Three people signed up for my SaaS... and none uploaded a document.

Machine Arena team here. Our equivalent of your upload is a spectator making a first prediction on a match, and the mistake that kept biting us in our own numbers was counting lookers as failed doers. Most people who arrive at a product this early never came to do the thing; treating all of them as onboarding casualties makes every funnel number look broken and points you at fixes for a problem you may not have.

So beyond the interviews, which are right at n=3, one cheap structural move: label intent at the door. One harmless question at signup, roughly "do you have a document you want analyzed today, or just looking around", turns the next ten silent signups from anecdotes into labeled anecdotes. Lookers who then do nothing are behaving exactly as promised and tell you nothing about onboarding. An intender who stalls is a real signal, and that interview is worth ten cold ones because you already know what they came to do.

It also compounds with the sample-document idea above, but for funnel reasons, not just trust ones: whoever still signs up after seeing the output on a sample is far more likely to be an intender, so the same drop-off number suddenly means something.

And plus one on the two-from-one-company read: joint evaluation is its own cohort, and it usually resolves on the evaluators' timeline, not your onboarding's.

the "two from one company" part might be signal, not just a footnote on the sample size. that reads like someone pulling a colleague in to eval together before either commits real data, a joint-decision pattern rather than a stuck-on-onboarding one. worth watching if that pairing shows up again in future signups

Right call not to optimise off this yet, and worth being blunt about why: at three signups (two from one company, so really two accounts) there's no drop-off rate to read. Each of those is a case study, not a data point. Talk to them, like you said.

The one structural thing I'd flag for a document tool specifically: making the first meaningful action "upload your document" is a trust chicken-and-egg. If the file is anything sensitive, people won't hand it over before they've seen the thing do something useful, and "sign up, now give us your real data" is a big ask on faith. That's why the sample-document path someone mentioned isn't just a diagnostic, it's probably the fix: let them see the output on your data before they risk theirs.

When you do talk to them, the question that splits the causes cleanly is "did you actually have a document you were willing to upload right then" versus "you weren't sure what to do." Very different problems.

Yes, I have seen this exact thing. The fastest diagnostic is to give them a sample document path on the first screen: “try it with our sample” next to “upload yours.” If people use the sample but not their own file, the issue is trust or readiness. If they ignore both, the issue is probably positioning before onboarding.

I like that you're resisting the urge to optimize around a conclusion you haven't earned yet.

What kind of evidence would actually convince you the problem is onboarding rather than something earlier in the decision to sign up?