Made a new data extraction & automation app would some insight

[harold-auto]

Hi everyone,
I've spent the last while building something to automate data entry: pulling data from invoices, receipts and other documents into whatever system it needs to land in.

I've worked in IT and been through several ERP implementations on the client side at SMEs, and the gap I keep seeing is that most tools in this space treat "admin" as just typing numbers in. It isn't — admins check data against internal rules, flag what's off, and make judgment calls that most software ignores.

So beyond just extracting the data, what I've built also:

• Maps everything into the column names you actually use, not generic defaults
• Flags missing or inconsistent fields instead of assuming it's all correct
• Applies validation rules — custom sums, conditional logic ("if this name appears in this column, do X")
• Learns from trickier or non-standard document layouts
• Pulls specific values out of free-text fields e.g. "6 x Hot Dog Tins" → it grabs just the 6, not the whole string

Has anyone here actually used tools in this space day to day? What's missing, or what winds you up about the ones you've tried?

 

[leaddrive]

Not invoices, but I build extraction tooling too (pulling contact details and site faults from local business websites), so one thing from that side that transfers: the feature people trusted most wasn't extraction accuracy, it was the tool saying "not found" and showing the timestamp of the check, instead of guessing. Every extraction tool I tried before building my own would rather fill a field with something plausible than leave it blank, and one wrong-but-confident value costs more trust than ten blanks. Your "flags missing or inconsistent fields" line is the bit I'd lead with in any demo, and I'd go further and store every extracted value with where on the document it came from, so the admin can click through and verify in a couple of seconds. The other thing that winds people up: tools that learn a layout silently, then a supplier changes their invoice template and the errors are silent too. A per-supplier "layout changed since last time" warning would be worth more than another integration. What are you seeing as the typical error rate on first pass with a new supplier?