Resume-ability: The Most Underrated Feature in Intake Design

An abandonment report records where an applicant stopped. It does not record why, so the reader fills that part in, usually with a decision the applicant never made. Dropped at step six. Must have gone somewhere else.

Sometimes. Often the phone rang, on the same device the application was open in. Or the bus arrived, or a customer walked up to the counter, or the void cheque turned out to be in a drawer at home. Nothing was decided.

Resumable applications save progress automatically and return an applicant to the step they left, on any device, without retyping anything. The case for building them rests on a distinction most funnel reporting cannot make: a stopped session is not the same as a declined offer. One is lost. The other is waiting.

Is abandonment refusal or interruption?

Nobody has published a measurement that splits abandoned credit applications into those two buckets. What follows is an argument built from adjacent evidence, and not a finding.

The closest public data comes from retail. Baymard Institute's cart abandonment research, last updated 22 September 2025 and compiled from 50 separate studies, puts average cart abandonment at 70.22% and finds that 42% of abandoners were browsing or not ready to buy at all. Among the shoppers who were ready to buy, the leading reasons are mechanical rather than financial: 18% left because the site required account creation, and 17% because checkout was too long or too complicated. Neither describes a person evaluating a product and deciding against it.

Financial flows have their own version of the same pattern. Open Banking Limited's options paper of 18 November 2019, written on the cost of the 90-day re-authentication rule, reported roughly 50% average completion across all payment service user authentications, with more complex multi-step journeys close to 0% conversion. Step count predicted the outcome.

The inference from there is still an inference. A refusal has a shape. It clusters at the screen where the applicant learns something they dislike, usually rate, term, fee, or a document they will not hand over. Interruption spreads thinly across every step in proportion to its length, and rises with step count rather than with any one step's content. Any lender can test which pattern their book shows in an afternoon, plotting exits per step against median time on step. If exits track duration rather than content, the population leaving is interrupted, and recoverable.

What does the interruption research actually show?

Gloria Mark, Victor González and Justin Harris, in a CHI 2005 study of 24 information workers, found that 57.1% of observed work segments were interrupted, and that people took an average of 25 minutes and 26 seconds to get back to an interrupted task, having passed through 2.26 other tasks along the way. The figure that matters most for intake design is a different one from the same study. 77.2% of interrupted work was resumed on the same day.

People come back. Later than anyone designing a session timeout imagines, and after doing several unrelated things first, but the large majority come back within hours. A session that expires after twenty minutes of inactivity is closing the door at almost precisely the average moment the applicant would have walked back through it.

Mark returned to the question in a CHI 2008 controlled experiment with Daniela Gudith and Ulrich Klocke, running 48 participants through interrupted and uninterrupted conditions. Interrupted participants finished the work faster, at the cost of measurably higher stress, frustration, time pressure and effort. Interruption does not remove intent. It taxes it. A flow that then demands twenty minutes of retyping charges the applicant twice for something they did not choose.

Both studies observe office work, not credit applications, and neither was built to explain form abandonment. What they describe is how people return to interrupted tasks, which is the mechanic an intake flow either supports or breaks.

Which applicants get interrupted most?

StatCounter's global stats recorded mobile at 52.57% of worldwide page views in July 2026, against 45.93% for desktop and 1.5% for tablet. Most applications now begin on the device that is also the biggest single source of interruption in the applicant's day.

Martin Pielot, Karen Church and Rodrigo de Oliveira, in a 2014 MobileHCI study, logged a mean of 65.3 notifications per participant per day and reported that messaging and email interrupted whatever the participant was doing in almost half of cases. Fifteen people took part, so treat it as directional.

The distribution matters more than the average. Pew Research Center, publishing on 8 January 2026 from a survey of 5,022 US adults fielded between February and June 2025, found that 16% of adults are smartphone dependent, meaning they own a smartphone but have no home broadband subscription. Among adults in households under $30,000 a year, that rises to 34%.

Set that against a lending book. The applicants least likely to have twenty uninterrupted minutes at a desktop are disproportionately the near-prime and subprime borrowers the growth plan is counting on. An intake flow that only really works in one sitting is means-testing its own funnel, and doing it in the direction nobody intended.

Where do applications actually stall?

Five stall points show up in almost every lending intake flow. Each needs a different recovery mechanic.

Stall point	What the applicant is doing	Recovery mechanic
Missing artifact	Void cheque, policy number, prior-year tax slip sitting at home	Server-side state plus upload from any device later
Co-applicant pause	Waiting on a spouse or business partner	Section-level invite so each party completes independently
Device switch	Started on a phone, needs a laptop and a scanner	State keyed to the applicant record, not the browser session
Ambient interruption	Call, customer, child, stop announcement	Field-level autosave and a resume link within the hour
Deliberation	Thinking about a large commitment	Long-lived link with a reminder cadence measured in days

What does the implementation sequence look like?

1. Persist server side, on every field, before an account exists. Local storage dies with the browser and does not survive a device switch. The applicant's mental model is that nothing typed can be lost; the job is to make that model accurate.

1. Capture a recovery handle in the first two screens. Email or mobile number, asked once, with the reason stated on screen. Everything downstream depends on reaching a part-completed record through something other than a cookie.

1. Show remaining work as a list rather than a bar. The GOV.UK Design System's "complete multiple tasks" pattern is written for exactly this case: longer transactions completed over several sessions, with the task list shown again at the start of every returning session.

1. Send the resume link at the moment of the stall, not the next morning. Kate Kaplan of Nielsen Norman Group, writing on 5 September 2021 about designing for long waits and interruptions, recommends showing progress completed alongside steps and time remaining, and giving fast access to recently used items. In intake, the return route is that item.

1. Name what is left, never what is done. "Two steps, about three minutes" survives being read on a lock screen. "68% complete" does not tell the applicant anything actionable.

1. Set session and link lifetimes against a published standard. NIST Special Publication 800-63B-4, published 31 July 2025, sets overall session limits of no more than 30 days at AAL1, no more than 24 hours with a one-hour inactivity limit at AAL2, and no more than 12 hours with a 15-minute inactivity limit at AAL3. A part-completed application holding a name, an address and a stated income is not AAL3 material.

1. Step up at the sensitive boundary rather than at the door. A resume link is an access credential. Returning an applicant to their own unverified answers carries one level of risk; showing a bank connection status, a decision, or an assembled document set carries another. Re-verify where the exposure changes.

1. Treat sign-out and the closed tab as the same event. GOV.UK One Login's design recommendations place a save-and-complete-later link under the primary action on every page of a service form, warn before sign-out that unsaved progress will be lost, and send a confirmation email carrying the link back in.

1. Instrument resumption on its own. Resume rate, time-to-resume distribution, and completion rate of resumers measured against first-pass completers. Aggregate funnel numbers hide all three, which is part of why the feature keeps losing roadmap arguments.

Harry Trimble and Rob Le Quesne, publishing the Government Digital Service's task list pattern on 4 April 2017 after five rounds of testing with 34 users, noted that one agency attributed over 60% of its user support volume to queries about the application process itself. Resume design and support cost are the same problem seen from two desks.

What we don't know

The central claim here has not been measured. No public dataset splits abandoned credit applications into interruption and refusal, and none of the sources above was built to make that split. The argument runs on retail checkout surveys, open banking authentication data, and workplace interruption studies, none of which is about a loan application on a phone at a bus stop.

There is also no credible published benchmark for lending application abandonment. Numbers circulate freely. The ones traceable to a method are samples of a single provider's own customers, which measure that provider's flows rather than the market.

And there is no controlled experiment from a regulator or an academic body on resume link timing, expiry, or copy in a financial application. The sequence above rests on standards guidance, interruption research from other settings, and the arithmetic of step count. Anyone running a large intake book with real experimentation capacity could settle three or four of these questions in a quarter.

Common questions

What is a resumable application?
A resumable application saves an applicant's progress automatically as they type and lets them return to the exact step they left, from any device, without retyping. The state lives on the server and is reachable through a link, so it survives a closed tab, an expired session, or a switch from phone to laptop.

How much abandonment is interruption rather than refusal?
No public dataset measures this for lending. Baymard Institute's cart abandonment research, updated 22 September 2025, found 42% of retail abandoners were browsing rather than ready to buy, and Open Banking Limited reported in 2019 that multi-step authentication journeys converted close to zero. Both point at friction, neither settles it.

How long does a resume link stay valid?
NIST Special Publication 800-63B-4, published 31 July 2025, sets overall session limits of 30 days at AAL1, 24 hours at AAL2 with a one-hour inactivity limit, and 12 hours at AAL3 with 15 minutes of inactivity. Match the tier to the data a part-completed record actually holds.

Do interrupted people actually come back?
In a CHI 2005 study of 24 information workers, Gloria Mark, Victor González and Justin Harris found 77.2% of interrupted work was resumed the same day, after an average gap of 25 minutes and 26 seconds. That is office work rather than credit applications, but the return behaviour is the relevant mechanic.

Which device do most applicants start on?
StatCounter recorded mobile at 52.57% of worldwide page views in July 2026, against 45.93% for desktop. Pew Research Center reported on 8 January 2026 that 16% of US adults are smartphone dependent, rising to 34% below $30,000 of household income. Cross-device resumption is not an edge case.

---

Carousel's intake flows save progress by default, so an interrupted applicant returns to where they stopped. See how verification fits your flow