The short answer
Session recordings showed merchants reaching the bank statement upload step on our funding application and leaving — not the unqualified ones, merchants who had already answered the revenue and amount questions. The step came off, cost per lead fell significantly, and the quality of what came through went up rather than down. Documents belong in the sales conversation, not in the application.
Asking for statements on the application is defensible on paper. A merchant who uploads four months of statements is obviously more serious than one who does not, the file arrives further along, and the reps stop chasing documents. We ran it, we watched what it did, and we took it off.
How it was found
Not from the conversion rate, which only says that fewer people finish than start. Microsoft Clarity records every session on the page, so the abandoned ones can be watched individually, and the pattern was unmistakable: merchants worked through the revenue question, the amount question and their contact details, arrived at the upload step, and left. Same step, over and over.
The important part of that pattern is who was leaving. A drop-off at the revenue question removes people below the floor, which is the form working. A drop-off at a document upload removes whoever cannot lay hands on four months of statements in the next few minutes, and that is a test of what a merchant is doing at that moment rather than of what their business looks like. The merchant with $90,000 a month going through the account, on a phone, in a van, is exactly who you want, and exactly who cannot do it.
What happened when it came off
- Cost per lead fell significantly. More of the merchants who reached the step completed, on the same spend and the same creatives.
- Quality went up rather than down, which was the opposite of the argument for keeping it. The gates that actually qualify — revenue, amount, time in business, state — were all in front of the upload, so nothing that was doing the qualifying was removed.
- The pixel got a bigger and cleaner sample. Every merchant who used to leave at the upload was a conversion event Meta never received, and they were disproportionately the busy ones.
- Statements still arrive. They arrive on the phone call, where a rep can explain why they are needed and walk the merchant through sending them, which is a conversation rather than a wall. Every exclusive lead we sell now arrives with 3 months of bank statements.
The general rule it points at
Friction on an application is only good when the people it removes are people you did not want. A one-time passcode qualifies because everyone it stops has an unreachable phone. A document upload does not, because it sorts merchants by whether they are at a desk. The test is not how much effort a step costs; it is whether the effort correlates with anything you are trying to find out.
It also points at the wider habit. There is a specific reason any funnel underperforms, and it is usually visible in the recordings within an afternoon. We would not have found this one from a conversion rate, a benchmark, or an argument in a meeting — only from watching about twenty people do the same thing and stop at the same place.