Every product team I've worked with has had a funnel chart of their onboarding, and most of them could recite the drop off numbers from memory. Sixty percent finish step one, thirty eight reach the integration screen, twelve connect a data source, and somewhere around nine become a real user. The chart is accurate, it updates nightly, and it will never tell you a single thing about why any of those people stopped.
That doesn't stop anyone from deciding they know. A big drop at the integration step gets read as "the integration step is too hard," and six weeks of engineering goes into simplifying it, and the number moves by about a point. It turns out a good chunk of the people leaving at that screen weren't stuck. They were fine with the screen and didn't have credentials for the system they needed to connect, and were waiting for someone in their IT team to get back to them. That's not a design problem, it's a sequencing problem, and no amount of staring at the funnel would have revealed it.
What behavioural data is actually good at
Analytics isn't the villain here, to be clear. It does something feedback can't: it measures everybody. Every user who abandoned is in the chart, including the ninety five percent who'd never answer a question, and that completeness is genuinely valuable. If you want to know whether a change made things better or worse, the funnel is the instrument, not a handful of comments.
What it measures is the outside of the behaviour. It knows the click happened, the page loaded, the session ended. It has no access to intention, and onboarding is almost entirely about intention. Two users who both leave at the same screen after ninety seconds can be having opposite experiences: one is confused and defeated, one has decided this looks fine and they'll come back on Monday when they've got the file ready. Same event, same duration, different futures. Aggregate a few thousand of those and you get a number that hides both.
The other thing it can't see is anything outside your product. Somebody who abandoned because their colleague said they'd already bought something similar, or because the pricing page made them want to check with a manager, exits your funnel identically to someone who couldn't find the button.
The questions that fill the gap
The useful onboarding questions are not satisfaction questions. Asking a new user to rate their onboarding experience out of five produces a number with no diagnostic content. What you want is what they were trying to do and where the attempt ran into something.
Three that consistently earn their place:
What were you hoping to get done first? Asked early, this tells you whether your onboarding is pointed at the same goal the user arrived with. A surprising amount of onboarding friction is really a mismatch between the tour you built and the job the person came for. They'll complete your setup checklist politely and then still not have done their thing.
Was there anything you tried to do and couldn't? This catches the invisible abandonment. Analytics will show you a page they left. It won't show you they were attempting to import two years of records and gave up when the format was rejected without an explanation.
What nearly stopped you signing up? Ask this of people who did make it through, because they can describe the hesitation the ones who left can't be asked about. The answers here are frequently about pricing clarity or trust rather than product.
Notice all three are about the user's task rather than your interface. That's deliberate. People are bad at diagnosing software and good at describing what they were trying to achieve, and the second is more useful anyway. What to ask users after signup covers the specific wording in more depth, and what to ask new users in the first week covers the days after.
Where to ask, and the honest limitation
In product beats email for this, because it's close to the moment and you don't need a working email relationship. A single question at the point of an action, or a small prompt on a screen where people commonly stall, gets response rates that a survey never will. The catch is that most in product widgets get ignored because they arrive generically and at bad moments, which why your in-app feedback widget is ignored goes into.
Here's the limitation nobody solves cleanly: the users you most want to hear from are the ones who left, and they're the least likely to answer. You'll disproportionately hear from people who got far enough to care. That's a real bias and it means onboarding feedback tends to overstate how well the middle of your funnel works. The partial fix is to ask a small number of people who abandoned, directly and personally, by email, with one question. A handful of replies from that group is worth more than a hundred from users who completed. Feedback from users who never contact support covers the general problem of the group that stays quiet.
Using both together
The way this fits together in practice is that the funnel picks the question and the answers explain the number. You look at the chart, find the step where people stop, and then go and ask the people who reached that step what was happening. Not "why did you drop off," which nobody can answer about themselves, but what they were trying to do at that point.
Do it in that order and analytics stops being a source of theories and becomes a way of choosing where to spend your curiosity. Do it the other way round, collecting feedback and then looking for numbers to support it, and you'll find them, because with enough dimensions you can always find a chart that agrees with you.
Reading the answers is its own small problem, since onboarding responses arrive in a trickle and in fragments. Qria is built for the reading side, holding structured responses together with the public reviews you're picking up on software directories, with the AI summarising what keeps coming up so a scattering of one line answers turns into two or three recurring themes. What usually falls out is that the reason for the drop off had been stated plainly by four different people over two months, in slightly different words, in a place where nobody was reading them together.
The funnel is a good instrument pointed at the wrong question. It tells you where to look. Someone still has to go and ask.
