How to improve form completion rate

Every form that nobody finishes has the same post mortem. Somebody wanted a field, nobody argued, and the field went in. Repeat that eight times over a year and you have a form that asks 24 questions to book a 20 minute call.

Field count is the one lever you can measure before you ship. Everything else here needs traffic to prove, so it goes first.

What completion rate means and how to measure it

Completion rate is submissions divided by the number of people who saw the form. That second number is the awkward one, because it does not exist inside the product. A response row is created when somebody submits, and the statuses it can carry are submitted, partially signed, signing completed, completed and failed. There is no started state and no abandoned state, so nothing in the responses table can tell you who left halfway.

The denominator has to come from your own analytics: page views on the form’s URL, or on the page that embeds it. Wire that up before you change anything, or every change below is a matter of opinion.

For a reference point outside your own numbers, Baymard Institute’s e-commerce benchmark of more than 340 top sites found the average checkout in 2024 carried 11.3 form fields, where by their analysis most sites need only 8. Their roll-up of 50 studies puts average cart abandonment at 70.22%, with a checkout that was too long or complicated named by 17% of respondents. Baymard does not publish the sample size for that reasons question on the page, so treat the ranking as directional rather than as a measurement.

Lever 1, cut fields: the 55-form table

The table at the foot of this post is the field count of every one of the 55 finished forms published on this site, counted from their field lists rather than estimated. The whole set runs from 6 fields to 29, with a median of 16.

The split by format is sharper than the split by subject. The 39 classic forms have a median of 18 fields; the 16 conversational ones have a median of 9.5. That is a fact about what people choose each format for, not about the formats. Nobody builds a 29 field form one question at a time.

Compare your form against its own kind. A newsletter signup at 16 fields is broken. A rental application at 16 is unusually lean. Two questions shrink a form fastest: does anything downstream consume this answer, and could we ask it later, once the person has committed. Nielsen Norman Group’s guidance makes the same point in reverse, that fields which can be derived elsewhere, collected later or dropped entirely should be.

Here is what the 55 forms are actually made of, counted by field kind:

Field kind	Times used	Forms using it
Short text	245	54 of 55
Single choice	162	53
Long text	111	52
Multiple choice	74	47
Dropdown	72	44
Full name	49	42
Date	45	30
Table	30	20
Number	27	18
File upload	19	17
Address	18	16
Signature	18	16
Rating grid	10	8
Photo upload	7	7

Short text carries 245 of the 890 answered fields across the set. That is where the savings are. Every short text field is a keyboard, a decision about format, and a chance to make a mistake, and a good proportion of them are asking for something a choice field could ask in one tap.

Lever 2, order: easy first, identity last

The first question sets the price. If it is easy and answerable without looking anything up, people start. If it is an email address, a proportion of them work out they are being captured and close the tab.

The exception is the signup, where the email is the transaction. Everywhere else, put the interesting question first and the contact details after. Across the 55 examples the median form has 10 required fields out of 16, and the ones that read best front-load the choice questions and leave name, email and phone in a block near the end.

Lever 3, one question at a time vs one page

A classic form shows everything at once, so a respondent can scan the whole ask and skip about. A conversational form shows one question per screen, so each answer costs less and there are more of them.

Short forms win on feel that way. Long ones lose the thing people need most, the ability to see how much is left and check what they already said. This is the choice most likely to move your number in either direction, so it has its own post: conversational form vs standard form sets out the decision rule and what each format can and cannot collect.

Lever 4, mobile: sticky submit, single column

Assume a phone. Two things matter most on a small screen.

The first is a single column. Nielsen Norman Group’s guidance is blunt about it: multiple columns interrupt the vertical momentum of moving down a form, and the only reasonable exception is a run of short related fields like suburb, state and postcode. The classic layout has a single column mode that enforces this; the single column form page shows what it produces.

The second is the submit button. It has two modes, sitting at the end of the form or fixed at the bottom of the screen, and on anything longer than two or three screens the fixed version removes a scroll that people otherwise have to perform from memory. The web.dev forms course makes the related point about keyboards: choose the input type so a phone offers the right keyboard, because a numeric field that opens a full QWERTY is a small tax paid on every entry.

Lever 5, the welcome screen and time estimate

A welcome screen is a trade. It costs one tap and buys the chance to say what the form is for and how long it takes. There are ten layouts and every one can carry an estimated time.

Publish the estimate only when it helps. “1 min” on a signup is an invitation. “15 min” on a performance review is honest and still worth showing, because the alternative is somebody starting it between meetings and abandoning at question nine. Do not guess low. A wrong estimate is worse than none.

Lever 6, errors and required marks

Mark required fields at the field. NN/g’s position is that people do not read the instruction at the top of the form and will not remember it while filling one in, so “all fields required unless marked” is a sentence that does nothing.

Two product specifics are worth knowing here. On a published classic form, a field’s description is only rendered for file upload and photo. Every other field silently drops it, so any explanation a respondent must read has to be a paragraph block above the field. And a conversational number field does not pass its minimum and maximum to the input, so the range is only enforced on submit, after every question has been answered. If a number has to be in a range, say so in the label.

The example: a six-field signup

The newsletter signup example is the shortest form in the set at six fields, and only one of them is required. It is conversational, so each question is its own screen, and the welcome screen carries a “1 min” estimate against a promise of one email a month.

What makes it work is that the five optional questions all change what gets sent rather than adding to a record. Somebody can answer the email question, skip everything else, and still be correctly subscribed. That is the shape to copy: one required question that completes the transaction, and everything after it framed as the respondent tuning their own outcome. The one question at a time builder is where that format lives.