AI automation for mid-sized businesses
We automate recurring processes for mid-sized businesses in Germany, Austria and Switzerland: receipts, requests, reports and data transfer between systems. Whatever the AI cannot decide safely goes to your team. Our office is on Schnellerstraße in Berlin-Schöneweide. There or on a video call, you talk directly to the developers who build your system.
What is AI automation?
AI automation means software takes over recurring processes in your business, and a language model only steps in where a text has to be understood, such as a request written in free form or a scanned document. Everything else is handled by fixed rules you can check.
Which processes we automate
AI automation for mid-sized businesses is usually about single processes that cost time every week: receipts, requests, reports and data copied between systems by hand. We build two kinds of solutions for that.
AI agents and custom AI systems
We build AI systems for your business: agents that qualify incoming requests, extract and classify documents, or handle a process across several systems. If a system should answer questions from your own documents, we use RAG (retrieval-augmented generation): the language model first looks up the relevant passages in your documents and then answers based on them. Every system is connected to your existing software and has fixed limits. How we build interfaces to your existing software is described on the Software development & integrations page. Law firms come with their own rules; the answer is in the common questions below.
Workflow automation with n8n
Our default tool for this is n8n. Anything that follows the same pattern can be handled by a machine: the report that used to be copied together, the data transfer between two systems, receipt processing, multilingual intake of requests. We cut the automation small and hand it over documented.
When a language model, when a fixed rule
A fixed rule gives the same result for the same input every time, and you can read why. A language model can’t offer that. So anything a rule can check, a rule checks.
When the system is unsure or a check fails, a person decides. This is called human-in-the-loop: the software handles the routine cases and passes everything else to your team, along with the reason it got stuck.
| Task | Approach | Who decides when in doubt |
|---|---|---|
| Read amount, date and invoice number from an XRechnung e-invoice | Fixed rule, the data is structured XML | Accounting, if a required field is missing |
| Read the same details from a PDF invoice or a scan | Text recognition (OCR) and a language model, then fixed validation rules | Accounting, if a check fails |
| Route a free-form request to the right contact | A language model classifies the request | Your team, if the model finds no clear match |
| Compile a monthly report from two systems | Fixed rule | The person responsible, if the numbers don’t match |
| Answer a question about your internal documents | Language model with RAG | Your team, before an answer goes out |
| Approve a payment | No automation, the software only prepares it | Always a person |
Which processes to start with
We usually start with the process that runs often and follows clear rules. Four typical candidates, roughly in order:
- Data transfer and reports: copying data from one system to another or compiling a monthly report. A fixed rule does this, no language model needed. If the numbers don’t match, the person responsible gets a message. A good first step, because the result is easy to check.
- Incoming invoices: fixed rules for e-invoices in XML format, text recognition and a language model for PDFs and scans, then fixed validation rules again. Unclear receipts go to accounting. The details are further down.
- Incoming requests: a language model reads free-form emails or forms and routes them. If it finds no clear match, the request goes to your team. A person writes or checks the reply to the customer.
- Questions about your own documents: a language model with RAG answers internal questions from manuals or contracts. This only pays off once the documents are kept up to date, and your team checks any answer before it goes out. More in our guide AI with your own company data (in German).
How receipt automation works
A schematic example for an incoming invoice, built with n8n. E-invoices in XRechnung or ZUGFeRD format are read straight from the XML. For a PDF invoice or a scan, text recognition (OCR) first makes the content readable, then a language model extracts supplier, date, amounts and tax rate. We adapt the steps to your software and your accounting, for example DATEV or Lexware Office.
For personal data in invoices we sign a data processing agreement under Art. 28 GDPR.
-
01 · Inbox
An email with an invoice arrives in a dedicated inbox, and an n8n workflow picks up the attachment.
-
02 · Read
An XRechnung is read directly, a PDF via text recognition (OCR) and a language model.
-
03 · Check
Fixed rules validate the required fields, check that net amount and tax add up, and flag invoice numbers seen before.
-
04 · Decide
If everything is clear, the invoice is booked into your accounting software. If anything is uncertain, it goes to your accounts team with the reason, and a person decides (human in the loop).
-
05 · Log
Every run is recorded so it can be traced later.
Where we deliberately don’t use AI
No language model decides on amounts that get booked, on releasing a payment or on changes to contracts. A model may read an invoice. Whether the amount is right is checked by a fixed rule, and a person releases the payment. Where a fixed rule can do the job on its own, we leave the model out: the rule is cheaper to run and can be checked.
What we don’t promise
We don’t promise miracles or an “AI transformation.” If a use case doesn’t hold up, we say so in the first conversation, before any money changes hands.
Further reading
These articles from our guides go deeper into topics on this page. They are in German.
- n8n and GDPR: self-host or cloud: where your data lives and what the contract has to say.
- Processing incoming invoices automatically: which steps fixed rules handle and where a person checks.
- AI with your own company data: how RAG works: when a language model can make good use of your documents.
- n8n, Make and Zapier compared: which tool fits your business.
How it works
-
Step 01
Before the project starts
First conversation
- · 15 minutes on the phone, on video or over coffee in Berlin, free of charge
- · We listen to what you are trying to do and tell you honestly whether we are the right people
- · If automation doesn't fit your case, you'll hear it here
-
Step 02
Before the project starts
Quote
- · We design the smallest solution that actually solves your problem
- · Scope, price and delivery date are in the quote before we start
-
Step 03
Within a few weeks
Build
- · The automation is cut small and connected to your existing software
- · For every process it is set in advance what happens when something goes wrong
- · Everything tested, with documentation a successor can read
-
Step 04
On handover day
Handover
- · Short training session and the finished documentation
- · Four weeks of free support on top
- · The code and the accounts belong to you
Who it fits, and who it does not
A good fit for:
- · Processes that always follow the same pattern: reports, data transfer between systems, receipt processing, intake of requests
- · You want a solution that is connected to your existing software and that belongs to you
- · You want to know scope and price before the project starts
Not a good fit for:
- · You expect miracle promises or an "AI transformation"
- · You want the software to move money or change contracts without a person signing off
Common questions
What does automating a process cost?
Scope and price are in the quote before we start. The first call is free.
What happens when the AI cannot decide something safely?
It goes to your team. For every process we decide what happens when something goes wrong, before it goes live.
Who owns the code and the accounts?
You do. We write the documentation so another developer can pick it up. If you ever stop needing us, you keep everything.
What happens after handover?
On handover day you get a short training session, the finished documentation and four weeks of free support. After that it's your call: a monthly support plan with us, or you run it yourselves.
Which processes are worth automating?
Anything that runs the same way every time: reports, receipts, requests, or data someone copies between two systems by hand.
Which tools do you work with?
Mostly n8n. We build your workflows with it, connect them to your software and look after them afterwards if you like. When a language model comes in is shown in the table further up.
Do you work as an n8n agency?
We build n8n workflows for your business and look after them afterwards if you like. n8n either runs self-hosted, on a server in the EU or on your own premises, or in the n8n cloud. n8n GmbH is based in Berlin, like us (see the n8n imprint). We are an independent service provider and not part of n8n GmbH. What speaks for self-hosting and what for the cloud is covered in our guide n8n and GDPR: self-host or cloud (in German).
Where does our data run?
In your projects, on servers in the EU by default, or on your own premises if you prefer. Which services are involved, including the language model, is listed before the start in the list of subprocessors attached to the data processing agreement under Art. 28 GDPR. You decide before anything goes live.
Do you also work for law firms?
Yes. Before we get access to any law firm data, we sign a confidentiality agreement in text form under § 43e BRAO (the German Federal Lawyers' Act). It explains that under § 203 (4) of the German Criminal Code, anyone assisting a lawyer who discloses a client secret without authorisation is personally liable to prosecution. That applies to everyone on our side who works on the project. We only see the data the task requires.
Who will we be talking to?
Directly with the developers who build your system. Our office is in Berlin-Schöneweide.
Do you only work for businesses in Berlin?
No. We work for companies in Germany, Austria and Switzerland, and most meetings happen on video. If you are in or near Berlin, you can visit us in Schöneweide, or we come to you.
Let's talk for 15 minutes.
Tell us where it hurts. We'll say whether we can fix it, and if not, who we'd ask instead.
The first call is free and you talk directly to the developers. We answer messages within four working hours.
Last updated: October 2026