Adverse action notice automation: every reason on every letter traced to the factor that decided it
Every notice out inside 30 days, each reason traced to the decision.
See one case, screen by screen ↓Software that turns each decline, counteroffer or line decrease into the notice the applicant receives, with the specific principal reasons, the ECOA statement and, when a credit report was used, the FCRA disclosure. Adverse Action Notice Automation adds six agents that read the decision, trace each reason to the factor or rule behind it, check the letter and keep the clock; the compliance analyst releases every notice.
How to choose AI solutions for lending →
Updated
Regulation B gives a lender 30 days from a completed application to send an adverse action notice with the specific principal reasons — and the clock runs while the file sits in manual review. When underwriting decides on day 24, compliance has six days left to send the notice.
The reasons themselves often come from a spreadsheet. The decision engine emits model reason codes, a code map turns each one into a sentence, and a template fills the letter. When a new model version gives a code a new meaning and the map is not changed, every letter it touches states a reason the model never scored — a sample-form phrase like “insufficient credit” — and nobody sees it until the monthly sample or an examiner does.
Estimated split, modelled on the hero case: 11.4 days median by hand, 2.1 days with the agents.
Six specialist agents compose each letter, trace every reason to the factor that decided it and check it against the rules; the analyst releases it.
Dana Whitlock’s $24,000 personal loan was declined after underwriting. Her notice is due Thursday, three days from now. Here is what happens next, screen by screen, in the working solution.
Elena’s home is the notice itself: Dana Whitlock’s decline, set as the letter Dana will receive, with the 30-day clock beside it — day 27 of 30, clock started Tue 15 Sep, due Thu 15 Oct. It is notice 1 of 296 due within three days, out of 1,846 open. Beside the letter, 14 rules from Regulation B, the FCRA, and Ohio and house rules tick as the letter is checked, and land on 12 of 14.
“Not specific. ‘Insufficient credit’ is a sample-form phrase from map v6.3. Model v7 scored R14 · length of credit history: oldest account 11 months, −41 points.”
The decision reader takes PL-2609-48213: Personal Loan Score v7 at 588 against a cutoff of 620, policy rule CP-4.2 failed at 46.2% debt-to-income against a 43% limit, clock from the completed application on 15 Sep. The reason tracer ranks one policy rule and five model factors and picks four principal reasons: CP-4.2, R14, R03 and R22.
Reason-map checker: “map v6.3 prints ‘insufficient credit’ for R14 — a sample-form phrase; model v7 documentation (MD-PLS7 §4.3) defines R14 as length of credit history · 214 notices affected.”
Elena presses Rewrite from the factor. “Insufficient credit” is replaced, live, by “Length of credit history: oldest account 11 months”. The margin note turns green — Accurate, Specific, reading grade 5.2 — and the two Regulation B checks that failed, specific principal reasons and only factors actually scored, pass. The letter now meets 14 of 14 rules.
Dana’s letter is not the only one. The map in use for Personal Loan Score v7 is v6.3, approved for v6, and it still prints “insufficient credit” for R14 — words carried over from v6, when R14 meant too little credit information; that factor moved to R38. The words were used 1,160 times in 30 days: 214 notices are held in the queue and 946 went out between 1 Sep and 9 Oct.
New words: “Length of credit history: oldest account [months] months” — the months filled from each applicant’s own R14 value.
Elena proposes map v7.1. Jun Park approves as the model owner and Wendell Okafor as the compliance officer — both must approve before a new map is used. The 214 held notices are rewritten with each applicant’s own months and checked again; none is held.
The reason-map checker never changes the map — two people approve.
The October accuracy sample traced 60 September letters again from scratch: 57 accurate, 3 not specific, all R14. The scan of every sent letter found 946 with the same words — 611 by e-delivery, 335 by mail, 931 applicants. The decisions do not change, only the words. A corrected notice is drafted for each, and whether they go is Wendell’s decision.
The notices queue is sorted by days left on the clock, which starts at the completed application, not the decision. Among those that need a person: Tomasz Kowal’s credit card decline, due tomorrow, was an underwriter override with no reason recorded; Deshawn Morrow’s must give the identity rule, not score factors.
Elena approves Dana’s notice for the 16:00 batch. The batch holds 1,785 notices across personal loans, auto, credit cards and HELOC — 1,142 by e-delivery, 643 by mail — with nine checks passed on every one: reasons traced, map version matching the model version, at most four reasons, the bureau disclosure, the ECOA statement, state statements, the legal clock, plain words and E-SIGN consent.
Elena signs with the meaning recorded — “I have reviewed the checks and release these notices” — ahead of the 16:00 cut-off. The clock keeper confirms 1,142 e-delivered and the rest with the print vendor for first-class mail that night.
“Delivered · every notice inside its 30 days.”
1,846 open clocks, 30-day and 90-day; 296 due in three days; open counteroffers on their 90-day clock; 14,270 notices delivered in the last 30 days, with 71 returned letters re-sent to the address on the bureau report. Median days from decision to notice: 2.1, down from 11.4 before July. Late in 2026: none.
The same notice, seen by the people who carry it — compliance, model risk, underwriting and correspondence — before and after.
Six specialist agents compose each letter, trace every reason to the factor that decided it and check it against the rules; the analyst releases it.
Reads each decision record as it is made: product, action, model version, score, cutoff, policy rules and the completed-application date that starts the clock.
Ranks the factors that drove each decision — the model’s own explanation and any failed policy rule — and picks up to four principal reasons, each traced to its factor.
Checks the words for every reason code against the model’s documentation for the version that decided, and holds any notice whose words do not match.
Checks the FCRA content when a credit report was used: score, range, date, key factors, the bureau’s contact, the free report and the dispute right.
Writes each letter from the approved template for its product, action and state, in plain words, with the ECOA statement, the federal agency and any state statement.
Keeps every 30- and 90-day clock, builds the daily batch, confirms delivery, re-sends returned mail and keeps the record 25 months.
Releases every notice; map changes need two approvers. The agents propose; a named person decides.
The compliance team can ask in plain words — about one applicant, one code or the week ahead — and every answer points to its source.
Why was Dana Whitlock declined?
Two things decided it. Her debt payments after the new loan would be 46.2% of her income — $2,389 of $5,170 a month — and policy rule CP-4.2 stops at 43%. And Personal Loan Score v7 gave 588 against a cutoff of 620, mostly because her oldest account was opened only 11 months ago (R14), she uses 87% of her limits (R03) and has 5 recent inquiries (R22). Those four are the reasons on her letter.
Which letters already went out with the vague R14 reason?
946 personal-loan letters sent between 1 Sep and 9 Oct said only “insufficient credit” for R14, while model v7 scored the length of credit history. The October sample found 3 of them. 611 went by e-delivery, 335 by first-class mail. No decision changes — only the words. A corrected notice is drafted for each one; Wendell Okafor decides whether they go.
What is due in the next three days?
296 notices, none late. 212 are ready for today’s batch, 79 are held on R14 and 5 need a person — the first is Tomasz Kowal, due tomorrow, whose decline was an underwriter override with no reason recorded. Deshawn Morrow’s, due Wednesday, must give the identity rule, not score factors.
Is Dana’s notice complete under Regulation B and the FCRA?
Yes — 14 of 14 required parts are there: the action, our name and address, four specific reasons, the ECOA statement and the CFPB’s address; the Keystone score 641, its range, date and four key factors with inquiries, the bureau’s contact, its free report within 60 days and the dispute right; the Ohio statement; and it leaves three days before its due date.
12 screens from the working solution, on its sample data. Pick one to see it large.
The rules every notice is drafted and checked against are settings. Changes are versioned and on the audit trail.
| Setting | Default | Choose from |
|---|---|---|
| Escalate when | 5 days left | 3 · 5 · 7 days left |
| Batch release | Daily at 16:00 | Daily at 16:00 · twice daily, 11:00 and 16:00 · hourly |
| Principal reasons on a letter | Up to 4 | Up to 3 · up to 4 · up to 5 when inquiries are a factor |
| Map must match the model version | On | Always on |
| Plain-language level | Grade 8 | Grade 6 · 8 · 10 |
| Map changes approved by | Model owner + compliance officer | Both must approve |
| Keep the record | 25 months | 25 · 36 · 60 months |
| Monthly accuracy sample | 60 notices | 30 · 60 · 100 notices |
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 lending data · “target” = the design goal, measured in the live solution · “estimated” = modelled on the hero case · “typical” = published rules (12 CFR 1002.9 and Official Interpretations: 30 days from a completed application; specific principal reasons; factors actually scored · 15 U.S.C. §1681m(a)). People, lenders and applicants named on this page are fictional — characters and sample data in the working solution.
Software that turns each decline, counteroffer or line decrease into the notice the applicant receives, with the specific principal reasons, the ECOA statement and, when a credit report was used, the FCRA disclosure. Adverse Action Notice Automation adds six agents that read the decision, trace each reason to the factor or rule behind it, check the letter and keep the clock; the compliance analyst releases every notice.
The reason tracer takes the model’s own explanation for the decision and any failed policy rule, ranks them by effect, and picks up to four principal reasons. The reason-map checker compares the words for each code with the model documentation for the version that decided, and holds any notice whose words do not match.
The map is checked against the new model documentation every night and on every new version. Where a code’s meaning differs, every notice that would use it is held, the letters already sent with it are counted, and new words are proposed for the model owner and the compliance officer to approve.
Yes. When a credit report was used, the bureau disclosure checker confirms the score, its range and date, the key factors — with inquiries listed whenever they are a key factor — the bureau’s name, address and phone number, and the free report and dispute rights.
Yes. Every month a sample of sent notices is traced again from scratch. When a reason is wrong, every sent letter that carried the same code and version is found and a corrected notice is drafted for each one; the compliance officer decides whether they go.
Yes. The agents never change a decision, change the code map or send a letter. The compliance analyst approves notices and signs each batch, a map change needs the model owner and the compliance officer, and corrected notices are the compliance officer’s decision.
Yes. When to escalate, the batch release time, the number of principal reasons, the plain-language reading grade, who approves map changes, templates by product, action and state, delivery, how long the record is kept and the size of the monthly accuracy sample are all settings.
The Agentic Solution Engine builds and deploys it from your requirements — your credit policy, reason-code maps, model documentation and notice templates — and it goes live once every quality gate has passed. We will walk you through it on a sample of your own declines 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 demoCard Dispute ManagementNo missed Reg E or Reg Z deadline; claims investigated in minutes.
1-min demoWe’ll run Adverse Action Notice Automation on a sample of your own declines, reason codes and templates.