The Loss-Draft Desk: Where an AI Servicing Agent Moves a Hazard-Claim Check Toward a Repaired House Without Sitting on the Borrower's Money

The Check Made Out to Two Parties

A hazard claim starts the way the worst days in a borrower's life start, with a fire, a burst pipe, a hurricane, or a tree through the roof. Some weeks later the insurance carrier issues a check, and the borrower who expected to sign it and hire a contractor discovers that the check is made out to them and to their mortgage servicer together, and that they cannot cash it without the servicer's endorsement. That is the loss-draft process, and for a lot of borrowers it is the first time they learn that the servicer has a claim on the insurance money at all. The servicer holds the proceeds and releases them as the repair progresses, and the borrower experiences that as either a well-run process that got their house fixed or a bureaucratic wall between them and money that is theirs.

We build the servicing agent that runs the loss-draft workflow on servicer platforms, from the first notice of loss through the final disbursement, with the servicer's people owning the judgment calls the process reserves for them. The reason this desk is a good place for an agent is that most of the work is tracking, chasing, and scheduling against a checklist and a set of thresholds, which is exactly the work that goes slow when it depends on a human remembering to follow up, and slow is where the compliance risk lives. A loss-draft desk that holds a borrower's repair money longer than the process requires is not a neutral delay. It is a harm, and it is the kind of harm the consumer-protection regime reaches. What follows is where the servicer's authority actually comes from, the control that keeps the desk honest, and the line the agent does not cross.

Where the Authority Over the Proceeds Comes From

The servicer's right to hold and control the insurance proceeds is contractual, and it sits in the security instrument the borrower signed at closing. The uniform security instruments the GSEs publish carry a property-insurance section, Section 5 in the standard Fannie Mae and Freddie Mac uniform instrument, that gives the lender the right to require that insurance proceeds be applied to restoration or repair of the property when the restoration is economically feasible and the lender's security is not lessened, and, when it is not feasible, to apply the proceeds to the sums secured by the loan. So the servicer is not doing the borrower a favor by getting involved and it is not free to do whatever it wants either. It is exercising a specific contractual right, and that right has a purpose written into it, which is to see the collateral restored, not to collect the borrower's insurance money.

The operating detail of how the servicer exercises that right comes from the investor. The Fannie Mae Servicing Guide and the Freddie Mac Servicing Guide both set out how a servicer handles an insured-loss event on a loan the GSE owns, including when the servicer may release proceeds without monitoring, when it has to monitor disbursement against completed work, what inspections are required, and how quickly it has to act. Those guides are the servicer's operating rules for the desk, and a servicer that services GSE loans is bound to them through its servicing contract. For loans in other investors' portfolios the equivalent investor guide governs, and the common structure across all of them is the monitored-disbursement threshold.

The Threshold Is the Control, and the Agent Runs It Consistently

The single decision that shapes the whole loss-draft process is whether a given claim is small enough to release without monitoring or large enough to require monitored disbursement. Below the threshold, the investor guides generally let the servicer endorse the check and release the funds to the borrower without holding them against repair progress, because the loss is small relative to the loan and the collateral risk is low. Above it, the servicer has to hold the proceeds and release them in draws tied to inspections that confirm the work is actually getting done, because that is how the servicer protects the collateral the proceeds exist to restore.

The reason to put an agent on this decision is that the threshold is exactly the kind of rule that humans apply inconsistently under load. Two adjusters at the same servicer, looking at similar claims on a busy day, will make different monitoring calls, and the borrower on the wrong side of that inconsistency experiences a harder process than a neighbor with the same loss. The agent applies the investor's threshold the same way every time, classifies each claim as monitored or non-monitored against the specific investor's rule for that loan, and produces the reason for the classification in the file, so the servicer can show that the monitoring decision followed the guide rather than the mood of the desk. Consistency here is not a nicety. A servicer whose monitoring decisions vary loan to loan is a servicer whose loss-draft program looks arbitrary to an examiner and unfair to the borrower who drew the slower path for no reason on the file.

Moving the Money at the Speed the Repair Actually Goes

On a monitored claim, the process is a sequence of draws, and the agent's job is to keep the sequence moving at the pace of the repair rather than the pace of the paperwork. At first notice of loss the agent opens the claim, sends the borrower the document package the process requires, the insurance adjuster's report, the contractor's estimate or contract, the borrower's and contractor's endorsements on the check, and whatever the investor guide adds, and it explains to the borrower in plain terms what the funds are for and how the draws will work. An initial disbursement commonly releases a portion of the funds to let the work start, and subsequent draws release against inspections that confirm the work reaching each stage. The final draw releases on the completion inspection that confirms the house is whole.

The agent tracks each claim against those milestones and chases the missing piece before the claim stalls, because the way a loss-draft desk fails the borrower is almost never a decision to withhold money and almost always a document that was never chased and a claim that quietly aged in a queue. The agent knows which document is outstanding on every open claim, follows up with the borrower and the contractor on a schedule, schedules the inspections when the work reaches a draw milestone, and releases the draw when the inspection clears. The measure we hold the desk to is the age of the outstanding funds against the stage of the repair, because a claim where the work is done and the funds are still held is the specific failure the process is supposed to prevent, and it is the one that turns into a complaint and, if it is a pattern, into a supervisory finding.

Why the Speed Is a Compliance Question, Not Just Service

The consumer-protection exposure on the loss-draft desk does not come from a single tidy rule the way the force-placed insurance process comes from Regulation X 1024.37. It comes from the general prohibition on unfair, deceptive, or abusive acts and practices under the Dodd-Frank Act, and from the state statutes several states maintain governing how quickly a servicer must release insurance proceeds and, in some states, whether it owes interest on funds it holds. The CFPB has treated servicer mishandling of insurance proceeds as a supervisory matter, and the shape of the concern is consistent: a servicer that holds a borrower's repair funds longer than the process requires, that imposes documentation demands beyond what the situation warrants, or that fails to release funds after the work is verified, is doing something the unfairness standard reaches, because the borrower is substantially injured by being kept from money that is meant to fix their home and they cannot reasonably avoid the injury.

So the speed and the consistency the agent brings to the desk are not customer-experience extras that sit next to the compliance work. They are the compliance work. A desk that releases the non-monitored claims promptly, applies the monitoring threshold the same way every time, chases the documents so claims do not age, and releases each draw as soon as the inspection clears is a desk that does not generate the pattern of held funds and stalled claims that the unfairness analysis is built to catch. We instrument the agent to surface the age of every open claim and the reason any claim is stalled, so the servicer can see a claim aging before the borrower has to call and ask where their money is.

The Line the Agent Does Not Cross

The place the agent stops is the place the decision stops being a workflow and becomes an exercise of the servicer's contractual and investor-governed judgment. The clearest example is the total loss or the near-total loss, where the property may not be economically feasible to restore and the security instrument gives the lender the option to apply the proceeds to the loan balance instead of to repair. That election, repair the house or pay down the debt, is a decision with real consequences for the borrower and for the investor, and it is the servicer's and the investor's to make under the instrument and the guide, not the agent's to make as a routing rule. When a claim presents as a total loss, a payoff-adjacent situation, or a case where the proceeds may exceed what a feasible repair would cost, the agent assembles the file and routes it to the servicer's people with the facts organized, and a human makes the application decision.

The agent also does not decide contested or unusual claims on its own, the borrower who disputes the monitoring requirement, the contractor whose draw request does not match the inspection, the claim tangled up with a loan already in loss mitigation or bankruptcy, where the loss-draft process intersects other protections the borrower has. Those go to a person. The agent's authority is to run the ordinary claim fast and consistently and to keep the money moving toward the repair, and the servicer's authority is to make the judgment calls the instrument and the investor reserve for a human, because an agent that could unilaterally elect to apply proceeds to a debt, or that could resolve a contested claim at machine speed, would be a faster way to make a consequential mistake, not a safer way to help a borrower rebuild.

The Failure Mode the Desk Is Built to Prevent

The pattern that shapes how we run the desk is not a dramatic one, which is why it is dangerous. It is a borrower whose home was repaired, whose final inspection would have cleared, and whose last draw sat unreleased for weeks because one document, a signed final lien waiver, a contractor's invoice, was missing and no one on a busy desk had chased it. The money was the borrower's, the work was done, and the funds sat in the servicer's control for a reason that amounted to nobody following up. No policy said to hold the money. The process just stalled, and a stalled process that keeps a borrower from their repair funds after the work is finished is precisely the unfairness the standard describes, substantial injury the borrower cannot avoid and that serves no legitimate purpose.

The decision from cases like that is that the agent owns the chase, not just the release. It knows the one outstanding item on every open claim, it follows up before the claim ages past its stage, and it puts a claim whose funds are held after the work is verified at the top of the exception list because that is the highest-harm state a claim can be in. The servicer still makes the judgment calls, but the agent makes sure a claim never sits stalled and invisible, because the invisible stalled claim is the one that becomes the complaint and, across enough borrowers, the finding. The loss-draft desk earns its keep by getting a borrower's house repaired and their money released on time, every time, and the way an agent helps is by never letting a finished repair wait on a document nobody remembered to ask for.