Skip to content
Sei

Regulation E Error Resolution with AI Agents: A 1005.11 Playbook for Dispute Intake

5 min readPranay Shetty
Share

Why Disputes Are the Workflow to Automate, and the One Most Likely to Bite

Error-resolution disputes are among the highest-volume, most repetitive workflows in retail banking, and they are almost entirely structured. A customer reports a transaction they did not authorize or a payment that posted wrong, the institution investigates, and a clock governs every step. That combination, high volume and structured inputs, is what makes the intake side a strong fit for an AI agent.

It is also the workflow where a small process miss compounds into a pattern. A single late provisional credit is a one-off. The same automation applying the wrong clock across thousands of disputes is the kind of systemic finding that draws an enforcer. We put agents on this workflow at banks and credit unions, and the design below is what keeps the productivity without creating the pattern.

The Clock Is Statutory and Does Not Care That the CFPB Pulled Back

It is worth being clear about why this is not a place to relax in 2026. The error-resolution duties live in the Electronic Fund Transfer Act and in Regulation E at 12 CFR 1005.11. They are statutory. A CFPB operating at reduced capacity does not change the text, and the EFTA carries a private right of action with statutory damages and attorney's fees, so a missed clock is a plaintiff's claim whether or not an examiner ever looks. State regulators that have stepped into the supervision gap read the same regulation. The institution that treats reduced federal examination as reduced exposure on Reg E has misread where the exposure comes from.

The Timing Rules an Agent Has to Enforce

The agent does not get to round these or treat them as targets. From 1005.11(c), the deadlines that govern a notice of error:

  • The institution investigates and determines whether an error occurred within 10 business days of receiving the notice
  • It reports the results to the consumer within three business days after completing the investigation, and corrects a confirmed error within one business day of the determination
  • If it cannot finish within 10 business days, it may take up to 45 days, but only if it provisionally credits the disputed amount within those first 10 business days and notifies the consumer of the amount and date of the provisional credit within two business days of crediting, giving full use of the funds during the investigation
  • For a notice involving an account in its first 30 days after the first deposit, the 10-business-day window becomes 20 business days and the 45-day window becomes 90 calendar days
  • For point-of-sale debit-card transactions and transfers not initiated within a state, the 45-day window becomes 90 calendar days

The consumer has 60 days from the statement to give notice of the error, and the institution may require written confirmation of an oral notice within 10 business days, but only if it told the consumer that requirement and an address at the time of the oral notice. These conditions are exactly the kind of branching logic a person under volume gets wrong and a timing engine does not.

What the Agent Does and Does Not Do

We draw the line in the same place across every compliance workflow we automate. The agent does the intake and the clock; the human owns the determination.

The agent handles the customer conversation, captures the structured claim, classifies the transaction type because the type sets the clock, opens the case in the system of record with the correct deadlines computed, requests and logs any written confirmation, drafts the customer-facing notices, and tracks every deadline to the day. It does not decide whether an error occurred, it does not deny a claim, and it does not sign the notice of results. The determination is a human's, recorded under that person's name. We keep the agent out of the decision because the decision is where judgment and liability sit, and because an agent that both investigates and adjudicates a dispute is the structure an enforcer most distrusts.

The Verbal-Claim Trigger Counsel Made Us Add

The first version of our intake agent waited for the customer to file a formal dispute through the structured flow. Compliance counsel at one of our bank customers caught the problem in review. Regulation E does not require a form. An oral notice that describes the error with enough detail is a notice of error, and the clock starts then, not when the customer later completes a form.

So we added a trigger that treats a qualifying verbal statement as the notice of error in its own right. When a customer says, in any words, that a transaction was not authorized or that a transfer posted incorrectly, and identifies the account and the amount or transaction well enough to investigate, the agent records that moment as the receipt of notice, timestamps it, and starts the clock immediately. It still guides the customer through the structured intake, but the clock does not wait for the form. We backtested this against historical chat and call transcripts and found verbal notices that the old form-first process had effectively ignored, which is exactly the gap that turns into a late investigation.

Provisional Credit Is a Control, Not a Courtesy

The 45-day and 90-day extensions are only available if the institution provisionally credits the account inside the first 10 business days and tells the consumer about it within two business days of doing so. Programs lose this by treating provisional credit as a discretionary customer-service gesture rather than the precondition it is for using the longer investigation window.

The agent enforces the dependency directly. If the investigation will not close within the short window, the system either provisionally credits and fires the two-business-day notice automatically, or it flags that the institution has chosen not to credit and therefore must finish inside the short window. There is no third state where the institution quietly takes 45 days without having credited, because that is the violation. The agent also tracks the written-confirmation interaction, since an institution that required written confirmation and did not receive it within 10 business days is relieved of the provisional-credit duty, and that branch has to be applied correctly rather than assumed.

The Notice of Results Most Programs Get Wrong

When the investigation finds no error, or a different error than the one claimed, Regulation E requires more than a denial. The institution has to mail or deliver an explanation of its findings within three business days of finishing, tell the consumer it will debit any provisionally credited amount, and inform the consumer of the right to request the documents the institution relied on. Denials that skip the documents-on-request language are a recurring exam and litigation theme.

The agent drafts the notice of results from the case record so the required elements are present by construction: the determination, the basis, the handling of any provisional credit, and the statement of the consumer's right to the underlying documents. A human reviews and signs. When the consumer later asks for the documents, that request is itself a tracked event with its own response obligation, not an email that sits in an inbox.

The Audit File

Because the real examiner in 2026 may be a state regulator or a plaintiff's lawyer pulling records, the file has to stand on its own. For every dispute we retain the timestamped notice of error and how it arrived, the transaction classification and the resulting deadlines, the provisional-credit decision with its date and the date of the consumer notice, the investigation record and the human who made the determination, the notice of results as delivered, and any documents-on-request activity. A bank that can produce this file for a sampled set of disputes in an afternoon is in a different position than one reconstructing dates from disconnected systems months later.

The Trade-Off

An agent on dispute intake does not reduce the investigation team, and we say so before a contract. It removes the clerical load, the missed verbal notices, and the timing slips, which is most of where Reg E liability actually comes from, and it leaves the determination with the people who should be making it. The return is fewer late credits, fewer defective notices, and a file that survives whoever decides to look. The institutions that get the most from it are the ones with dispute volume high enough that the timing errors were already happening quietly. For a small shop with light volume, the case rests on accuracy and on the audit file, not on headcount, and we make that distinction on the first call rather than overselling it.

Pranay Shetty

Pranay Shetty

CEO & Co-Founder

Related Articles

Compliance

Governing Personal-Agent Traffic in Mortgage: Model Risk, Fair Lending, UDAAP and the Audit Trail

When borrowers' agents start talking to a lender's systems, the lender has a new channel to govern — one where the other party is software, the lender's own agent may be answering, and the record of what happened is easy to lose. How existing frameworks apply: GSE AI governance requirements, model risk and third-party risk guidance, ECOA and fair lending, UDAAP, and record retention under Regulation B and Regulation Z.

4 min read
Read more
Compliance

The Exam Starts With a Document Request, Not a Conversation: Preparing an AI-Assisted Mortgage Operation for Supervision

When a CFPB or state examiner opens a supervisory exam, the first thing you get is an information request, and an AI-assisted operation has to answer questions the exam procedures were written before anyone was asking. What examiners actually request, how the modules in the CFPB manual map onto an automated file, and the evidence that lets you answer "how did the machine decide this" in a sitting instead of a scramble.

6 min read
Read more
Compliance

When the Model Writes the Ad: Mortgage Marketing Copy Under the MAP Rule and Reg Z 1026.24, and the Review Gate Before a Generated Line Ships

Generative AI now drafts mortgage emails, landing pages, and social copy at a scale no compliance team has reviewed a piece at a time. The moment a model writes a sentence about a rate or a payment, two regimes bite: Reg Z 1026.24 triggering terms and the MAP Rule, Regulation N at 12 CFR 1014, which bars material misrepresentation about a mortgage credit product and makes you keep every materially different version for 24 months. Here is the gate we put between the model and the send, the 'no closing costs' line it caught, and why AI turns the recordkeeping rule from a burden into a byproduct.

5 min read
Read more

You Ain't Seen Nothin' Yet

Book a Demo

Pack up some of your complex historical files — any loan type, any investor. We run them through intake, income and condition clearing, and in 30 minutes you see every condition we created and cleared efficiently for your own team, and why.

  • Any loan type, any agency guideline or custom investor overlays.
  • Every finding cited to the guideline or document it came from

Case study highlight

Loan review fell from 4 hours 50 minutes to 47 minutes per file, and 78% of items now clear at high confidence — so QC touches only the exceptions.

Pre-Close QC Agent · Better Mortgage— read the case study

Please provide your full name so we know how to address you.

Tell us which company you represent so we can personalise our response.

Use your work email. One of our experts will get back to you.

Which desks would you like to discuss?*

Choose the desks you’d like us to cover. Pick “Not sure yet” if you’d rather we worked it out on the call.

Roughly, so we bring the right person to the call.

Complete the verification to submit the form.