The three numbers, and why they differ
Opens: how many times the form was opened in a browser. Starts: how many people answered at least one question. Completions: how many reached the end. The biggest gap is almost always between the first two, and it's about the intro and the first question — not the questionnaire.
Completion rate calculated on «starts» is a kind number: it excludes exactly the people who opened and left, which is who a badly made form loses. Calculated on opens it's lower and truer.
The three typical drops
A drop at the open: visible when opens are many and starts are few. The cause is nearly always expectation — the form looks longer than promised, or it isn't clear who's sending it. You fix that outside the form, in the invitation.
A drop at one specific question: a required question asking something personal, or an open question placed too early. Fix it by making it optional, moving it later, or removing it.
A slow, steady decline: no single question is guilty, the form is simply too long. Nothing to fix here — something to cut.
How to read it without fooling yourself
With conditional logic the curve stops being monotonic: a question almost everyone skips looks like a cliff and isn't one. Before panicking, check whether that question sits inside a branch — in drop-off, «reached a question» means the path went through it, not that everyone had to answer.
The other trap is volume: across twenty responses, two abandonments on the same question are chance, not a signal. Under a hundred responses, drop-off is for forming a suspicion, not for deciding.
The one fix that nearly always works
Move the hardest question after an easy one. Someone who has already answered has invested something and carries on; someone who meets the biggest effort first closes the tab. It doesn't change the content of the survey, and it's often enough.
Look at where the curve falls: at the open it's the invitation, at one question it's that question, steadily throughout it's the length.