The Abandonment Curve: Where Applications Die, Step by Step

Aggregate completion rate is the most quoted number in intake analytics and the least useful one. "We complete at 58%" establishes that a problem exists and says nothing about where it lives.

The information is in the curve. Completion plotted step by step attributes a loss to a screen rather than to a mood, and the loss is nowhere near evenly spread.

An application abandonment curve is the completion rate of an application measured per step rather than end to end. The published evidence shows loss is not uniform. Identity verification, bank connection, consent and document upload each fail at different rates and for different reasons. No credible per-step benchmark exists for lending, so the curve has to be measured in-house.

What is an application abandonment curve?

An application abandonment curve plots entries, completions and exits for every screen, so each step carries its own conversion number. A lender who knows the application completes at 58% has one fact. A lender who knows the identity step converts at 71% and the document step at 44% has a work order.

The distinction matters because the fixes are not interchangeable. A step losing people to effort responds to field reduction. A step losing people to fear responds to a sentence of copy. A step losing people to availability, where the document sits at home or behind another login, responds only to removing the document. Read the mechanism wrong and you buy months of work on the wrong screen.

Which steps have real published evidence behind them?

Several step types have been measured by named public bodies. None of the measurements come from a lending funnel, a real limit rather than a technicality, so read each for mechanism before number.

What the public record contains for each step type in a financial application:

Step type	Best available public measurement	Source and date	What it does not tell you
Identity verification	48% success rate against a 90% projection	UK National Audit Office, 5 March 2019	The measure excluded pre-provider drop-outs
Bank connection	Roughly 50% authentication completion; multi-step journeys close to 0%	Open Banking Limited, November 2019	Current rates; predates a 2021 rule change
Consent comprehension	45.2% do not read terms; 77% did not feel informed	Financial Services Consumer Panel, March 2018	Whether that causes exit here or later
Forced account creation	18% of shoppers abandoned over required account creation	Baymard Institute	Whether the size holds where enrolment is normal
Flow length	17% abandoned over a checkout that ran too long	Baymard Institute	Where in the flow length became intolerable
Document upload	No published measurement	None located	Everything

Why is the identity step the best-documented one?

The UK National Audit Office's investigation into GOV.UK Verify, published 5 March 2019, is the clearest public measurement of an identity step at scale. It reported a verification success rate of 48% at the beginning of February 2019, against a 2015 projection of 90%.

The more useful part of that report is a definitional footnote. The National Audit Office noted in March 2019 that the success rate did not count applicants who dropped out before choosing a provider, for example because they lacked the required documentation. That 48% measured people who had already committed, so true step conversion was lower by an amount nobody published.

Most internally reported identity-step numbers carry the same blind spot. It is the commonest reason a dashboard flatters a flow that is losing people badly.

Two further figures from the same March 2019 report give texture. Of the 70% of Universal Credit claimants who attempted to sign up through Verify, only 38% could verify their identity online. At DVLA services, 8.3% of users gave information that did not match DVLA-held data and could not access the service. That second number is the shape of a data-matching failure: not refusal, not effort, just a mismatch between a life and a reference file.

What does the bank-link step lose?

Open Banking Limited's November 2019 options paper on the 90-day re-authentication requirement is the most detailed public measurement of bank-connection friction available. It recorded roughly 50% average completion across all customer authentications, with complex multi-step journeys close to 0% conversion.

That zero is the finding worth carrying into any flow design. Step count at the connection screen does not degrade conversion gradually. Past a point it removes it entirely.

The same November 2019 paper found roughly 13% of active customers dropping off every 90 days under that rule, and around 30% of connections broken as a result. The Financial Conduct Authority removed the requirement in November 2021, so read the figures as evidence of what connection friction costs, not as a live benchmark.

The Open Banking Implementation Entity reported on 4 August 2020 a direct correlation between making app-based authentication available and higher consent success rates. Where an applicant authenticates inside their banking app instead of typing credentials into a browser, more of them finish.

There is also a ceiling no design work removes. BIS Papers No 168, published by the Bank for International Settlements on 30 March 2026, found that when Brazilian consumers were shown benefits including improved loan terms and higher credit limits, only 27% would share data to obtain them. In India, the same paper noted, only 10% reported using the account aggregator mechanism. Some applicants decline regardless of the screen.

Where does the consent step actually fail?

The consent step usually fails without producing an exit, which is why it rarely shows on a curve at all.

Research for the Financial Services Consumer Panel, conducted by Dr Edgar Whitley and Dr Roser Pujadas of the London School of Economics and published in March 2018, surveyed 191 participants and found 45.2% do not read terms and conditions at all, 40.9% only skim-read them, and 77% did not feel informed. Among the non-readers, 41.9% said the text was too long.

So the consent screen converts, and it converts by people agreeing to something they have not read. That cost surfaces later as a revoked connection or a complaint, neither of which appears in step-level conversion.

Is the document step the biggest drop?

Almost certainly, and the public record does not prove it.

The closest openable evidence is adjacent. The Baymard Institute's checkout research found 18% of US online shoppers abandoned an order because the site wanted them to create an account, and 17% because checkout ran too long or too complicated. Baymard also found the average US checkout shows 23.48 form elements by default, against an ideal of 12 to 14.

Those numbers describe typing and enrolment friction. A paystub request is a different category of ask. It moves the applicant out of the flow, into an email archive or a filing cabinet or another login, and the return trip is optional.

What we don't know

This is the honest column, and it runs longer than the evidence column.

There is no credible published abandonment rate for loan applications, and no published per-step lending funnel benchmark. Figures circulating on both come overwhelmingly from vendor materials whose sample, definition of an "application start" and measurement window are undisclosed, which makes them unusable as evidence even where they are accurate. None are cited here.

Nothing above is Canadian. Every step-level figure comes from the UK, Brazil, India or US e-commerce.

Nothing above is from a lender either. Government identity verification and online checkout are the closest measured analogues to a credit application, and those populations differ in motivation, urgency and consequence of failure. Treat the mechanisms as transferable, the magnitudes as not.

The document step, which practitioner experience suggests is the largest single loss in most financial funnels, has no public measurement at all. A striking gap, for the step most people would name first.

How do you plot your own curve?

1. Instrument entries, completions, exits and time-in-step per screen, not per session.
2. Count the pre-step drop-out the National Audit Office flagged in March 2019, applicants who leave before a step formally begins. Excluding them gives a step credit it has not earned.
3. Split mobile and desktop. They diverge hardest at typing and document steps.
4. Classify the largest drop by mechanism before touching it: effort, fear, availability, mismatch, or waiting on someone else.
5. Track document-fallback usage as a rate, not a failure, given the 27% willingness ceiling the Bank for International Settlements reported in March 2026.
6. Re-measure after every change, holding the other steps fixed. Curve improvements arrive a point or two at a time.

A lender who can say "our income-document step converts at 41% on mobile and 63% on desktop" holds better information than any published benchmark will supply, and is the only party who can produce it.

Common questions

What is an application abandonment curve?
An application abandonment curve is the completion rate of an application measured step by step rather than end to end. It records entries, completions, exits and time-in-step for each screen, which attributes a loss to one specific ask rather than to the flow as a whole.

What is a normal loan application abandonment rate?
No credible published benchmark exists. Figures circulating for loan application abandonment generally originate in vendor materials that do not disclose sample, definition or measurement window. The practical answer is to instrument the funnel per step and compare each step against its own prior months.

Which application step loses the most people?
Public evidence points to identity and bank-connection steps, with the UK National Audit Office recording 48% identity verification success in March 2019 and Open Banking Limited recording roughly 50% authentication completion in November 2019. Document upload is likely worse and has no public measurement.

Does asking for account creation hurt completion?
The Baymard Institute's checkout research found 18% of US online shoppers abandoned an order because the site required them to create an account, and 17% because checkout was too long. Lending applicants expect more enrolment than shoppers, so treat the direction as transferable, the size as untested.

How many applicants will connect a bank account at all?
BIS Papers No 168, published by the Bank for International Settlements on 30 March 2026, found only 27% of Brazilian consumers would share data even when shown improved loan terms and higher credit limits. That ceiling makes a document fallback a design requirement, not a courtesy.

---

Carousel instruments every step of the intake flow so the drop-off is visible rather than inferred. See how verification fits your flow