AI invoice processing & document data extraction
Stop typing invoices by hand. AIK reads PDFs and scans, pulls out the fields you need, validates them, and exports the result — on your own servers, with local or cloud AI.
How the pipeline works
- Collect — drop files in a folder, receive them by email, or upload them; photos and scanned PDFs included.
- Extract — an AI model reads each document (vision for scans and photos, OCR fallback) and returns structured fields: numbers, dates, totals, line items.
- Validate — required fields, min/max, formats, math checks (line items vs. total, VAT-aware), anomaly and new-vendor flags, even PDF signature checks.
- Export & act — write to Excel, CSV or JSON, post to your API or ERP, insert into a database, or email a report with the file attached.
Invoice data extraction software you run yourself
Most invoice data extraction software sends every document to someone else's cloud. AIK does the same job — invoice number, dates, supplier, VAT, totals, line items — on your own server if you want, with a local AI model, and checks the result before it reaches your books.
Accurate, and it tells you when it isn't
Extraction is only useful if you can trust it. AIK flags records that need a human look — a total that doesn't add up, a vendor it has never seen, an amount far from the usual range — instead of silently pushing bad data downstream. Approvals can pause a run until someone signs off.
What it checks on every invoice
- The sums — line items against the total, and VAT against the base, so a wrong total is caught before it reaches your books.
- A changed bank account — when a supplier's IBAN differs from their last accepted invoice, the invoice is held. A swapped IBAN is the most common invoice fraud.
- Fields the AI is unsure of — the model marks what it could not read with confidence; those invoices wait for a person and the doubtful fields are highlighted.
- A supplier you have never paid and an amount far from the usual — flagged, not passed through.
- Bulgarian specifics — VAT rates 20 / 9 / 0 %, the ЕИК and VAT number format, and a look-up in the Commercial Register.
What you get in Excel
One row per invoice, with these columns: file, supplier, ЕИК, invoice number, date, net amount, VAT, total, IBAN, document kind (invoice, credit note or debit note) and the invoice a note corrects. If you want other columns or another order, say so in plain words and the sheet follows your layout. Your accounting software imports from that file.
Credit and debit notes
A credit note reduces what you owe, so its amounts are written with a minus, even when the document prints them without one. That keeps the sheet's totals right. The VAT and sum checks work on notes too. A note that does not say which invoice it corrects stays out of the sheet and waits for you, with the reason next to it.
Questions people ask
Which files can it read?
Text PDFs, scanned PDFs and photos taken with a phone. Scans and photos are read as images by a vision model, with OCR as a fallback.
Does it work with Bulgarian invoices?
Yes — Cyrillic text, Bulgarian VAT rates, the ЕИК and VAT number checks, and a Commercial Register look-up. The same pipeline reads English and other languages.
How accurate is the extraction?
Accuracy depends on the model and the document quality, so AIK does not ask you to trust it blindly: the sums are checked, fields the model is unsure of are marked, and anything doubtful waits for a person instead of going to your books.
Can it catch invoice fraud?
It catches the most common kind: a changed bank account. An invoice whose IBAN differs from that supplier's last accepted one is held until someone confirms it. New suppliers are flagged too.
Does it read credit notes?
Yes. The sheet says whether each document is an invoice, a credit note or a debit note, and which invoice a note corrects. A credit note's amounts carry a minus, and a note without the original invoice number waits for you to check it.
Where does the extracted data go?
Excel, CSV or JSON files, your own database, your ERP or any system with an API, or an e-mail with the report attached — one or several at once.
Do our invoices leave our servers?
Not if you choose a local AI model — AIK can run entirely on your own server. With a cloud model, only the document text or image is sent to that provider, and personal data such as IBANs and e-mails in the text can be masked first — on a scan or photo they are covered with a box before it is sent (a value too blurry to read stays visible).
Do I need to program anything?
No. You list the fields you want and tick the checks on the Document Types page; AIK builds the rest. Invoices come ready-made.
Can I test it on our own invoices first?
Yes. You can run it on a few of your own files and compare AI models side by side on them before you switch anything on.
Step by step, for an accountant: How to extract invoice data from PDF to Excel automatically (no code).
Keep your financial data in-house
Invoices are sensitive. With AIK you can run the whole thing self-hosted and even use local AI models (via Ollama), so documents and extracted data never leave your network — ideal for finance teams under GDPR. Read about on-premise deployment →
Intelligent document processing and invoice OCR — and AP automation for small business
Invoice OCR software turns a picture into text; intelligent document processing understands which text is the invoice number, the supplier and the total, and checks them. AIK does both in one pipeline, so a small business gets AP automation — capture, checks, approval, export — without an enterprise project.
Runs reliably, day after day
Schedule it to process new files automatically. Each tool runs isolated with its own timeout, health checks confirm everything works after an update, and safe rollback (app + database) means a bad change is one click away from undone.