I've done similar before and was still blown away by the bad data.
Somewhat unrelated, but still a hell of a story in the power of human input into data...
Working in the healthcare industry during COVID, federal law had 18,000 of our employees required to submit proof of vaccination to continue working in our hospitals and clinics. All they had to do was get their vaccination certificate PDF off the government website, type in their staff number, and upload the form, we then submit this information as the employer to confirm that these people do indeed work for us and are safe to continue doing so.
56% managed to do it. The rest were all sorts of shit. Most common were people that took photos of their computer screen, converted the photo to PDF, and uploaded that. Next most common was people print the PDF, scan it, then upload the scan PDF.
We had thought of everything to make a simple download then upload as easy as possible, including a 3 step video, and yet they went above and beyond in unimaginable ways. The people that genuinely didn't know what to do hit the support link so they could be guided through it and did things perfectly in a couple minsβthe self-confessed computer illiterate people were not a problem at all.
Thanks to training a form detection bot, I got it down to under 2000 remaining in a day, and the looming threat of "You have to do this or we can't legally give you work and pay you until you do" quickly sorted out the rest.
People will ALWAYS fuck things up in ways you've never thought of before. Reading the short, clear, and user friendly instructions for the simple job doesn't work and they'll get angry that something went wrong, every fucking time.