The 24-Hour Rule: Application Momentum and the Physics of Drop-Off
Completion probability decays every hour an application sits. What three separate research literatures say about momentum, and what transfers to lending.
Alfred BEditorial Reviews
A credit application is not really a document. It is a state of attention, held by one person, for a limited time, and it starts leaking the moment it is created.
Funnel reporting hides that. A file that funds on day nine and a file that funds in eleven minutes land in the same bucket at month end, so the number on the board says pull-through while saying nothing about what the applicant went through.
Application momentum is an applicant's willingness to keep going, and it decays with elapsed time rather than with effort spent. Drop-off is front-loaded inside a session and delay-driven between sessions. Harvard Business Review research from March 2011 found that contact within an hour was nearly seven times as likely to qualify a lead as contact an hour later.
The published evidence for this sits in three separate literatures, none of them about lending, plus one regulator dataset that is.
What does "application momentum" actually mean?
Application momentum is the probability that an applicant who has started will finish, measured as a function of elapsed time rather than of steps completed. It behaves less like a balance and more like a half-life.
Shortening a form by three fields is a step-count improvement. Answering in four minutes instead of four hours is a time improvement, and the time improvement is the one the evidence keeps rewarding.
Three clocks run at once in any application: seconds inside a screen, hours between an applicant's action and yours, days from start to a decision. Each has its own research and its own failure mode.
Why is drop-off worst at the very beginning?
The seconds-scale evidence is the oldest and the most counterintuitive. Nielsen Norman Group published an analysis by Jakob Nielsen in September 2011 of research by Chao Liu, Ryen White and Susan Dumais covering 205,873 web pages and more than two billion recorded dwell times, and it found that 99% of pages show a negative aging effect. In plain terms, the risk of leaving is highest at the start and falls the longer someone stays. Nielsen put the average visit at a little under a minute and identified the first ten seconds as the window where the decision to abandon is mostly made.
That shape is the useful part. Applicants who survive the opening moments become progressively more likely to finish, which means the first screen carries a wildly disproportionate share of total loss.
Load time compounds it. Think with Google published research by Daniel An in February 2018, drawn from Google and SOASTA analysis conducted in 2017, finding that the probability of a mobile visitor bouncing rises 123% as page load time goes from one second to ten seconds.
None of this was measured on credit applications. The curve shape is what transfers, not the numbers.
What happens once a file sits for an hour?
Here is the study the whole 24-hour idea traces back to. Harvard Business Review published research in March 2011 by James Oldroyd, Kristina McElheran and David Elkington analyzing 1.25 million sales leads received by 29 business-to-consumer and 13 business-to-business companies in the United States.
Firms attempting contact within an hour of an enquiry were nearly seven times as likely to qualify that lead as firms attempting contact an hour later, and more than 60 times as likely as firms that waited 24 hours or longer. Qualifying meant reaching a meaningful conversation with a decision maker.
The behavioural part of that paper gets quoted less and deserves more attention. Among companies that responded at all within 30 days, the average response time was 42 hours. Only 37% responded inside the first hour, 24% took longer than a day, and 23% never responded at all.
The decay curve is steep and the median operator sits on the wrong side of it.
One honest caveat, stated up front rather than buried: this research measures the seller's clock, not the applicant's. It tells you what happens when a firm delays. It does not directly measure what happens when an applicant pauses a half-finished file and the flow waits patiently for them.
Does the day-scale evidence agree?
It does, from a very different method. Federal Reserve Bank of New York Staff Report 836, published in February 2018 by Andreas Fuster, Matthew Plosser, Philipp Schnabl and James Vickery, used market-wide loan-level data on US mortgage applications and originations. Technology-led lenders processed applications about 20% faster after controlling for loan, borrower and geographic characteristics, with purchase mortgages moving 7.9 to 9.2 days faster and refinances 9.3 to 14.6 days faster.
The authors state plainly that faster processing did not come at the cost of higher defaults. That clause is what makes the finding usable in a risk committee.
The evidence for time-based decay, by timescale and source:
| Timescale | What the published evidence shows | Source |
|---|---|---|
| Seconds | 99% of pages show a falling exit risk the longer a visitor stays; the first 10 seconds carry the highest risk | Nielsen Norman Group, Sept 2011 |
| Seconds | Bounce probability rises 123% as mobile load time goes from 1 second to 10 | Think with Google, Feb 2018 |
| Hours | Contact within an hour was ~7x more likely to qualify a lead than an hour later | Harvard Business Review, Mar 2011 |
| Hours | More than 60x more likely to qualify than a wait of 24 hours or longer | Harvard Business Review, Mar 2011 |
| Hours | Average response was 42 hours; 23% of firms never responded | Harvard Business Review, Mar 2011 |
| Days | Technology-led mortgage lenders processed ~20% faster, without higher defaults | NY Fed Staff Report 836, Feb 2018 |
| Outcome | ~2.3 million of 12.1 million 2017 mortgage applications closed incomplete or withdrawn before a decision | CFPB Data Point, May 2018 |
What does the regulator record show?
One dataset in US lending counts unfinished files directly. The Consumer Financial Protection Bureau's Data Point on 2017 mortgage market activity, published in May 2018, reported about 12.1 million home mortgage applications from 5,852 institutions, and stated that approximately 2.3 million of those were closed by the lender as incomplete or withdrawn by the applicant before a decision was made. On the report's own figures that is roughly one application in five.
The Bureau's December 2024 report on 2023 activity counted about 10 million applications and 5.7 million originations, with denial rates of 9.4% on home purchase and 32.7% on refinance. It groups withdrawn and incomplete files inside the application total without breaking them out.
What we don't know
No credible published abandonment rate exists for loan applications. Percentages circulate widely, and the ones we traced ran back to vendor marketing or trade articles recycling unattributed figures rather than to a regulator, a statistical agency or a peer-reviewed source. A number with no methodology behind it is worse than no number, because it gets budgeted against.
The mortgage figures above are the closest published proxy, and they are not the same thing. A file recorded as withdrawn or incomplete under the Home Mortgage Disclosure Act had already become a formal application. Everything that died before that point is invisible in the data, and the record carries no timestamps for how long the file sat before it died.
There is also nothing Canadian. Statistics Canada released results from the Survey on Financing and Growth of Small and Medium Enterprises on 20 February 2025 reporting that 88.2% of SMEs had their largest debt financing request fully or partially approved in 2023, with no measure of time to decision anywhere in it.
My own view, which a cautious committee would soften: the abandonment rate is unknowable at the industry level and mostly uninteresting anyway. The number worth owning is your own median hours from last applicant action to next lender action, cut by hour of day, because that is the metric that moves when someone changes something.
How do you defend momentum by design?
Four moves, in rough order of payoff.
- Put the highest-drop-off decision first and make it survivable. If a bank connection or an identity check is going to lose people, losing them in minute one costs less than losing them in minute nine.
- Measure the pause, not just the completion. Time between the applicant's last event and the file's next event is the single most diagnostic number in intake, and almost nobody reports it.
- Never let a file need a human to advance during hours when no human is there. In private credit I watched a stips queue build every Friday afternoon and clear Monday at ten, so a Friday applicant lost 66 hours to a calendar.
- Treat re-entry as a first-class path. Resumable state, one link, no re-authentication marathon.
Momentum is an asset with a carrying cost, and the cost is paid in hours whether or not anyone is counting them.
Common questions
What is application momentum?
Application momentum is the probability that someone who has started an application will finish it, expressed as a function of elapsed time rather than steps completed. It decays while a file waits, which is why intake speed and response latency are treated as design constraints rather than as operational nice-to-haves.
Where does the 24-hour rule come from?
It traces to Harvard Business Review research published in March 2011 covering 1.25 million sales leads at 29 business-to-consumer and 13 business-to-business US firms. Firms contacting within an hour were more than 60 times as likely to qualify a lead as those waiting 24 hours or longer.
What percentage of loan applications are abandoned?
No credible published figure exists. The Consumer Financial Protection Bureau reported that about 2.3 million of 12.1 million 2017 US mortgage applications closed incomplete or withdrawn before a decision, but that counts formal applications only and carries no timing data at all.
Is faster underwriting riskier?
Federal Reserve Bank of New York Staff Report 836, published February 2018, found technology-led mortgage lenders processed applications about 20% faster than other lenders and states that faster processing did not come at the cost of higher defaults. Speed and credit quality moved independently in that sample.
Which single metric best tracks drop-off?
Median elapsed hours between the applicant's last action and the lender's next action, split by stage and by hour of day. Averages hide the problem because one stalled file distorts them. Median exposes the queues, the handoffs and the overnight gaps where momentum quietly dies.
Carousel's intake flows are built to hold momentum from the first field to a decidable file. See how verification fits your flow
