Forms: the last metre of the journey and the first leak

One in two visitors who start filling in a form never finishes it. Not "never found the site", not "never understood the offer": they found it, they read it, they clicked, they started typing and they left halfway through.

The figure comes from Zuko (opens in a new tab), a form analytics tool that publishes its aggregated data: across 93,022,997 tracked sessions in 17 industries, 51.71% of people who start a form complete it. Search, content, advertising and design all exist to bring someone to exactly that point. The form is the last metre of the journey and it is almost always the part nobody looks at.

The same data shows a sharp gap by device: 54.48% completion on desktop, 47.53% on mobile. Close to seven points, across every industry measured but one.

On the ecommerce side, the Baymard Institute (opens in a new tab) compiles 50 studies into an average cart abandonment rate of 70.22%. That raw number says very little, since half of those people are window shopping. The breakdown of reasons is far more useful. Excluding the 42% who report browsing with no intention to buy, here is what remains.

Three of those reasons come straight from form design: the mandatory account, the length of the checkout and the errors. None of them is about the product or the price. They are about what people are asked to do in order to hand over their money.

On length, Baymard measures (opens in a new tab) an average of 11.3 fields in a checkout flow in 2024, against 12.7 in 2019. The trend is going the right way but slowly. Their estimate of the bare minimum for most sites: 8 fields.

The advice is everywhere and it is not wrong. A clean experiment even backs it up: in October 2013, MarketingSherpa ran a live test (opens in a new tab) of a download form on roughly 300,000 recipients. The only difference between the two versions was one extra field asking for job title. The result: 58% conversion without the field, 51% with it, close to twelve percent lost, at a 99% confidence level. The test dates from 2013 and ran on a professional audience: the order of magnitude is instructive, the exact value does not transfer as is.

Where the advice becomes misleading is when people conclude that field count is the variable to optimise. It is not the count that costs, it is what each field demands. An email field the browser fills on its own costs almost nothing. A "VAT registration number" field costs an extra tab and two minutes of searching. A "how did you hear about us?" field costs a decision, and a decision is the most expensive thing you can ask of someone who has already decided to contact you.

The most quoted story in the field illustrates the misunderstanding well. Expedia supposedly gained 12 million dollars a year by removing an optional "company" field. The figure has been circulating for fifteen years and its only public source is a talk by Joe Megibow, then head of analytics at Expedia, reported in late 2010 (opens in a new tab). No published protocol, no methodology. Treat the number for what it is: a conference-stage figure.

What is genuinely instructive is the mechanism described. Customers filled the "company" field with the name of their bank, then entered that bank's address below rather than their own. Billing address verification failed and the payment was declined. The field was not driving people away by simply existing, it was manufacturing cascading errors in the fields that followed. A field does not cost you because it is there, it costs you because of what it triggers.

Then there is the trade-off nobody names. Shortening a form increases the number of submissions, not automatically the revenue. A two-field contact form fills a calendar with unqualified meetings and moves the sorting work downstream, where it costs human hours. So the right question is not "how many fields" but "which information has to be known now". Anything that can wait for the first conversation should leave the form.

Abandonment never happens "in the form" in general. It happens at one of three precise moments and each one is fixed differently.

Before the first keystroke

Nobody has typed anything yet, so the decision rests on perceived cost. A twelve-field form displayed as one block drives people away before they even try, while those same twelve fields spread across four screens get filled in without the question of abandoning ever coming up.

This is the "one thing per page" (opens in a new tab) principle formalised by the GOV.UK teams and now written into their service manual (opens in a new tab): one question per screen. Their reasons are documented, lower cognitive load, better behaviour on mobile and simpler error handling. They also state the limit themselves: for a form filled in daily by the same people, splitting it up turns into a clicking chore. The principle applies to what occasional visitors fill in, which covers the overwhelming majority of forms on a company website.

Two details also shift perception at that moment. Mark the optional fields rather than starring the required ones: an asterisk on every line reads like a full interrogation. And show real progress as soon as the flow runs past two screens.

During data entry

Here the highest-return lever is not a redesign, it is an hour of your developer's time. Every field on a form can be labelled so the browser knows what it holds and fills it in automatically. The Chrome team analysed (opens in a new tab) thousands of address and credit card forms across millions of page loads: people who use that autofill abandon 75% less often and spend roughly 35% less time on the form. Chrome states the limit honestly, this is an observed correlation, not a controlled experiment.

Zuko measured the same thing across 215 forms (opens in a new tab): 71% completion with autofill against 59% without. The detail is more instructive than the average. Of those 215 forms, 160 converted better, 31 showed no difference and 24 converted worse. The cause is always the same: badly labelled fields, where the browser drops the information in the wrong place. Autofill is not a gift, it is a contract. It works when the work has been done and it hurts when it has not. Worth asking for explicitly at your next technical update.

At the moment of the error

This is the harshest breaking point, because the person has already invested time. The only published experiment comparing when an error message should appear dates from 2009: Luke Wroblewski and the agency Etre (opens in a new tab) tested six variations of the same form on 22 participants wearing eye trackers.

Twenty-two participants and a 2009 web: the sample is small and the context has aged. Take it for what it is, the only side-by-side comparison ever published on the question, not a universal law.

The content of the message matters as much as its timing. A message that states a fact is useless, a message that names the expected correction solves the problem. "Invalid format" leaves the person searching, "The number must contain 10 digits, no spaces" hands them the answer. And a field must never be cleared after an error: making someone retype a full address because a postcode was badly formatted is handing them a reason to leave.

According to Statcounter (opens in a new tab), mobile accounted for 51.47% of worldwide web traffic in June 2026. So most of your forms are filled in with a thumb, often standing up, on an imperfect network. And that is where the completion rate is lowest. The seven-point gap does not come from screen size, it comes from a series of details that never show up on a mockup.

None of these show up when you resize a browser window. They show up when you fill in the form yourself on a real phone, with a real thumb, on a real mobile network.

Forms are where accessibility obligations and data protection obligations overlap most directly. WCAG 2.2 (opens in a new tab), the reference the European Accessibility Act relies on since it became applicable on 28 June 2025, added criteria that target them explicitly: information already provided in the same process must not be requested again, signing in must not require a memory test and pasting must remain possible in a password field. The real cost of those exclusions is the subject of another article (opens in a new tab).

The GDPR adds a constraint that pulls in the same direction as conversion, which is rare enough to be worth noting. The data minimisation principle, article 5.1.c (opens in a new tab), requires collected data to be adequate, relevant and limited to what is necessary. Every field therefore has to be justifiable by a purpose. A short form is not only more effective, it is also a far easier position to defend under scrutiny.

Then there is the grey area: pre-ticked boxes, marketing consent bundled into acceptance of the terms, a decline button made deliberately discreet. Those practices earn addresses in the short term and are now being penalised, as detailed in the article on what regulators punish in 2026 (opens in a new tab).

A standard analytics tool stops at the page. It tells you how many people saw the form and how many submitted it. Between those two numbers sits a black box and that is exactly where everything happens. Four indicators are enough to open it, and they come either from a dedicated tool or from a handful of events set up by your developer.

Putting the measurement in place before touching the form avoids the usual trap: changing five things at once, then watching a number go up without knowing which change produced it. The general method for tying an interface change to a result is covered in this article on measurement (opens in a new tab).

A form is the one place on a site where the relationship reverses. Everywhere else you give: information, a demonstration, proof, a reason to stay. Here you ask. And you ask at the exact moment the person has decided you are worth it. Which makes it the moment where losing them costs the most.

That is what makes the gap so striking. A visual redesign gets seen, discussed and budgeted. Having eight fields labelled gets seen by nobody and takes an hour. The available data suggests the second intervention weighs more than the first on the one metric that decides whether a visitor becomes a customer. The rest of your site works to bring someone all the way to that field: it would be a shame if it were the only part nobody thought about.