The Stips Problem: Why Conditional Approvals Go to Die
Conditional approvals die between yes and funding. What US mortgage data shows about the loan stipulations process, and what nobody publishes.
Alfred BEditorial Reviews
A conditional approval is a promise with homework attached. Someone still has to produce the missing month of bank statements, the void cheque that matches the account on the application, the tax assessment that agrees with the income already keyed into the file. The credit work is finished. The money cannot move.
A loan stipulations process is the workflow that collects and clears the outstanding items a lender requires before an approved file can fund. No published figure measures how often those items get satisfied. United States mortgage disclosure data is the closest public proxy that exists, and in 2024 it recorded 5.4% of applications ending as files closed for incompleteness.
That gap between yes and money is the least instrumented part of lending. Origination software counts approvals. Servicing counts fundings. The stretch in between usually lives in an inbox, and inboxes do not report.
Where do stipulations come from?
Stipulations arrive from three directions. Some are gaps between what the applicant supplied and what credit policy requires. Some are conditions the credit decision itself attaches, such as a personal guarantee or a paydown. Some are items that were valid when the file was approved and had gone stale by the time it was ready to fund.
The third category is generated by the lender's own calendar rather than by anything the applicant did wrong. Freddie Mac's announcement of 26 May 2022 describes automating a ten-day pre-closing verification of employment, a stipulation that exists purely because time passed between approval and closing.
In private credit I watched a stips queue build every Friday afternoon and clear Monday at ten. Nothing in that queue was hard. Most of it was one document, one signature, one clarifying sentence from a broker. The files did not die of complexity. They died of nobody being on the other end of the request.
What does the public record show about files that die after approval?
The Home Mortgage Disclosure Act dataset is the only large public record in the United States of what happened to individual credit applications. Across all US states and territories in 2024, lenders reported 10.72 million applications, and 22.9% of them ended without funding for a reason other than a denial. That is more than the 19.3% denied outright.
The 22.9% covers three of the dataset's labels: application withdrawn by applicant, file closed for incompleteness, and application approved but not accepted. We computed the shares from the FFIEC and Consumer Financial Protection Bureau HMDA Data Browser, covering action-taken codes 1 through 5 for 2024, accessed August 2026, summed across state codes and excluding the records that carry no state code.
The Consumer Financial Protection Bureau's Data Point on 2022 mortgage market activity, published September 2023, said the same thing in its own words, reporting around 11.0 million closed-end site-built single-family applications including 2.5 million that lenders closed as incomplete or the applicant withdrew before a decision.
How 10.72 million US mortgage applications ended in 2024, by disposition and loan purpose, computed from the FFIEC HMDA Data Browser:
| Disposition | All applications | Home purchase | Refinance |
|---|---|---|---|
| Loan originated | 57.7% | 67.1% | 50.7% |
| Application denied | 19.3% | 11.3% | 21.4% |
| Withdrawn by applicant | 14.3% | 15.3% | 17.0% |
| File closed for incompleteness | 5.4% | 3.5% | 7.5% |
| Approved but not accepted | 3.3% | 2.9% | 3.4% |
| Applications counted | 10,724,547 | 5,264,046 | 2,894,727 |
Purchase and refinance do not sum to the total because home improvement and other-purpose applications sit outside both columns.
Why do refinances lose more files to incompleteness than purchases?
Refinances end as closed for incompleteness at more than twice the rate of home purchases. The 2024 HMDA data puts home purchase at 3.5% and refinance at 7.5%, with cash-out and rate-and-term refinances sitting close together. Withdrawals move the same direction, 15.3% against 17.0%.
A purchase has a closing date, a seller, a deposit and a moving truck, and everyone involved is pushed by a calendar nobody controls. A refinance has none of that. The applicant can put the request down for a week and lose nothing they can feel.
Stipulation fulfilment looks mostly like a function of external pacing, and lenders inherit whichever pace their product comes with. Most refinance files that die incomplete are files where the lender's own follow-up was the only clock in the room, and it was not enough.
Does the same pattern show up in business lending?
Business lending has no equivalent public record, so the answer is partial. The Federal Reserve's Small Business Credit Survey, released 3 March 2026 and covering 6,525 firms surveyed between September and November 2025, reports that 42% of applicants received the full amount of financing they sought, 36% received some or most, and 22% received none.
Those are credit outcomes rather than fulfilment outcomes. Nothing in the survey separates a partial approval from a full approval that never funded because the borrower stopped answering.
Canada's closest measure has the same limitation. Statistics Canada reported on 20 February 2025, from the 2023 Survey on Financing and Growth of Small and Medium Enterprises, that 88.2% of SMEs had their largest debt financing request fully or partially approved. That is an approval statistic, silent on everything between the approval and the deposit.
What does a working loan stipulations process look like?
A loan stipulations process works when each condition is a tracked object with an owner, an acceptance rule, a due date and a status, rather than a line in an email thread. The design question is not whether the list exists. It is whether anything in the system notices when the list stops moving.
Four properties separate a tracked stipulation from an email thread.
- Each condition is a discrete record with its own state, so "three of seven cleared" is a readable fact rather than something reconstructed by scrolling.
- Acceptance is defined before the request goes out. A stip reading "bank statements" that comes back as a screenshot generates a second round trip that was avoidable.
- The request has a home the applicant can return to. A resumable link beats an attachment chain, because the applicant who lost the email is otherwise a lost file.
- Ageing is visible and owned. Time since last applicant action is the most diagnostic number here, and it is almost never on a dashboard.
None of that is exotic. It is ordinary queue discipline, applied to a queue most lenders never modelled as one.
Can stipulations be pre-empted at intake?
Some can, and the US secondary market has spent a decade proving it. Fannie Mae's Day 1 Certainty offers representation and warranty relief on income, employment and asset data validated through its Desktop Underwriter validation service, which moves the verification from a post-approval document request to a check performed at application.
Freddie Mac said on 26 May 2022 that adopters of its automated offerings, including the Asset and Income Modeler, can shorten cycle times by as much as 15 days, and on 15 May 2025 that lenders maximizing automation originate loans $1,500, or 14%, cheaper. Both figures come from the sponsor of the program.
Independent work points the same way. Federal Reserve Bank of New York Staff Report 836, published February 2018 by Andreas Fuster, Matthew Plosser, Philipp Schnabl and James Vickery, found technology-led lenders reduced processing time by about ten days, roughly 20%, with refinances moving 9.3 to 14.6 days faster, and reported that default rates did not rise with the speed.
The stipulations that survive automation need a human decision or a third party's signature, and those are worth designing a queue around. The ones that are only a document somebody already holds on the applicant's behalf are a collection problem, solved in one corner of the market and adopted unevenly across the rest.
What we couldn't verify
No published stipulation fulfilment rate exists. We searched regulators, central banks and statistical agencies in the United States, Canada and the United Kingdom and found no dataset or paper reporting how many conditions get satisfied, in what time, or how many approved files fail for want of one.
Figures circulating on this point trace back to vendor materials with undisclosed samples, and none are used here. No public dataset carries the time between approval and funding. HMDA records the year and the outcome, not the dates in between, so the 5.4% incompleteness share tells you a file died and not how long it took.
Canada publishes no application-disposition data. Section G of OSFI's E2 Mortgage Loans Report collects application and approval counts quarterly from deposit-taking institutions, and we located no public release of those figures. CMHC's Residential Mortgage Industry Data Dashboard publishes originations, amortizations and arrears with no application outcomes at all.
Withdrawn by applicant is a mixed bucket. It holds rate shoppers who went elsewhere, buyers whose purchase fell through, and applicants who gave up on a stips list, and the HMDA record does not separate them. The 14.3% is an upper bound, not a measurement.
Common questions
What is a loan stipulation?
A loan stipulation is an outstanding item a lender requires before an approved file can be funded, such as an updated bank statement, a signed guarantee, proof of insurance or a corrected form. Stipulations sit between the credit decision and the disbursement, and they are usually tracked separately from the underwriting itself.
What percentage of conditional approvals never fund?
No published figure exists. The closest public proxy is US mortgage disclosure data, where 5.4% of 2024 applications ended as files closed for incompleteness and 14.3% were withdrawn by the applicant, per the FFIEC HMDA Data Browser. Neither category isolates stipulation failure specifically.
Why do lenders ask for documents again after approving a loan?
Some conditions come from the credit decision itself, and some are re-verifications triggered by elapsed time. Freddie Mac's May 2022 announcement describes automating a ten-day pre-closing verification of employment, which exists because the gap between approval and closing can outrun the freshness of the original check.
Does Canada publish loan application outcome data like HMDA?
No equivalent public dataset was located. OSFI's E2 Mortgage Loans Report collects quarterly application and approval counts from deposit-taking institutions under Section G, with no public release found, and CMHC's Residential Mortgage Industry Data Dashboard reports originations and arrears rather than application outcomes.
Can verification at intake reduce stipulations?
Partly. Fannie Mae's Day 1 Certainty provides representation and warranty relief on income, employment and asset data validated at application, and Freddie Mac reported in May 2022 that automated adopters can shorten cycle times by up to 15 days. Conditions requiring a human decision or a third party remain.
Carousel builds the intake and verification layer that turns a stips list into a tracked queue instead of an email thread. See how verification fits your flow
