Force-Placed Insurance Under Regulation X 1024.37: What an AI Servicing Agent Has to Get Right Before the Charge Posts

The Servicing Workflow That Produces the Most Avoidable Findings

Force-placed insurance is the servicing action that costs borrowers the most money for what is, almost always, a paperwork problem at the servicer or the carrier. The CFPB has run examinations and brought enforcement actions on force-placed practices repeatedly, and the rule itself, 12 CFR 1024.37, is more prescriptive than most of Regulation X. The institutions that take findings on this are not failing because the rule is ambiguous. They are failing because the noticing engine, the carrier's data feed, the escrow analysis, and the credit-bureau furnisher are not synchronized, and the AI agent that sits on top of the servicing queue inherits whichever one is wrong first.

We build the agent that handles hazard-insurance lapse and reinstatement on mortgage servicing portfolios. The design below is the one that survives a state mortgage exam and the one the borrower-attorney bar cannot use against the servicer in a subsequent action.

The Reasonable-Basis Standard, Not the Lapse Date

The rule at 1024.37(b) prohibits a servicer from charging a borrower for force-placed insurance "unless the servicer has a reasonable basis to believe the borrower has failed to comply with the mortgage loan contract's requirement to maintain hazard insurance." The standard the agent has to encode is "reasonable basis," not "the carrier file said the policy lapsed." A carrier file that shows a lapse is one input to a reasonable-basis determination. It is not the determination.

The agent's job at the moment the carrier flags a lapse is to assemble the evidence the rule expects the servicer to consider: the borrower's escrow record, including any disbursement that was sent to the carrier in the period preceding the alleged lapse; the inbound mail and document log, including any declarations page the borrower submitted that the imaging queue may not have indexed; the call and chat transcripts in the trailing window, including any conversation in which the borrower mentioned the policy; and the carrier's prior cycle data, which sometimes shows the policy renewed under a slightly different name or policy number the matching engine did not connect. Only after the agent has run these checks and the evidence still supports the conclusion that coverage lapsed does the noticing engine arm.

A reasonable-basis check that the agent runs and writes into the file is the artifact that defends the eventual charge. A check that no one ran is the artifact a borrower attorney will use to argue the charge was imposed without a reasonable basis, and the rule's affirmative-defense posture is not friendly to a servicer whose file does not show the work.

The Two Notices and the Clocks They Run

1024.37(c) requires the servicer to deliver an initial written notice at least forty-five days before charging the borrower for force-placed coverage, and 1024.37(d) requires a second reminder notice at least thirty days after the first and at least fifteen days before charging. The content of each notice is specified at the field level, including the statement that the lender does not have evidence of insurance, the request for the borrower to provide evidence, the cost or estimated cost of force-placed coverage, the fact that the coverage may be more expensive than the borrower's prior coverage, and the fact that the coverage may not cover the borrower's interest in the property.

The agent maintains both notices in versioned templates that the compliance team has approved, and it composes the borrower-specific fields from the system of record rather than letting the model write any of them. The cost figure, the carrier name, the policy effective date, the property identifier, and the borrower's prior coverage are all system data. The narrative connective text is templated. The model's role is delivery scheduling, channel selection per the borrower's preferences within the bounds the rule permits, and the recordkeeping. The model does not write the dollar amount, the date, or the carrier name, because every one of those is a figure the rule and the regulator measure against.

The clocks themselves are the place a servicing program most often slips. The forty-five-day clock runs from the date the initial notice is delivered, not from the date the lapse was detected or the date the notice was queued. The thirty-day-and-fifteen-day spacing is two separate constraints on the second notice, and the rule's "delivered" language imposes a presumption of receipt the agent has to compute correctly for the delivery method used. The institutions we work with that get findings on this are almost never the ones whose mail vendor delivered late; they are the ones whose system marked the notice as sent on the day the queue produced the file, rather than on the day the mail vendor actually mailed it. The agent reconciles the queue-produced date against the carrier-confirmed mail date and uses the later one for clock purposes, because that is what the rule reads.

The Renewal Notice and the Annual Cycle

1024.37(e) requires the servicer to deliver an annual notice at least forty-five days before charging the borrower for the renewal or replacement of existing force-placed coverage. The renewal-notice path is structurally identical to the initial-notice path and operationally separate, because the renewal cycle does not begin from a lapse event; it begins from the anniversary of the force-placed coverage in effect. The agent runs a parallel state machine for accounts already in force-placed status that produces the renewal notice on the renewal anniversary minus the forty-five-day window, regardless of any other event on the account. A servicer that loses the renewal cycle because the system was waiting for a lapse event that never came is in violation of (e) even when (c) and (d) are clean.

Bona Fide and Reasonable Charges

1024.37(h)(1) prohibits the servicer from charging the borrower for force-placed coverage that is not bona fide and reasonable. Charges that are commission revenue back to the servicer or its affiliates from the placement carrier sit at the center of the CFPB's historical enforcement on this rule, and the institution's bona-fide-and-reasonable posture is a documented one at the placement level. The agent's role here is bookkeeping: it writes the premium amount the borrower is charged, the carrier identification, the coverage period, and the policy benefits into the loan record and the disclosure, and it surfaces any mid-cycle change in the carrier's rate as an exception the operations lead reviews against the servicer's bona-fide-and-reasonable analysis. The agent does not validate the analysis itself, because that is a business and legal determination at the placement level, but the agent's record is what an examiner will use to test whether the servicer applied its analysis consistently.

The Refund Duty That Most Failure Modes Live Under

1024.37(g) requires the servicer, within fifteen days of receiving evidence demonstrating that the borrower had hazard insurance in place during a period for which the servicer charged the borrower for force-placed coverage, to cancel the force-placed coverage, refund any premium charged for the duplicative period, and remove any related fee. The refund duty is mechanical and time-bound, and it is where servicers we have audited most often miss the rule by a margin large enough for a complaint to materialize.

The agent's design for this turns the moment the borrower submits evidence into a hard interrupt. Evidence can arrive as a declarations page faxed in, an email forwarded by an agent, a document uploaded through the portal, a verbal confirmation in a call the agent classifies into an evidence-of-coverage category, or an inbound carrier file confirming retroactive coverage. Each path lands in the same case type, with the same fifteen-day clock starting at the timestamp the evidence was received by the servicer in any form, not at the timestamp the operations team reviewed it.

The fifteen-day clock drives three workstreams that have to complete inside it: the cancellation of the force-placed policy with the placement carrier, the refund of the premium and any related fee to the borrower's account, and the correction of any escrow analysis that ran against the now-canceled charge. A refund that the operations team processed in nine days but did not push to the borrower's account because the escrow re-analysis was queued for the next cycle is still a violation, because the borrower has not been made whole. The agent enforces the three-workstream completion check inside the clock and surfaces any one of them that has not closed by day twelve as an escalation, so the failure mode is caught before the rule's deadline rather than after.

The Credit-Bureau Furnishing Loop That Is Easy to Miss

When force-placed insurance is later canceled because the borrower had coverage all along, the related charges and any escrow shortfall they produced may have caused a delinquency the servicer furnished to the credit bureaus. The FCRA Section 623(a) accuracy duty and the duration finding in the CFPB's April 2024 Supervisory Highlights both reach this situation, and we have written separately on the FCRA furnisher-accuracy architecture. The agent's connection to that pipeline is the part of the force-placed flow that programs most often miss.

When the cancellation and refund complete, the agent runs a furnishing correction in the same case rather than waiting for the next cycle. Any past-due reporting that resulted from the now-reversed charges is suppressed, the bureaus are notified, and the change is logged with the basis. A servicer that refunded the borrower but kept the bad tradeline up for another two months is a servicer whose force-placed program does not survive a sampling test.

The Failure Mode We Engineered Against

The pattern we changed on the first program we ran this design on was that the lapse-detection feed from the carrier and the inbound evidence-of-coverage queue were two separate inboxes that lived on two different sides of the operations house. The lapse feed armed the noticing engine within hours. The evidence-of-coverage queue ran on a daily cycle that the operations team triaged when capacity allowed. Borrowers who emailed a declarations page on day forty of the forty-five-day clock had a charge post on day forty-six because their email was queued and the lapse feed continued running uninterrupted.

What we changed is that any inbound communication classified into the evidence-of-coverage category is a hard interrupt on the force-placed clock for that loan. The clock pauses, the case routes to the reasonable-basis re-evaluation, and the charge does not arm until the operations team makes the call. The fifteen-day refund clock starts simultaneously if the evidence supports continuous coverage. The agent does not let two adjacent inboxes process a borrower's account on conflicting timelines, because that is the structural problem that produces the avoidable charges this rule was written to prevent.

What the Force-Placed File Contains

For each account where force-placed coverage was contemplated, initiated, or canceled, the file includes the carrier's lapse signal with the date and the underlying policy data; the reasonable-basis evidence-gathering log with the inputs the agent considered; the initial notice with the delivery method, the producer date, and the carrier-confirmed mailed date; the second notice with the same three timestamps; the renewal notice in each subsequent year force-placed coverage stayed in effect; the placement record with the carrier, premium, coverage period, and the cross-reference to the servicer's bona-fide-and-reasonable analysis; every inbound communication classified into the evidence-of-coverage category with the timestamp and the case it interrupted; the refund and cancellation events with the fifteen-day clock measurement; the escrow re-analysis output and the borrower statement that reflects it; and the credit-bureau furnishing correction with the date and the bureaus notified.

A servicer that can produce this per account in a sampled exam is the one whose force-placed program holds up. A servicer whose file requires reconstruction from three systems and a manager's memory is the one whose program does not.

The Trade-Off

A force-placed program that runs this way places fewer policies, refunds faster when the placement was wrong, and adds latency to the moment a charge first arms because the reasonable-basis evidence has to assemble before the notice goes out. The placement carrier sees fewer premium dollars and the operations team sees more case work in the trailing window of every lapse. The return is fewer findings on noticing, fewer cure costs on refunds the rule's clock caught, fewer credit-bureau corrections the bank has to push under pressure, and fewer borrower complaints that escalate to the AG's office because the borrower's documented coverage was ignored. Force-placed insurance is the servicing program where doing the rule's work upfront is dramatically cheaper than doing the rule's cleanup later, and the agent's contribution is to make the upfront work actually happen on every loan, every cycle.