What this package is
This package turns one kind of document into data you can actually use. Drag a batch of files in, the system reads them, and you see exactly what it pulled from each one. Anything wrong you correct right there, without opening the original to compare line by line. Confirm, and the rows land in your own Excel or Google Sheet automatically.
The scope is one document type and up to eight fields, and I write every field name into the specification on the day we agree the work. Anything beyond that is quoted before it is built. Before handover I run one accuracy pass against your real documents, which matters more than it sounds: clean, well-lit samples always read better than something photographed at an angle in a warehouse. A system tested only on tidy samples fails on its first real day.
What this package does not include is a line-by-line item breakdown and intake from chat apps. Both are in the Check package.
See it working
Documents read, last seven days
Every number here is counted from the real documents on the intake page. Confirm one and come back.
Showing today's 12 documents of 1,247 in the system
This package gets one daily summary. Check adds a full spending dashboard with history by vendor and category.
This is a working model of the delivered screens. All the data lives in this page and it is not connected to anything. The real one reads the files you upload and writes to the sheet or system your team already uses.
Everything you get
This list is the entire scope of the package. Nothing is hidden in a contract.
- Reads one document type. Receipts, or tax invoices, or purchase orders.
- Pulls up to 8 fields: vendor, date, document number, net, tax, total and two of your own, named in the spec on day one.
- An upload page. Drag a batch in, see exactly what was read from each file, and correct it there.
- Results go straight into Excel or Google Sheets, no copying by hand.
- A daily figures page: how many documents were read, what they came to, and how many need a second pair of eyes.
- One accuracy pass against your own real documents before handover, not against tidy samples.
- Full handover pack in English: user guide, technical notes, README and code comments.
- 60 days of support after handover, with replies inside the working hours I publish, counted in your time zone.
- The per-page charge for reading documents, paid to the AI provider on your own account so the system is genuinely yours. The going rate is roughly 3 to 6 cents a page, so 500 documents a month lands around $20 to $30. I work the figure out from your real volume before you commit.
- A 100% accuracy guarantee. No system does that honestly. So I build a review step for the documents the system is unsure about, instead of your team checking every single one.
- Handwriting. Printed and digital documents read very well. Handwritten fields are not accurate enough for me to promise anything, so I do not.
- Fully automatic account coding. The system can suggest a category, but someone in accounting should confirm it before it is posted.
- A company behind me. I work alone, there is no support rota and no security certification on the wall. If you need 24/7 cover or a signed enterprise SLA, I am not the right person for it.
Questions about this package
How many sample documents do you need?
Ten to twenty, and real ones. The messier the better: crooked photos, creases, a signature across the total. I want the hard cases on day one, not at handover. If you send me only your cleanest documents I will estimate the accuracy too high, and you are the one who pays for that.
Can it read handwriting?
Not accurately enough for me to guarantee it. Printed documents read very well. Fast handwriting still fails at a rate that makes the output unsafe to post into accounts. If your source is handwritten forms, I will propose data entry inside the system instead, which is both more accurate and cheaper.
These documents are confidential. Is this safe?
The system runs in your own accounts, nothing stays with me, and I can set the originals to be deleted automatically once they are read. Security work is my main line anyway, and the Approve package ships with a written security review of the system being handed over.
How would I know if it read something wrong?
Each document carries a confidence level. Low confidence gets pushed to a person, high confidence goes straight through, so the review effort lands where it belongs. In the Check package the system also does the arithmetic, and anything that does not add up is raised even when the system was confident.
Can it write into the sheet we already use?
Yes, into the sheet your team already opens daily, as long as the column headers have settled. If they have not, I will help settle them once before we start, because changing headers later means changing both sides.