing leaves until Priya sends it; Lena is the first to accept.
The cause and fix writer drafts the October early default report for the credit risk committee on Tue 10 Nov: what happened, the rate without the ring and the app defect (1.08%), the thread through all three, fixes and owners. Priya confirms all 87 causes and sends it; Daniel signs.
“Explained in 2 days · July report took 4 days · a sampled review used to take about 6 weeks.”
The same vintage, seen by the five people who explain it, fix it and sign it — what their month looked like, and what it looks like now.
Six specialist agents re-read every early default's file and propose its cause; the credit risk analyst confirms.
Watches servicing and opens a case the day a loan goes 30 days late on any of its first three payments, with the whole file assembled — application, documents, verification, decision record, fraud checks, channel and every payment attempt.
Re-reads the application, documents and verification results as an underwriter would, after the fact, and lists every discrepancy — income, employer, identity, account, address — and the underwriting exception each loan went through.
Links applications and loans of the last 180 days by device, bank account, phone, address, employer and document template, finds rings, and lists applications still in the pipeline that carry the same signals.
Replays each decision under the policy version in force at funding and the one before, finds the rule whose change made the difference, and replays the whole vintage for any proposed setting.
Checks that what the borrower chose was set up: the autopay authorisation, the first due date against the signed contract, the welcome letter, statements and every payment attempt — and reports where the bank, not the borrower, broke a step.
Proposes one cause per loan — fraud, credit, onboarding or hardship — with its confidence, its runner-up and the evidence, groups loans into patterns, drafts each fix for its owner and the monthly report.
Confirms every cause; the director signs the report. The agents propose; a named person decides.
Everyone on the team can ask in plain words — or tell it what to send. Every answer cites the file, record or rule it rests on.
Why did the August vintage jump?
August’s early default rate is 1.81% — 87 of 4,812 loans — against 1.06% for February to July. It is three problems, not one: fraud, 34 loans, 19 of them one ring through Fernway FW-2207; credit, 29, of which 21 were approved only under CP-26-07; onboarding, 16, of which 13 had autopay never saved on Android 6.14; and hardship, 8. Without the ring and the app defect the rate is 1.08%.
Show me the Fernway ring
Ring R-0926 is 19 loans, $338,400, funded 4–19 Aug through Fernway sub-source FW-2207. They share 4 devices, 3 receiving bank accounts opened in July, and pay stubs for two employers made from one template. Not one payment has been made on any of them. Two applications with the same signals are in the pipeline now — $41,000; I would hold them today.
What would a 42% debt-to-income cap cost us?
For scores 640–679, a 42% cap with a $1,800 residual-income floor, replayed on August: 268 of the 412 loans booked above the old cap are declined — $3,210,640 of volume — and 17 of the 21 early defaults are avoided. The 144 loans kept would default at 2.8%, against 1.62% below 40%. Going back to 40% avoids all 21 but gives up all 412.
Which underwriting exception shows up most?
Income verification waived — 36 of the 87 August early defaults (41%) were approved on the instant-decision path IV-W, which accepts stated income with a pay-stub image and no payroll or bank-data check: 19 ring loans, 9 first-party frauds and 8 CP-26-07 credit misses. Waived loans defaulted early at 3.25% (36 of 1,108); verified loans at 1.38%.
13 screens from the working solution, on its sample data. Pick one to see it large.
How an early default is defined, explained and routed is a setting — versioned and on the audit trail.
| Setting | Default | Choose from |
|---|---|---|
| Early default | 30+ days past due on any of the first 3 payments | Any of the first 3 payments · 30+ days past due on the first payment · 60+ days past due in the first 6 months |
| Review clock | 5 business days | 3 · 5 · 10 business days |
| Bulk confirm | 90% or more | 90% or more · 95% or more · Never — one at a time |
| Risk appetite | 1.25% | Early default rate at the first-payment read, by vintage — raised or lowered in Settings |
| Fair-lending review before a credit rule proposal | On | On · off |
| Refer every confirmed ring to the BSA team | On | On · off |
| Lender-error late marks corrected with the bureaus | On | On · off |
| Partner early default trigger | 3.0% | 2.5% · 3.0% · 4.0% |
Every figure is labelled: a target the solution is built to, an estimate, a typical published result, or a proven one.
“demo” = seen in the working solution, on its sample vintage · “target” = the design goal, measured in the live solution · “estimated” = our estimate, including the sampled-review baseline of about six weeks · “typical” = published figures (Point Predictive study via Auto Finance News — 70% of early payment defaults tied to fraudulent applications). People, the bank, partners, merchants and loans named on this page are fictional — characters and sample data in the working solution.
A review of every loan that misses one of its first payments, to find out why it failed — fraud, a credit miss, an onboarding slip or a real hardship — and send the fix to the team that owns it. Early Payment Default Review opens a case the day a loan goes 30 days late on any of its first three payments and has agents re-read the whole file.
A vintage chart says the rate went up; it does not say why. Here every file is re-read and the root-cause tree branches the vintage by cause, underwriting exception, fraud indicator, channel or partner, so a ring, a loose rule or a waived check shows up as a pattern.
The link analyst searches every loan and application of the last 180 days for shared devices, receiving accounts, phone numbers, addresses, employers and document templates, groups linked loans and lists applications still in the pipeline with the same signals. It never blocks an account or files a report; confirmed rings go to your BSA team.
Yes. The credit rule tracer replays each decision under the policy version in force and the one before, and names the rule whose change made the difference. For a proposed change it replays the whole vintage — loans declined, volume given up, early defaults avoided — and every proposal passes fair-lending review first.
Yes. The onboarding checker looks at the autopay authorisation, the first due date against the signed contract, the welcome letter, statements and every payment attempt. In the working solution it found 13 borrowers whose autopay the app never saved; the fix sent to loan servicing re-enrols them, waives the late fee and corrects the late marks with the bureaus.
Yes. The agents propose; the credit risk analyst confirms or changes every cause, each fix goes to a named owner who accepts it, and the Consumer Credit Risk Director signs the monthly report. The agents never contact a borrower, change a rule, block an account, waive a fee or alter credit reporting.
Loan servicing, loan origination, your decision engine and its policy versions, fraud checks and device intelligence, channel and partner data, and the app release log. It reads from each what an early default's file needs.
The Agentic Solution Engine builds and deploys it from your requirements and documents — your early default definition, credit policy versions, owners and a recent vintage — and it goes live once every quality gate has passed. We will walk you through it on your own early defaults first.
Applications verified in minutes, each figure tied to its document.
1-min demoSynthetic Identity Fraud DetectionRings caught at application, not at charge-off, each hold explained.
1-min demoAdverse Action Notice AutomationEvery notice out inside 30 days, each reason traced to the decision.
1-min demoWe’ll run Early Payment Default Review on a recent vintage of your own early defaults.