238 of 240 invoices, extracted free
A builder tested new browser AI tools on 240 real client invoices. The files were scans, phone photos, and Word exports. He ran the test on a $600 laptop with no special graphics card. An on-device model in Chrome read each invoice in chunks through a small extension. The job was to return vendor, line items, subtotal, tax, and total as clean data for the accounts system. The tool had to fail loudly on a page it could not read.
Start with one job only. Fill in the three lines. Keep the bot on that job.
The one job this bot does: ______________________________
What it may do on its own: ______________________________
What it must ask me first: ______________________________
Two lines from this case, side by side. No new figures.
A builder tested new browser AI tools on 240 real client invoices. The files were scans, phone photos, and Word exports. He ran the test on a $600 laptop with no special graphics card. An on-device model in Chrome read each invoice in chunks through a small extension. The job was to return vendor, line items, subtotal, tax, and total as clean data for the accounts system. The tool had to fail loudly on a page it could not read.
The local model extracted 238 of 240 invoices cleanly. Two badly rotated phone photos failed, and the same two had broken his old cloud flow. Cost was $0 per invoice, against about 1.1 cents per invoice all-in for the cloud flow. His rule: under 200 documents a month, cloud is fine. From 200 to 2,000, local first with cloud fallback works best.
The after line is the source's claim, not our measurement.
smalloffice.space · Get a case like this in your inbox each morning. Subscribe free
Read this case on your phone now, and keep it for later. Type your email. We send this case to that address, and you get the next cases each morning too.