Borrower Fallout: The Silent Leak in Mortgage Pipelines

Two people in the same mortgage company use the word fallout every week and mean different things by it. On the capital markets desk it means locked loans that did not fund, priced into a hedge before the market opens. In operations it means the applicant who stopped answering. Nobody reconciles the two.

Borrower fallout is the share of mortgage applications that neither fund nor get denied. Across 96,173,152 United States applications reported for 2019 through 2024, our computation from the FFIEC HMDA Data Browser puts that share between 21.5% and 23.6% in every single year, through a complete rate cycle. The band is narrower than most operators assume.

Fallout is the only large loss in a lending pipeline that one department models to the basis point and another does not measure. Secondary marketing has a fifty-year quantitative tradition. Intake has an inbox.

What does fallout mean, and to whom?

Fallout has two established meanings in mortgage lending, and they use different denominators. In secondary marketing it is the share of rate locks that never fund, measured daily against a hedge position. In lending operations it is the share of applications that never reach a decision or a closing table. Both are real.

The secondary marketing meaning is the older and the better instrumented one. The Office of the Comptroller of the Currency's Mortgage Banking booklet, version 1.0 published February 2014, describes it plainly: falling interest rates may cause borrowers to seek more favourable terms and withdraw applications before the loans close, and if customers do not close, commonly known as fallout, a bank may be unable to originate enough loans to meet its forward sales commitments.

Dean Brown, chief executive of Mortgage Capital Management, writing in The Mortgage Professional's Handbook, gives the trading-desk version: fallout is the percentage of price-protected locks that cancel, withdraw or expire, against total price-protected locks taken in a fixed period. His worked example runs 19.9%, 20.0% and 15.9% across three months, an 18.5% quarterly average, and he notes a floor, that at least 6% of loans fall out no matter what anyone does.

The hedging definition starts counting at the lock, so everything an applicant does before the lock, including giving up on a document request, sits outside the frame. The operations definition starts at the application and has no published benchmark at all.

Why aren't published pull-through numbers comparable?

Published mortgage pull-through figures range from roughly 57% to roughly 97%, and none of them contradict each other. Each uses a different denominator: all applications, locks taken, loans past underwriting, or loans already at the closing table. Comparing figures across sources produces a number that describes nothing in particular.

Four denominators, four sources, all current. Milliman's introduction to mortgage pipeline hedging, published August 2025 by Jonathan Glowacki and Nate Dorr, defines pull-through as the probability that a lock results in a funded loan, using 90% as its baseline and flexing to 95% when rates rise and 85% when they fall. Mortgage Capital Trading's pipeline hedging whitepaper, vendor research distributed through the Mortgage Bankers Association and carrying a 2022 copyright, models pull-through by stage: roughly 75% at lock, 85% past underwriting, 97% at closing. Optimal Blue, also vendor research, reported purchase pull-through of 85.7% and refinance pull-through of 69.2% for December 2025, published 13 January 2026. Our computation from the FFIEC and Consumer Financial Protection Bureau HMDA Data Browser puts originations at 57.7% of 2024 applications.

Every one of those numbers is correctly calculated. A lender who benchmarks an application-based ratio against a lock-based one will conclude the operation is failing when it is performing normally, or the reverse. Conversion adds a fifth denominator, usually leads to applications. Before a fallout number means anything in a meeting, somebody has to say what sits under the division line.

Does borrower fallout move with interest rates?

At the rate lock, yes. At the application, barely. Our computation from the FFIEC HMDA Data Browser shows non-denial fallout on United States mortgage applications holding between 21.5% and 23.6% in every year from 2019 through 2024, while lock-stage pull-through reported by Optimal Blue moved several points from one month to the next.

The period covers as violent a rate cycle as the mortgage market has produced. Freddie Mac's Primary Mortgage Market Survey recorded the 30-year fixed rate at 2.65% for the week ending 7 January 2021, the lowest in a survey running back to 1971, and at 7.09% for the week ending 17 August 2023, the highest in over twenty years. Application volume swung from 15.1 million to 23.3 million and back to 10.0 million.

How 96,173,152 United States mortgage applications ended, by year and disposition, computed from the FFIEC HMDA Data Browser for action-taken codes 1 through 5, summed across all state and territory codes and excluding records carrying no state code, accessed August 2026:

Year	Applications	Originated	Denied	Withdrawn	Closed incomplete	Approved, not accepted	Non-denial fallout
2019	15,088,520	61.9%	16.7%	14.1%	4.7%	2.6%	21.5%
2020	22,700,550	64.5%	12.5%	14.7%	5.5%	2.9%	23.0%
2021	23,349,060	64.6%	12.5%	14.2%	6.0%	2.7%	22.9%
2022	14,299,358	58.8%	17.5%	15.5%	5.2%	2.9%	23.6%
2023	10,011,117	57.0%	20.1%	14.6%	5.3%	3.0%	22.9%
2024	10,724,547	57.7%	19.3%	14.3%	5.4%	3.3%	22.9%

Originations look volatile in that table, moving 7.6 points between 2021 and 2023. Almost all of that movement is the denial rate, which tracks credit mix and the collapse of easy refinances. Take denials out of the denominator and originations sit in a 71.3% to 74.2% band across all six years.

The rate-driven fallout that secondary marketing hedges is real, and it concentrates between the lock and the closing table, where a borrower holds an option and rate news tells them when to use it. Optimal Blue's May 2026 figures, reported by National Mortgage News on 9 June 2026, had purchase pull-through at 76.7% and refinance at 65.3%, down 539 and 1,332 basis points in a single month after the 30-year conforming rate rose 13 basis points. Nothing in the application-stage data moves like that.

The stable band is the interesting half. Roughly one application in four dies without a credit answer whatever rates do, which makes the recoverable portion operational rather than macroeconomic. It cannot be hedged and it does not wait for the cycle.

Where in the pipeline do applications actually leak?

Four distinct exits, and the public record separates only three of them. Withdrawn by the applicant is the largest at 14.3% of 2024 United States applications in our HMDA computation, files closed for incompleteness account for 5.4%, and approved but not accepted for 3.3%. The fourth exit, a borrower locking with a competitor, hides inside the first.

Withdrawn is a mixed bucket and its size makes the mixing expensive. It holds the purchase that fell apart, the shopper who found 20 basis points elsewhere, and the applicant who could not face another document request. No public dataset splits the 1,529,521 withdrawals recorded for 2024.

Files closed for incompleteness have the clearest operational cause, and the same dataset shows refinances losing files that way at more than twice the purchase rate. See why conditional approvals go to die.

Approved but not accepted is the quietest exit and the most expensive per file, because everything has already been spent. The Mortgage Bankers Association reported on 15 May 2026 that total loan production expenses ran $11,898 per originated loan in the first quarter of 2026, across 324 reporting companies. A file that walks away at that point returns none of it.

What does document collection contribute to fallout?

More than the public record can prove. No dataset attributes a withdrawal to a specific document request, so the honest position is that document collection is a plausible major contributor with circumstantial evidence behind it and no measurement at all. The gap is the same one this series flagged in the abandonment curve.

The circumstantial case is decent. Refinances, which lack a closing date to force the pace, lose more files at every applicant-driven exit. Fannie Mae announced on 6 March 2024 that Desktop Underwriter would validate income, employment and assets from one 12-month asset verification report, an enhancement whose stated purpose was reducing applicant paperwork.

The measurement problem is structural. Both of the richest borrower-experience datasets in United States mortgage lending sample only completed loans. The National Survey of Mortgage Originations, run jointly by the Federal Housing Finance Agency and the Consumer Financial Protection Bureau, drew 62,359 first-lien mortgages originated 2013 through 2024 in a public use file updated 27 May 2026. J.D. Power's 2025 United States Mortgage Origination Satisfaction Study, published 12 November 2025, surveyed 10,067 customers who had originated or refinanced. Everyone who fell out sits outside both frames by construction.

Anyone who wants the number instruments their own funnel, the argument made per step in the abandonment curve and per channel in why emailed documents lose files.

What does recovery design look like?

Recovery design means instrumenting the four exits separately, then treating each as a queue with an owner rather than as attrition. The cheapest available gain is knowing which of the four exits a lost file went through, a question most origination reporting cannot answer today.

1. Record a reason code at every exit, chosen by whoever closes the file, and keep the codes to five or fewer so they get used honestly.
2. Split the withdrawn bucket into competitor loss, transaction collapse and applicant abandonment. Those three share a label and nothing else.
3. Measure time since the last applicant action, not time in stage. A file waiting on the lender and a file waiting on the borrower age identically on a dashboard and mean opposite things.
4. Track pull-through on two denominators at once, applications and locks, and publish both internally with their definitions attached.
5. Give the applicant a resumable place to return to. An emailed attachment chain has no address, and a borrower who lost the email is a lost file.
6. Treat approved-but-not-accepted as its own report line with a named owner.

Rate shopping earns a line of its own because lenders overestimate it. The Consumer Financial Protection Bureau's January 2015 report on consumers' mortgage shopping experience, covering 1,922 purchase borrowers surveyed in early 2014, found roughly 77% applied to only one lender. Freddie Mac research published 16 February 2023 found rate dispersion more than doubled in 2022, so shopping pays better than it did, and it stays a minority behaviour. Most files that leave are not being taken. They are being lost.

What does Canada publish on mortgage fallout?

Almost nothing. Canada has no public equivalent of the United States application-level disclosure dataset, so no Canadian fallout, withdrawal or incompleteness figure exists to cite for mortgage lending, at any level of aggregation, from any federal source located for this article.

The closest published number sits in Canada Mortgage and Housing Corporation's Residential Mortgage Industry Report, Spring 2026 edition, which states that the approval rate for same-lender refinance applications fell from a peak of 98% in the second quarter of 2021 to 79% in the fourth quarter of 2025, after a 70% reading in the third quarter that was the lowest since the data began in 2016. That is an approval rate. It says nothing about what happened to approved files afterward.

Bank of Canada work published in January 2025 by Odae Al Aboud, Saarah Sheikh, Adam Su and Yang Xu describes the enhanced RESL2 dataset from the Office of the Superintendent of Financial Institutions, roughly six million mortgages worth $1.7 trillion as of September 2024. That dataset covers the stock of outstanding loans. Applications that never became loans are not in it.

What we couldn't verify

No published application-to-funding fallout benchmark exists for any market outside the United States disclosure data used here. Figures circulating for mortgage pull-through are almost entirely lock-based, vendor-published, or both, and where the sample and the denominator are both undisclosed they are not cited in this article.

Nothing published anywhere splits the withdrawn category by cause. Our HMDA computation counts 1,529,521 United States applications withdrawn by applicants in 2024 and cannot say how many of those borrowers went to a competitor, lost a house, or simply stopped.

No public dataset carries the elapsed time between application and each exit, so the claim that ageing predicts fallout rests on operator experience rather than published evidence. The HMDA record carries a year and an outcome.

The 6% fallout floor from Dean Brown's handbook chapter is a practitioner estimate with no published derivation, and it describes locks rather than applications. It appears here because it is the only floor anyone has put in print. The same-lender refinance approval series in the Canada Mortgage and Housing Corporation report does not name its underlying data source.

Common questions

What is borrower fallout in mortgage lending?
Borrower fallout is the share of mortgage applications that neither fund nor get denied. It covers applications withdrawn by the applicant, files closed for incompleteness and approvals the borrower never accepted. In secondary marketing the same word means locked loans that never fund, a narrower and later measurement.

What is a normal mortgage pull-through rate?
There is no single answer, because the denominators differ. Optimal Blue vendor research reported 85.7% purchase pull-through on locks for December 2025, Milliman's August 2025 hedging paper uses a 90% baseline on locks, and originations were 57.7% of all 2024 applications in the FFIEC HMDA data.

Does fallout rise when interest rates fall?
At the rate lock, yes, because a borrower holds an option and cheaper offers appear. At the application level the effect is small. Our computation from the FFIEC HMDA Data Browser shows non-denial fallout between 21.5% and 23.6% in every year from 2019 through 2024, across a full rate cycle.

How much does a lost mortgage application cost a lender?
No published figure isolates it. The Mortgage Bankers Association reported on 15 May 2026 that total loan production expenses ran $11,898 per originated loan in the first quarter of 2026 across 324 companies, a cost lenders recover only on files that fund.

Does Canada publish mortgage fallout or pull-through data?
No federal Canadian source publishes application-level dispositions. Canada Mortgage and Housing Corporation's Residential Mortgage Industry Report, Spring 2026 edition, reports a same-lender refinance approval rate of 79% for the fourth quarter of 2025, which measures approvals rather than what happened to files after approval.

---

Carousel builds the intake and verification layer that makes each of those four exits a measured event rather than a missing file. See how verification fits your flow