Measuring Form Completion: Starts, Errors and Finishes

Measure a form in four numbers: starts (people who began it), finishes (people who submitted successfully), the completion rate between them, and, where the tooling allows, the points of loss - the field or step where starters leave, and the fields where errors cluster. Those four numbers separate the three diseases that look identical from the outside: a traffic problem (few starts), a form problem (starts that do not finish), and a field problem (loss or errors concentrated at one question). Judge changes against the completion rate for comparable traffic over whole weeks, and keep page metrics out of it: the form's rate answers only for the form.

Define a start consistently - the first field focused or the first character typed are the practical choices - and have your analytics or form tool record it as an event, alongside the successful submission. Without a start count, a form receiving few submissions might be unvisited or unfinishable, and the two failures demand opposite work: better traffic and offers in the first case, better form craft in the second. Google Analytics support documentation covers events and key events for this kind of tracking; form platforms often count views and starts natively. Whichever source you use, write the definition in the log and never quietly change it mid-comparison.

For multi-step forms, the per-step counts locate the loss precisely, as the multi-step guide describes. For single forms, error data is the map: fields generating frequent errors, and fields corrected repeatedly before submit, mark the questions visitors find unclear or unforgiving - read alongside the error guide, they form a prioritised fix list written by your own visitors. Where the tooling offers session recordings, watch a handful of form sessions at the worst point before redesigning anything; ten minutes of watching people hesitate teaches more than a month of rates.

Form completion splits sharply by device, because the mobile experience differs in keyboard, screen and context, as the mobile forms guide details. Always read completion for phone and desktop separately: a healthy blended rate can conceal a phone rate half the desktop one, on the device most of your visitors use. The same applies, more cheaply, to traffic source where volumes allow - a campaign sending colder visitors will show a lower completion that is the audience speaking, not the form. Segment first, conclude second, in that order, every time.

When you cut a field, rewrite a question or fix an error message, pre-commit the read: completion rate for the same device mix, over whole weeks, with enough starts that the counts mean something. Illustration: on a form with 100 starts a week, a change needs several weeks before a real improvement separates from ordinary week-to-week wobble; declaring a winner after 40 starts is reading tea leaves. The foundations bench's discipline on baselines and read windows applies unchanged. Log the change, the date and the expected effect before the read begins, so memory cannot improve the story afterwards.

A completed form can still fail the business: submissions that arrive as spam, enquiries too thin to answer, quote requests missing the one fact pricing needs. Add one quality measure beside completion - the share of submissions a human judges usable, sampled monthly - and watch the pair together. A redesign that lifts completion by admitting junk has moved the cost from the visitor to the inbox, not removed it. The balance between enough detail and too much, from the quote guide, is steered by exactly this pair of numbers: completion rate and usable share, read together, never alone.

Fold the four numbers into the monthly review: starts, finishes, completion by device, worst field or step, usable share. Trends matter more than any single month; a completion rate sagging across a quarter is a form ageing through accumulated fields, a tool update, or a shifting audience, and the log will usually say which. The parent guide, forms and lead capture, places this routine as the bench's steering wheel: audits propose, measurement disposes, and no form change is argued from opinion once the numbers exist.