Documents and invoices
I connect your inbox, scanner and KSeF to your accounting system: every document is parsed, checked against named rules, approved by a person and posted. You keep a working pipeline, an error queue and a handover pack any other developer can pick up.
3 weeks
to the first flow in production
×4
cheaper per invoice - market benchmark
100%
of documents pass named rules
A typical day at a wholesaler: 287 documents in, 6 waiting for a person in the error queue, nothing posted without passing the rules.
Four situations accounting offices and wholesalers describe most often.
It looks the same in an accounting office and in a warehouse: someone opens a PDF and retypes the tax ID, the amounts and the line items into Comarch Optima or Subiekt. The market benchmark for manual processing is around 10.9 USD per invoice against 2.8 USD with automated capture. At 500 documents a month that gap shows up in people's calendars, not just in a cost sheet.
One customer sends a PDF, the next a photo of a delivery note, the third an Excel sheet, the fourth types the line items into the body of the email. A salesperson rekeys all of it and gets product codes and quantities wrong. The format changes the day a supplier upgrades their system.
KSeF is already mandatory: since February 2026 for the largest taxpayers and since April 2026 for the remaining VAT payers. Invoices from foreign suppliers, receipts, warehouse documents and orders never enter it and arrive exactly as before. Half the document flow stays outside the register.
A duplicated invoice, a bank account outside the VAT register, a price that is not the contracted one: all of it turns up on the last day. The clean-up runs over email because nothing shows which document is stuck and why. There is no record of who approved what either.
Not a deck and not a prototype. Things that stay with you after handover.
A mailbox in Google Workspace or Microsoft 365, a scanner folder, a network drive and the KSeF API. Attachments, email bodies and spreadsheets travel the same route. Every document gets an id and an archived copy before anything else happens to it.
From an invoice I pull the seller, the tax ID, the number, the dates, net, VAT, gross, the bank account and the line items. From an order: the customer, product codes, quantities and prices. The OCR model is picked per document type, because a faxed scan and a system-generated PDF are two different jobs.
Duplicate, missing tax ID, bank account outside the Polish VAT register (biała lista), price off the agreed list, quantity above what was ordered. Each rule has a name, so you can see which one stopped a document. Rules are described in plain language in the docs, so an accountant can read them without a developer.
Anything above the agreed limit, every new supplier and every document with a warning waits for a person. You approve in the panel or in Slack, in one click, with the original next to the extracted fields. Nothing moves on without that approval.
Comarch Optima and ERP XL, Subiekt GT and nexo, enova365, wFirma, ifirma, BaseLinker. Through the API where your licence allows it, through an import file where it does not. Sales invoices can go on to KSeF through its API.
A screen listing the documents that failed, with the reason and a retry button. Plus the repository, a flow diagram, a runbook and the list of credentials. You do not need my phone number for this to keep running next year.
Every stage ends with something you can check on your own documents.
You give me 40-60 real documents from the last month and access to a test database. I check what can be read automatically and what will always stay manual. The output is a fixed scope and a fixed price, before any contract.
I build the intake: a mailbox in Google Workspace or Microsoft 365, the scanner folder, the KSeF API. Documents are split into fields, the tax ID and bank account are checked, and everything is matched against your customer records. I show accuracy on your files, not on a demo.
I wire the write-back into Optima, Subiekt, enova365 or wFirma, through the API or an import file depending on what your licence allows. A gate sits above it: amounts over the limit and new suppliers wait for a human click.
The pipeline runs alongside the manual process for a few days and we compare results on the same documents. Then we switch over and you get the repository, the diagram, the runbook and every credential. From that day you can run it yourself or with me.
Packages and pricing
One stream of documents from intake to a posted record in your accounting system. Usually purchase invoices or orders from the inbox.
Accounting office: 300-800 purchase invoices a month.
Two to four streams plus write-back into the ERP: the document, the customer record, the warehouse document and the order.
Wholesaler: email orders, purchase invoices and BaseLinker.
Keeping a live pipeline healthy: monitoring, rule fixes and reacting to changes on the vendors' side.
Polish integrators publish roughly €500-9,000 a month to maintain builds like this, depending on scope.
Fixed net prices, agreed after I look at your real documents; in Poland I invoice in złoty (12,000 zł, 24,000 zł and 890 zł a month). The only separate cost is OCR and model usage on your own account.
I would rather say it before the contract than three weeks in.
If what you actually need is a store wired to stock and an ERP, that is a separate service: Services
Describe the task in two sentences. You get a fixed price, a timeline and a plan. If it isn't a fit for us, we say so straight away and point you elsewhere.
One document flow costs €2,800, several flows with ERP write-back cost €5,600, and care after launch is €210 a month. The price is fixed and quoted after I look at your real documents, before any contract. Polish integrators publish bands of roughly €19,000-58,000 for a pipeline that writes into an ERP, because they scope the whole company; I do one process end to end.
Yes, if your licence gives you an API or an import format. Comarch Optima and ERP XL, Subiekt GT and nexo, enova365, wFirma, ifirma and BaseLinker all have documented interfaces and I work with each of them. Before quoting I check on your own installation whether the write-back goes through the API or through an import file, because that changes the scope.
Because part of your cost documents will never pass through KSeF. Invoices from foreign suppliers, receipts, warehouse documents, contracts and orders still arrive by email as PDFs, scans or spreadsheets. KSeF tidies up domestic invoices; the rest of the flow stays on your desk, and that is the part I automate.
A person is, because nothing reaches the books without approval. Every document passes named rules: tax ID, VAT register, duplicate, match against the order. Anything above the agreed limit or from a new supplier waits for approval, and documents the pipeline does not understand land in the error queue with a reason attached. Every action is logged: who, when, and against which version of the file.
Three weeks from handing over sample documents to the first day in production. One week goes to extraction and validation, one to write-back and the approval gate, one to running in parallel with the manual process. With several flows and ERP write-back it is usually five to six weeks.
€210 a month for care, plus your own bill for OCR and the model. Vendor pricing is public: Google Document AI charges 1.50 USD per 1,000 pages of basic OCR, Azure Document Intelligence charges 10 USD per 1,000 pages on its prebuilt invoice model and 30 USD on a custom one. At 500 documents a month that is single-digit euros, not hundreds.
You are left with a working system and full documentation. The code sits in your repository, the pipeline runs on your infrastructure in the EU or in your own cloud account, and the keys and credentials are yours. The handover includes the flow diagram, the validation rules written out, and a runbook: what to do when a supplier changes a file format or an API stops responding.
What clients say
A few words from founders and teams we've shipped for.
Launched our store in 11 days — multilingual, fast, and already on page one for three of our main keywords.
The Telegram bot handles the first conversation with every lead now. My sales rep only steps in when the deal is hot.
Leads land in our CRM with every field already filled. No more copy-pasting from chats — the pipeline moves itself.
Shipped to the App Store and Google Play in three weeks. The handover docs let my team take it from there without me.
Set up our Google and Meta ads from zero. Cost per lead dropped by about a third in the first month.
He rebuilt the site and six weeks later ChatGPT was recommending us by name. Didn't know that was even possible.
Free 30-minute call. No sales pitch - just a real conversation about what you need and whether we're the right studio to build it.
No sales pitch — just a clear read on your scope.