A per-question drop-off chart is the most useful thing a form tool can give you and the easiest to misread. It shows a bar for each question and how many people got that far, and there is almost always one bar that is obviously shorter than the one before it.
The natural conclusion is that the question at that point is bad and should be deleted. Sometimes that is right. Often it is not, and deleting it removes information you needed while leaving the actual problem in place.
The chart tells you where. It never tells you why, and there are four distinct whys that produce an identical shape.
One: the question is too expensive
Some questions cost real effort. Describe your project. Paste your current process. Upload a document. These are legitimately hard, and a drop at one of them is people deciding the answer is not worth the work.
The tell is the time data. If people sit on the question for a long while before leaving, they tried. They read it, thought about it, and gave up. That is an effort problem, and the fixes are to make the question smaller, make it optional, or move it after the point where you have their contact details so an abandonment still leaves you with a lead.
The fix that almost never works is rewording it. If the cost is genuine, nicer phrasing does not reduce it.
Two: the question is confusing
This looks similar on the chart and completely different in the timing. People arrive, hesitate briefly, and leave. They are not weighing up the effort. They do not know what is being asked.
Internal vocabulary is the usual culprit. A field labelled account type on a form aimed at the public. A dropdown with options that all sound like the same thing. A date field that does not say whether it wants the event date or today's date.
The cheapest diagnostic in existence is to show the question to one person outside your business and ask them what they think it means. If they pause, you have your answer, and the fix is a rewrite that takes five minutes.
A short pause before leaving means confusion. A long one means the question was simply too much work.
Three: the question is intrusive
Phone number is the classic. So is company size, budget, date of birth, and anything that reads as qualification rather than service. People do not find these hard or confusing. They find them presumptuous at that point in the relationship.
The signature is a clean, sharp drop with very little time spent, on a question that is trivially easy to answer. Nobody struggles to type a phone number. They decline to.
Three fixes work, in this order. Move it later, so the person has already invested. Make it optional, which costs you less than you think. Or explain why you want it, in six words, right under the field. A phone field that says so we can confirm the slot loses far fewer people than a bare one.
Four: it is not the question at all
The fourth cause is the one people miss, and it is common. The bar is short because something broke there, or because a previous answer sent people somewhere unexpected.
Validation is the usual suspect. A phone field that rejects a legitimate format, a postcode field that will not take a space, a required upload that fails silently on a phone. From the chart these look exactly like an intrusive question. From the person's side they look like a website that does not work.
Conditional logic is the other. If answering yes to question four skips to question nine, question five will show a drop that is not a drop at all. Before you change anything, check whether a branch explains it. This catches a surprising number of false alarms.
Separating the four
You do not need sophisticated tooling to tell these apart. You need three numbers per question and a little patience.
- Time on question before leaving. Long means effort, short means either confusion or refusal.
- Whether the question is hard. A phone number is not hard, so a drop there is refusal, not effort.
- Whether anyone submits it with an error first. Repeated failed submissions mean validation, not reluctance.
- Whether the question sits after a branch. If it does, check the branch before you blame the question.
The first bar is a different animal
Almost every form shows its biggest loss between the view and the first answer, and that one is not a question problem at all. It is the gap between what the page promised and what the form turned out to be.
People who arrive expecting a quick enquiry and find a nine-question intake leave immediately. So do people who clicked something ambiguous and landed somewhere they did not intend. Neither is fixed by editing question one, although a lighter opener genuinely helps.
Treat that first bar as feedback on the page above the form and the link that got them there, not on the form itself. It is frequently a marketing problem wearing a form problem's clothing.
Mobile and desktop are two different charts
If your tool lets you split by device, always do it, because an averaged chart hides the most actionable thing in the data. A drop that appears moderate overall is often severe on mobile and absent on desktop.
The causes are specific and fixable. Fields that need a keyboard and precision. Date pickers that behave badly on small screens. Uploads that require a file the person does not have on their phone. Long option lists that mean scrolling past the submit button.
Most small businesses now get well over half their form traffic from phones, so a mobile-only problem is a majority problem that the average makes look minor.
Partial responses are the transcript
If your tool stores answers as people go rather than only on submit, the abandoned responses are the single most useful thing in the account, and hardly anybody reads them.
They tell you what people managed before they stopped. An abandonment that already contains a name and an email is a lead you can follow up. A dozen abandonments that all stop with the same three questions answered is a diagnosis more precise than any chart.
Read twenty of them. Not aggregated, not summarised, just read them. You will find the problem faster than by staring at bars, and you will also find a few enquiries worth answering.
Change one thing
The temptation once you have a diagnosis is to fix everything at once: reword three questions, make two optional, move the phone field and shorten the whole thing.
Then completion moves and you have no idea which change did it, which means you cannot repeat it on the next form. Worse, if two changes pulled in opposite directions, the net effect can look like nothing happened.
Change the single biggest drop. Wait for enough submissions to mean something, which at typical small-business volumes is usually a few weeks rather than a few days. Then change the next one. It is slower and it is the only way to actually learn anything.
How much data is enough
Small-business form volumes make it very easy to act on noise. If forty people saw the form and twelve finished it, a bar that looks alarming can be four people who happened to have a bad morning, and rebuilding a question on the strength of that is worse than doing nothing.
There is no clean threshold, but a rough working rule is that you want at least fifty people to have reached a question before you take its drop-off seriously, and a couple of hundred before you believe a difference of a few percentage points between two questions. Below that, use the chart to generate hypotheses and use reading actual abandoned responses to test them.
The other half of this is time. Traffic is not evenly distributed across a week, and a chart built from Tuesday and Wednesday can look nothing like one built from the weekend. Wait for whole weeks rather than whole numbers of days.
Views, starts and completions are three different numbers
Most confusion about form analytics comes from collapsing these. A view is somebody who loaded the page. A start is somebody who answered something. A completion is somebody who submitted. The ratios between them point at completely different problems and the same overall percentage can come from very different situations.
A low start rate against high views is a promise problem, living on the page rather than in the form. A high start rate with a low completion rate is a form problem, and that is where the per-question chart earns its keep. A low view count with excellent conversion is a traffic problem and no amount of form work will touch it.
Write the three numbers down before you open the drop-off chart. It takes a minute and it stops you spending a fortnight fixing the wrong layer.
Seasonality and campaign mixing
A form that is linked from four places is really four forms, and averaging them produces a chart that describes nobody. Traffic from a price-comparison page behaves differently from traffic from your own newsletter, and the difference usually shows up in exactly the questions that matter.
If your tool records where submissions came from, split the chart. If it does not, the cheap version is to use a different link for each source and compare submission counts, which gets you most of the insight for none of the effort.
The same caution applies across time. A drop that appeared in January on a form aimed at the wedding trade is not a design regression, and quite a few forms get rebuilt every year in response to nothing more than the calendar.
The dip near the end
Most forms show a small extra loss at the final step, and it is usually mistaken for fatigue. Occasionally it is. More often it is the last screen asking for something the earlier ones did not prepare anybody for.
Consent checkboxes, marketing opt-ins, terms links and a suddenly required phone number all cluster at the end, because that is where they get added when somebody remembers they are needed. To the person filling it in, the form changed character at the last moment, which is exactly when they have the most to lose by walking away and therefore notice it most sharply.
If your final step has more than one new demand on it, move something earlier. A consent box on the same screen as the questions it relates to reads as ordinary. The same box alone at the end reads as a catch.
When the chart is fine and the leads are bad
Sometimes completion is healthy and the submissions are useless. That is not a drop-off problem and the chart cannot help you with it, but it is worth naming because people reach for the same tool.
A form that converts brilliantly and produces unqualified leads is usually a form that has removed all the friction, including the useful kind. A budget question or a timeline question will cost you completions and improve the average submission, and that trade is often worth making deliberately.
Decide which number you are optimising before you start moving things. A form is not trying to be completed by everyone. It is trying to be completed by the right people.