Service · Data and analytics

Intelligent document processing

Since domestic invoices started flowing through KSeF, accounts teams receive them in structured form and no longer retype them. Plenty of paper remains, though: invoices from suppliers in Germany, Czechia or China, WZ delivery notes packed with shipments, CMR consignment notes, contracts and handover reports. We recognise these documents automatically, extract their data and stage it in your ERP for approval.

Volume
decides whether it pays
A person
approves every document
ERP
data goes straight in
Your layouts
used to train the model

Included in this service

Volume is what makes this worthwhile. A handful of pages a month is quicker to key in than any system is to maintain.

Talk the scope through with an engineer

Collecting documents

Attachments landing in a common mailbox, warehouse scanner output and mobile snapshots, even crooked or dim ones.

Recognising the type

Invoices, credit notes, WZ notes, CMRs and contracts are distinguished automatically and sent down separate paths.

Reading the fields

Number, dates, amounts, currency, the supplier's EU VAT ID, plus every line with its quantity. The model is trained on your suppliers' layouts.

Matching line items

Supplier product codes are mapped to your own codes and quantities checked against the purchase order, so the PZ receipt in the ERP is created without hunting.

Approval screen

The original and the extracted data side by side, with any fields the system is unsure about highlighted.

Posting to the ERP

Once approved, the document goes into Comarch ERP, Symfonia or enova365, and the original is archived with a link.

How we work together

A pilot covers one kind of document plus adaptation to your layouts, and we give its schedule after reviewing samples.

01

Counting documents

Monthly volumes per type and the staff hours currently lost to manual entry.

02

Sample set

Real documents in their full variety, the worst included: creased, stamped over the total, in three languages.

03

Tuning the model

Training happens on an EU-hosted service, with accuracy scored against documents held back for testing.

04

Go-live

We connect the document flow, train your team over Teams and monitor results closely for the first weeks.

Reading takes the least time; matching takes the most. Extracting from a WZ note that the supplier shipped 40 units of item “ART-2231/B” is trivial nowadays. The real work is making the system know that this is your code 10-557, packed in fours. We build that mapping table during the pilot, and it decides whether a colleague approves a document in seconds or in minutes.

Questions and answers

For invoices from Polish suppliers, OCR is no longer needed because the data arrives structured. What remains are foreign invoices, stock documents, consignment notes and contracts. In trading and manufacturing firms that is often more paper than the domestic invoices ever were.

Current models handle German, English, Czech and Italian well, as well as English-language invoices from Chinese suppliers. We measure accuracy separately per language and per major supplier, so it is clear where a colleague needs to look more carefully.

Contracts and handover reports contain names, signatures and sometimes PESEL numbers. We process them in an EU-region service under a data processing agreement and on terms that exclude training the provider's models on that data. The processing description is written so your data protection officer can add it straight to the record of processing activities.

There is no single cut-off. We count the hours warehouse and accounts staff spend retyping each month and compare them with the cost of building and running the system. A few documents a month rarely justify it; a steady, large stream often does.

Stop retyping documents

Describe the paperwork you receive and the monthly numbers. You will get a straight answer on whether automating it makes sense.

Hours
Mon-Fri 8:00-18:00 CET, reply within one working day
Meetings
Online via Teams or Google Meet

We set strictly necessary cookies only: they keep the site running and remember the city you chose. Nothing here is used for advertising or tracking. More in our privacy policy.