Invoice and receipt data extraction into Pohoda, Money and KROS.

02 / 08

Who extraction is for: not just accounting firms

Data extraction pays off for any company where someone regularly retypes invoice data into accounting software or collects and sorts documents for an external accountant. Accounting and tax firms that process documents for their clients use it too, often together with a client portal or a custom CRM. Here are examples from other industries:

• Manufacturing: invoices for materials and energy from regular suppliers, where each cost has to be assigned to a job or cost center.
• Wholesale and warehousing: invoices from goods suppliers that need to be checked against the purchase order and goods receipt before they're paid.
• Service businesses: technicians' receipts for spare parts and fuel, which today arrive in an envelope once a month.
• Logistics: foreign invoices for tolls, fuel and vehicle repairs in different currencies and languages.
• Online stores: invoices from foreign suppliers and services, such as advertising or subscriptions, that only arrive by email.

03 / 08

Which documents the system reads

The system reads PDFs, phone photos and scans. Documents can be sent to a single email address, such as invoices@yourcompany.com, or employees can photograph them on their phone or upload them to a folder on Google Drive or SharePoint. Each type of document needs different care:

• Incoming PDF invoices: supplier, company ID, VAT ID, invoice number, payment reference, issue, delivery and due dates, IBAN, line items, and the tax base and VAT for each rate.
• Receipts: seller, date, amount and VAT, even from faded or crumpled paper. On Slovak e-kasa receipts, the system also reads the QR code printed on each one.
• Foreign invoices: a different language, a different currency and often no VAT, with a reverse charge note instead. The system records the currency and exchange rate according to the rule your accountant sets, and checks an EU supplier's VAT ID in the VIES system.
• Credit notes and advance invoices: the system has to tell them apart from regular invoices, otherwise the balance owed to the supplier wouldn't add up.
• E-invoices in XML: in Slovakia, from 1 January 2027, VAT-registered businesses will have to send invoices to other Slovak businesses as structured XML. Those, like other structured e-invoices, don't need extracting; the system reads them directly.

04 / 08

How I make sure the numbers add up

Every extracted document goes through checks that aren't done by the language model but by plain calculation in code, which always gives the same result for the same numbers. A document that fails any check stops and waits for a person:

• Totals: the line items must add up to the tax base, and the tax base plus VAT to the amount due.
• VAT by rate: for each rate, the tax is recalculated from the base and compared with the amount on the document. An invoice with two rates has to match on both.
• Rounding: suppliers round VAT per line or for the whole document, and the system knows both methods. It accepts a one-cent difference caused by rounding and records it as rounding, but nothing bigger.
• Duplicates: the same supplier with the same invoice number, the same file sent twice, or the same invoice arriving from the supplier and again in a forwarded email. Pohoda also has its own duplicate check for XML imports, which I turn on as a second safeguard.
• Company ID in the register: the system verifies a Slovak supplier by its company ID in the Register of Legal Entities, and a Czech one in ARES, so the name and address are right. A new supplier gets a new entry in the address book instead of someone fixing a typo in an old one.
• IBAN: when a known supplier suddenly sends an invoice with a different bank account, a person always reviews it. A changed account number on an invoice is a well-known fraud technique.
• Dates: the due date can't be earlier than the issue date, and a document from last year won't quietly end up in the current period.

05 / 08

A person reviews unclear documents, and the system learns from the corrections

Unclear documents appear on a simple screen: the invoice on the left exactly as it arrived, the extracted data on the right. For each field you can see where on the document the system read it, and any field that failed a check is highlighted. You correct it, confirm, and the document goes on to your accounting software.

Corrections turn into rules. If you change the account coding or cost center for your electricity supplier the same way three times, the system will suggest it on its own next time. The rule only takes effect once you confirm it, and you can switch it off at any time. So the language model never gets a free hand: it suggests, the rules check, and a person decides.

I can also add rules that only your company has, such as assigning an invoice to a job by its purchase order number, checking it against the goods receipt in your warehouse, or requiring a manager's approval for costs above a certain amount.

06 / 08

Import into Pohoda, Money, KROS and ABRA Flexi

I pass each verified document to your accounting software by whatever route your software and its license offer. The original document stays in the archive, and for every record you can see where and when it came from. If the software runs on an office computer, the system connects through your network or a VPN. How I safeguard such a connection so nothing gets lost during an outage is described on the system integration page.

• Pohoda: XML import into the invoices agenda, run automatically or through POHODA mServer. For every import, Pohoda returns a response with the result for each document, so the system knows which document went through and which needs fixing.
• Money S3: a paid API module through which the document passes an import queue. Money S4 and S5: an API module that runs on your machine.
• KROS, common in Slovakia: documents imported into OMEGA from XML, Excel or a text file.
• ABRA Flexi: a REST API with separately purchased access.
• Other software: if it has no API, I prepare an import file in the format it expects, and your accountant imports it themselves.

07 / 08

Doklado, Digitoo or a custom solution: which one when

If you have ordinary invoices, one company and a standard process, buy an off-the-shelf tool. It costs less than development and you can start straight away. Doklado is a digital archive and document processing service; according to the vendor, it automatically extracts the amount, VAT, date, supplier and payment reference and sends documents to software such as Pohoda, OMEGA or Money S3. Digitoo reads documents using AI and lists direct connections to Pohoda, Money S3, ABRA Flexi and ABRA Gen, among others. KROS Digitálna kancelária, aimed at Slovak accounting firms, also extracts invoice data.

If one of them covers your process, I'll tell you on the first call and there's nothing to build. A combination often works too: an off-the-shelf tool extracts the documents and I add only the missing piece, such as the check against your warehouse. A custom solution, or an add-on to an existing tool, pays off when at least one of these applies:

• Invoices need to be checked against a purchase order or goods receipt in your warehouse system before they're booked.
• Costs are split across jobs, cost centers or vehicles by rules only your company has.
• Documents arrive from several places at once: email, technicians' phones, a supplier portal.
• The data should also go into your own CRM, ERP or management reports, not just into accounting.
• You have many unusual documents, such as foreign toll or fuel card statements, and want your own checks for them.

08 / 08

A test on 30 to 50 of your documents, price and terms

Before going live, I test the extraction on 30 to 50 of your real documents. You pick a mix of what you actually receive: regular invoices, receipts, foreign documents, credit notes and bad scans. I compare the results with how your accountant booked them and show you every document where the system made a mistake or sent it for review. I adjust the checks and rules, repeat the test, and only then does extraction go live.

Extraction connected to one accounting program is a single process automation, a few days of work, and usually starts at €500. Custom rules, a connection to your warehouse or another system, and a review screen for several people add to the price. Running costs, meaning hosting, backups and language model calls, range from tens of euros to the low hundreds a month depending on the number of documents.

After the first free call, you get a specific price and delivery date, with the quote split into parts. Once you approve it or any part of it, I build a free prototype on your documents so you can check it will work before you make a final decision. You pay 20% when you order and the rest after handover. By default, I run the system on my own accounts for a monthly fee and always fix its bugs for free.

Invoices from sole traders and employees' receipts contain personal data, so before launch we sign a data processing agreement under Article 28 GDPR. I keep data in the EU wherever possible and call language models through business APIs that, under the providers' terms, don't use the data for training by default. If you want to automate other processes around your documents too, see business process automation or AI agents for business.