Completion rate

Defining "started"

Completion rate is only as meaningful as its denominator, and "started" is ambiguous. Landing on the first screen is not the same as answering the first question, and neither is the same as clicking through to page two. Pick one definition, write it down, and use the same one next quarter — a completion rate that improved because you redefined the denominator has improved nothing.

The most useful definition for survey design work is "gave at least one answer", because it excludes people who opened the link out of curiosity and isolates the population who genuinely intended to take part.

Where people drop out

Drop-off is rarely spread evenly. It clusters, and the clusters are diagnostic.

• The first screen. Usually a mismatch between what the invitation promised and what the survey looks like, or a visible progress indicator revealing it is longer than expected.
• The first free-text box. Typing costs an order of magnitude more effort than tapping. See open-ended questions.
• A grid. A matrix with ten rows on a phone is where people decide they have done enough.
• Anything that feels personal too early. Demographics at the top of a survey read as gatekeeping. Put them at the end, where the respondent has already invested effort and understands why you are asking.
• A validation error they cannot clear. A postcode rule that rejects a valid BFPO address, or a required field the respondent genuinely cannot answer, ends the session immediately.

A worked example

A 40-question consultation shows 2,000 starts and 700 completes — 35%. Splitting drop-off by question shows 500 of the 1,300 losses at a single ten-row matrix on page three. Replacing that grid with three separate questions gated by conditional logic — so most respondents see one of them rather than all ten rows — is a bigger intervention than rewriting the invitation email, and it costs an afternoon.

The general principle: completion rate is a property of the survey, so fix it in the survey. Compare with response rate, which is a property of the invitation and is fixed elsewhere.

Common mistakes

• Treating partial responses as worthless. Someone who answered 18 of 40 questions has given you 18 answers. Whether you can use them depends on your analysis, not on whether they pressed submit.
• Chasing the number by removing questions you need. A 100% completion rate on a survey that no longer answers your question is not a win.
• Ignoring who drops out. If the people abandoning at the matrix are disproportionately older or on phones, your remaining sample has shifted. That is sampling bias arriving through the back door.
• Comparing across survey types. A two-question pulse check and a 40-question consultation do not belong on the same chart.

In NumoForms

Results show response counts and completion for each survey. Three product features exist mainly to move this number: conditional logic, so people are not asked things that do not apply to them; save and continue, which issues a resume link so a long survey can be finished in two sittings rather than abandoned; and the conversational layout, which presents one question at a time and auto-advances on one-tap questions. Test responses can be cleared before you go live, which keeps your baseline honest — clearing them also removes any uploads and partials they created.