Automatic document reading with AI
Invoices, freight bills (CT-e), contracts and receipts arrive by email or upload, and the AI extracts the data, checks it and posts it to the ERP. The team only reviews what the AI flags as doubtful.
How it works today, without automation.
Dozens of PDFs arrive by email every day. Someone opens them one by one, copies the tax ID, amount, due date and items into the ERP, and checks again at month-end, when the books don't add up.
- 01Hours every week spent typing what is already written in the document.
- 02A typo in an amount or due date turns into a fine, interest or a wrong charge.
- 03Documents sit in the inbox waiting for someone to have time.
- 04Nobody knows for sure which documents have already been posted.
The flow, step by step.
-
01
Document arrives
By email, upload, shared folder or WhatsApp.
-
02
AI reads and extracts
Identifies the document type and extracts the fields: tax ID, amounts, dates, items.
-
03
Checks the data
Compares it with the order, the supplier and the company's rules.
-
04
Posts to the ERP
Creates the entry, the payable or the right record in the system.
-
05
A person reviews exceptions
Only what doesn't match goes to review, with the reason shown.
What would be delivered.
Deliverables
- Reading of invoices, freight bills (CT-e), contracts, bank slips and receipts.
- Checking rules against orders, suppliers and amounts.
- Automatic posting to the ERP or finance system.
- Review queue for exceptions and a history of every document read.
- Google Drive or OneDrive
- ERP
- Finance
- SEFAZ (NF-e and CT-e XML)
Stages
-
Discovery2 to 3 days
Document types, required fields and checking rules.
-
Reading and rules2 to 3 weeks
Extraction models, checks and ERP integration.
-
Testing with real documents1 week
We run documents from recent months and measure accuracy.
-
Go-live and tuning1 week
Production launch with the review queue active.
The final timeline is set in the proposal, after discovery.
Questions about this use case.
Can the AI read scanned documents or photos?
Yes. Digital PDFs, scanned documents and legible photos. When the invoice XML exists, it is used first because it is more accurate.
What if the AI gets an amount wrong?
The checking rules compare the document with the order and the supplier. Anything that doesn't match is not posted: it goes to a person for review.
Does it work with my ERP?
It works with ERPs that have an API or file import. During discovery we confirm how to post to your system.
Which documents can be automated?
Invoices, freight bills (CT-e), bank slips, payment receipts, contracts, purchase orders and other documents with a similar structure. Freight bills can also be checked against the freight table and quote before posting.
Want to see how it would work in your company?
You explain the problem. We build the solution. A straight answer, no generic proposal.
See also.
AI customer service on WhatsApp
Answers customers 24 hours a day, checks orders in your system and hands over to a person when needed.
// 03Order and lead triage
Requests from several channels classified and sent to the right person.
// 04Administrative routines on autopilot
Bank reconciliation, registrations and collections done by AI, with human approval where it matters.
// 05Assistant for company documents
The team asks and the AI answers based on internal manuals and policies.
// 06Reports that build themselves
The weekly summary comes out ready, without anyone merging spreadsheets.