Fintech partner oversight for sponsor banks — every partner’s loans tested every week, not sampled every quarter
Quarterly spreadsheet reviews become weekly evidence, ready for the examiner.
See one case, screen by screen ↓Software a sponsor bank uses to monitor the fintech partners that lend under its charter — their loans, adverse action notices, complaints, marketing and BSA controls — and to keep the evidence its examiners ask for. Fintech Partner Oversight has six agents test each partner’s weekly files in full and draft findings; the bank’s oversight manager decides on each one.
How to choose AI solutions for lending →
Updated
A sponsor bank originates its fintech partners’ personal loans, point-of-sale loans and cards under its own charter, so it answers to its examiners for each partner’s underwriting, adverse action notices, complaints, marketing and BSA controls. Many oversight teams test that once a quarter: each partner sends a loan tape, a complaint log and a folder of ads, and two analysts sample them in spreadsheets and write the pack for the risk committee.
A partner like Fernquill decides about 4,900 applications a week, and a quarterly review samples 25 to 60 loans from the whole quarter. A credit model changed without the bank’s approval, a notice template that gives the old reason, an ad that says “no credit check” — each one can start the week after a review and run for three months, until the complaints arrive or the examiner finds it first.
Estimated split for one quarterly partner review by hand, against one week’s run with the solution.
Six specialist agents test every partner's loans, notices, complaints and ads each week and chart them against control limits; the oversight manager decides.
Brannock Valley Bank has six fintech lending partners and $1.84B outstanding. It is Monday 12 October, the FDIC is on site in 35 days, and Fernquill’s tape is late. Here is what happens next, screen by screen, in the working solution.
The weekly run started at 06:00 with five of six tapes in. The agents charted Cobalt Mesa Pay, Nimbleleaf, Hearthway, Pocketrise and Wrenfield Auto — 120 tests, one signal — and drafted two findings for her review. At 08:52 Jordan Pike uploads Fernquill’s tape for 5–9 October: 4,918 decisions, 1,342 funded for $14,612,400, 3,576 declined.
“Charting it runs all 24 tests on every loan, notice and complaint and plots this week’s points against Fernquill’s own limits.”
Dana presses Chart Fernquill’s week and the points land day by day. The share of funded loans priced above the program grid has a centre of 0.3% and an upper limit of 0.8%. Monday is 0.0%. Tuesday is 18.5% — 52 of 281 loans — and every day after it stays far above the limit.
“212 loans priced above the program grid since Tuesday.”
She opens the 212 loans: score, grid tier, grid APR, APR charged and the points over, loan by loan — $2,531,800 funded, 3.4 points over the grid on average, about $173,600 of extra interest over the loans’ lives. Every APR is under the 36% cap, so the cap test passed. Only the grid test caught it.
Cause: from Tue 07:00 the pricing engine reads model v7.2’s cash-flow band, which the grid does not know.
Every decline is compared with the notice that went out. On 412 of 3,576, the decision’s principal reason is CF-03, “Irregular deposits in linked bank account”, and the notice says “Income insufficient for amount requested”. All 412 were decided by credit model v7.2, live since Monday 5 October.
Reg B §1002.9(b)(2): the notice must give the specific principal reasons for the decision. The agent flags the mismatch; a person decides.
Five tests for six partners, each a small control chart against that partner’s own last 12 weeks. Fernquill shows four signals: pricing above the grid, approval rate 27.3%, notice accuracy 88.5% and complaints 8.7 per 10,000 accounts. Its fair lending ratio, 0.91, stays inside its limits. Hearthway’s complaints are above the line again, at 27.5.
Fernquill’s score falls from 84 to 68: 16 of 24 tests passed. The reporter drafts five findings, three High and two Medium, each with its rule and its evidence — the notices, the unapproved model change under program agreement §6.3, the 212 loans, double-debit complaints that went 9 → 14 → 22 → 31 a week, and an ad.
The reporter drafts. It never accepts, closes or escalates an issue.
Of 38 live creatives sampled, one is not approved: social ad FQ-SOC-0917, live since 17 September and seen 1.2 million times. “No credit check” — Fernquill pulls a full credit report on every application. “Rates from 8.99% APR” — a rate claim must show the range, 8.99%–35.99%.
Dana accepts the five findings; Jordan Pike sees them in the partner portal. She escalates the three High findings to the Partner Risk Committee with a recommendation: return to model v7.1 unless v7.2 is approved by 16 October, send corrected notices to the 412 applicants and reprice the 212 loans by 26 October, and test daily until the score is above 70.
“Martin Ayres, Chief Compliance Officer, signs the escalation; the committee meets Thu 15 Oct.”
The remediation request to Jordan Pike lists every finding with its fix and due date: corrected notices to the 412 and CF-03 mapped to its own reason; v7.2 submitted with validation and fair lending testing, or back to v7.1; the 212 loans repriced from day one with excess interest refunded; the remaining 284 double-debited accounts refunded and three CFPB complaints answered; the ad taken down.
Evidence for the 16 November exam updates with the morning’s work: 45 of 48 items current, 94%, with 13 weeks of dated tests per partner and every finding carrying its evidence. The three gaps left are named: two Hearthway findings past due, a Pocketrise finding past due, and a committee minute on the Hearthway escalation.
The same Fernquill week, seen by the five people who carry it — the bank’s oversight team, its compliance and BSA officers, its auditor, and the partner.
Six specialist agents test every partner's loans, notices, complaints and ads each week and chart them against control limits; the oversight manager decides.
Reads each partner’s weekly loan tape, reconciles funded loans to the bank’s sub-ledger and tests every loan against the program and the credit policy: APR range and cap, Military Lending Act MAPR, state limits, fees, score and DTI, and the model version against the approved list.
Compares every decline’s principal reason codes with the notice that was sent, and checks the 30-day timing, the number of reasons and the credit score disclosure.
Reads partner complaint logs and the CFPB portal feed, themes every complaint, and watches trends, spikes and the 15-day response clock.
Samples live ads, emails, checkout and landing pages and compares them with the bank’s approved library and the advertising rules.
Checks OFAC screening coverage, identity and device links, and how fast fraud referrals are worked.
Turns failed tests into findings with severity, rule, evidence and a proposed owner and due date; drafts scorecards, remediation requests, the examiner pack and the board report.
Accepts and assigns; the CCO signs escalations. The agents propose; a named person decides.
The oversight team can ask in plain words — or tell it what to send. Every answer points back to the tape rows, notices, complaints or ads it rests on.
Why did Fernquill’s score drop from 84 to 68?
Three domains fell on this week’s tape: adverse action 88 → 46 (412 notices give the wrong reason), marketing 86 → 62 (an unapproved “no credit check” ad) and loan tape 90 → 72 (model v7.2 on every decision; 212 loans above the pricing grid). Complaints also fell 84 → 66 on the double-debit spike. The other 16 tests passed, including the 36% APR cap, the Military Lending Act and OFAC.
Which loans were priced above the grid, and since when?
212 of the 1,074 loans Fernquill funded from Tue 6 Oct to Fri 9 Oct. Monday was on the grid; from Tue 07:00 the pricing engine read model v7.2’s cash-flow band, so the share above the grid stepped from 0.3% to 18.5–21.8% a day — far above the upper control limit of 0.8%. The 212 average 3.4 points over their grid rate on $2,531,800 funded; every APR is still under the 36% cap, so only the grid test catches it.
Is the double-debit problem getting worse?
Yes. Double-debit complaints went 9 → 14 → 22 → 31 a week since Fernquill moved its payment processor on 14 Sep. 1,186 accounts were debited twice on 1 Oct for $318,940; Fernquill has refunded 902. Three CFPB portal complaints are past the 15-day response — the oldest is 21 days.
Draft the remediation request to Fernquill
Drafted for Jordan Pike, each fix tied to its evidence: corrected notices to the 412 applicants and CF-03 mapped to its own reason; submit v7.2 under §6.3 or return to v7.1; reprice the 212 loans to the grid from day one and refund excess interest; refund the last 284 accounts and answer 3 portal complaints; take down FQ-SOC-0917 today.
13 screens from the working solution, on its sample data. Pick one to see it large.
The tests every partner is held to are settings the oversight team owns.
| Setting | Default | Choose from |
|---|---|---|
| Cadence | Weekly · Monday 06:00 | Weekly, Monday 06:00 · Daily, 06:00 · Weekly, Monday and Thursday |
| Partners on Elevated | Tested daily until the score is back above 70 | On · Off |
| Control limits | ± 3 standard deviations · last 12 weeks | ± 3 SD over 12 weeks · ± 3 SD over 26 weeks · ± 2.5 SD over 12 weeks |
| Pricing grid tolerance | 0.00 points | 0.00 · 0.05 · 0.10 points |
| Adverse action mismatch threshold | 0.5% of declines | Any share, in steps |
| Complaint spike | 2× the 13-week average | 1.5× · 2× · 3× the average |
| Due dates by severity | High 10 business days · Medium 30 days · Low 60 days | High 5, 10 or 15 business days · Medium 20, 30 or 45 days · Low 45, 60 or 90 days |
| Escalation to the Partner Risk Committee | Martin Ayres, Chief Compliance Officer | Chief Compliance Officer · Partner Risk Committee chair |
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 partner data · “target” = the design goal, measured in the live solution · “estimated” = our estimate · “typical” = published figures (Klaros Group via Sphinx: FDIC formal enforcement odds Q1 2023–Q1 2024, 15% for fintech partner banks vs 1.8% for other banks). Regulatory context: Interagency Guidance on Third-Party Relationships, 88 FR 37920 (6 Jun 2023) · 12 CFR 1002.9. The bank, partners, people, applicants and ads named on this page are fictional — characters and sample data in the working solution.
Software a sponsor bank uses to monitor the fintech partners that lend under its charter — their loans, adverse action notices, complaints, marketing and BSA controls — and to keep the evidence its examiners ask for. Fintech Partner Oversight has six agents test each partner’s weekly files in full and draft findings; the bank’s oversight manager decides on each one.
Third-party risk tools mostly track questionnaires, contracts and due diligence files. This tests the partner’s actual loan decisions, notices, complaints and live ads every week, from loan-level data, and charts the results against that partner’s own control limits.
For each partner and test, the centre line and limits come from that partner’s own last 12 weeks — the centre ± 3 standard deviations by default. A point outside the limits, 7 in a row on one side of the centre, or 6 rising or falling is a signal, and clicking it opens the loans, notices or complaints behind it.
Every decision on the tape. In the working solution, all 4,918 of Fernquill’s decisions for the week were tested against 24 tests, including the full 3,576 adverse action notices.
No. The agents read and test; they never edit a partner’s data or a decision, approve a creative, answer a complaint or file a SAR. The partner oversight manager accepts and closes findings, the Chief Compliance Officer signs escalations, and fraud referrals go to the BSA officer.
Each accepted finding with its owner, due date and evidence in the partner portal, and the rules that apply to them. Remediation requests list exactly what to fix and by when.
The examiner pack is kept current each week: dated tests per partner, the issue log with owners, complaint analysis, marketing reviews, BSA and OFAC testing and committee minutes, matched to the exam request list, with the remaining gaps named.
The Agentic Solution Engine builds and deploys it from your requirements — your program agreements, pricing grids, credit policy, notice templates and a few weeks of partner tapes — and it goes live once every quality gate has passed. We will walk you through it on your own partners’ files 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 Fintech Partner Oversight on a few weeks of your own partners’ tapes, notices, complaints and ads.