Data extraction with AI
PDFs, scans, e-mails and websites become usable data
Invoices, delivery notes, contracts, forms, product catalogues: valuable information sits in documents nobody wants to type up. AI-assisted extraction pulls it out reliably and hands it to your systems in a structured form.
Classic text recognition (OCR) delivers characters but no understanding. Modern language and vision models, on the other hand, recognise what an invoice number is, which line item belongs to which price and whether a contract contains a notice period, even with documents that look different every time.
I build extraction pipelines that read documents from mailboxes, scanners, folders or websites, recognise the relevant fields with AI, validate them against rules and write them to your target systems. Uncertain recognitions are flagged and submitted for review.
What you get
-
No more typing
Receipts, forms and lists are captured automatically. Your team only reviews instead of entering.
-
Works with changing layouts
Unlike template-based recognition, AI copes with hundreds of supplier formats without setting up each one individually.
-
Validation included
Totals, tax rates, mandatory fields and duplicates are checked automatically. Anomalies go to a review list.
-
Straight into your systems
Export to DATEV, Lexware, ERP, CRM, Excel or database, the way your processes need it.
Understandable real-world examples
Capturing delivery notes and weighing slips
- Starting point
- A building materials dealer receives dozens of hand-annotated delivery notes every day. Quantities are typed up in the evening, errors only surface at invoicing.
- Solution
- Photos or scans of the notes are read by AI (including handwritten quantities), matched against purchase orders and posted to the inventory system. Deviations are reported immediately.
- Result
- Captured the same day, deviations detected before invoicing, no more evening shift for data entry.
Making contracts searchable
- Starting point
- A property management company has 800 lease agreements as PDF scans. Questions like "Which contracts expire in 2027?" mean days of searching.
- Solution
- All contracts are run through an AI extraction once: parties, term, notice periods, rent, index clauses. The result is stored in a database with a search interface.
- Result
- Reports in seconds, deadlines monitored automatically, the archive is usable for the first time.
Collecting prices from supplier websites
- Starting point
- A buyer compares prices from twelve suppliers every week, whose websites and price lists are all structured differently.
- Solution
- A collector fetches the price lists and websites, an AI maps the items to the company's own item numbers and writes the prices into a comparison table. Unusual jumps are highlighted.
- Result
- Price comparison automatically every Monday, the buyer negotiates instead of searching.
How we work together
-
1
Free initial consultation
30 minutes by phone, video or on site in and around Eisenach. You describe what costs time or causes friction today. I tell you honestly whether and how AI can help.
-
2
Analysis and concept
I look at the workflow, the data and the systems involved. You receive a short concept with effort, benefit and risks in plain language.
-
3
Implementation in small steps
First a working prototype, then the build-out. You see results early and keep control over budget and direction.
-
4
Rollout and support
Training for your team, documentation, monitoring. On request I take care of maintenance and further development.
Frequently asked questions
- How accurate is the recognition?
- For legible documents field accuracy is typically above 95 percent. More important is that the system knows when it is uncertain: those cases are submitted for review so that no wrong data enter the system unnoticed.
- Does it work with handwriting?
- Modern models read legible handwriting reliably, for example quantities or signature fields. With hard-to-read handwriting the hit rate drops and the review list takes over.
- Are we allowed to read websites automatically?
- That depends on content, terms of use and purpose. For price lists of your own suppliers with their consent it is unproblematic. I point out legal limits and implement technically considerate solutions.
- Where are the documents processed?
- Depending on sensitivity, in the EU with providers under a data processing agreement or locally on your server with open-source models. Personal data are only processed with an appropriate legal basis.
Related services
AI process automation and optimisation
Invoices, orders, enquiries, reports: much of it is still handled manually. I analyse your workflows, find the steps with the biggest leverage and automate them with a mix of classic software and AI models.
Learn moreMCP development: connecting AI assistants to your systems
The Model Context Protocol (MCP) is the open standard that lets AI assistants like Claude or ChatGPT access company systems: calendar, inventory, CRM, databases, line-of-business applications. I develop MCP servers that expose exactly the functions you want to expose.
Learn moreWebsite development with AI
A website has to convince two audiences today: people and machines. I build fast, maintainable websites with clean structure, optimised for search engines and prepared so that ChatGPT, Perplexity and friends recommend your business correctly.
Learn moreLet's talk about your use case
In a free 30-minute call we find out whether it is worth it and what a first step could look like.
On site for businesses in Eisenach and within 100 km, remote beyond that.