Software engineering from Munich · since 2010
Automate business processes with AI
The work in your processes takes minutes. The process takes days. We build the automation that closes that gap: inside the systems you already run, in production. Not as a pilot.
13 minutes of work. Four to six days of cycle time.
More than 98% of the cycle time is waiting. That is where the leverage is, and it is usually not where digitisation projects aim. A new portal makes the typing prettier. It does not make the typing unnecessary, and it does not get the invoice out of the inbox. The figures above are illustrative, not measurements from a client project. We measure your actual numbers in the first step.
You get an assessment either way, including “this one is not worth it.”
Progress Software partner (NASDAQ: PRGS) since 2019. Working with Rohde & Schwarz, H&Z Management Consulting, CAST Software.
Use cases
What we automate
Five patterns that show up in almost every mid-sized company, and where the waiting time comes from.
Documents
Documents somebody retypes
Invoices, delivery notes, purchase orders, inspection reports, claims, contracts. An AI system reads them (including poor scans and layouts it has never seen), matches them to the right record, and hands the data to your ERP or DMS.
You keep the approval. What goes away is the retyping and the searching.
Inbox
Shared inboxes a process depends on
orders@, service@, claims@. These are the entry point of a business process but get treated like a mailbox. An AI system classifies every incoming message, extracts the relevant data, creates the record in the target system and drafts the reply.
The record exists the moment the mail arrives, not the next time someone opens the inbox.
Review
Review and approval steps
Plausibility checks, four-eyes review, quality control, deviation detection. Where a person today checks records or images against rules and experience, a model does the triage: the unambiguous cases pass through, the unusual ones land on a desk, with a stated reason.
We built exactly this for a German mechanical engineering company: an AI system that detects production defects during manufacturing control.
Reporting
Recurring reports and forecasts
Monthly reports, forecasts, capacity planning, anomaly detection in time series. Reports that someone assembles by hand from several systems get produced automatically, and raise a flag when something falls out of range, instead of waiting for someone to look.
Knowledge
Knowledge sitting in folders and in people’s heads
Manuals, proposals, project documentation, standards, maintenance records. An assistant that works strictly on your own documents answers questions with citations, traceable to the page.
Runs entirely inside your infrastructure if you need it to.
How we work
Four steps, in this order
Each step decides whether the next one is worth taking. That is why the order matters, and why you can stop after step 1 and after step 2.
01
1 day · free
Process screening
We look at two or three of your processes and work out where automation pays and where it does not. You get an assessment either way, including “this one is not worth it.”
02
1–2 weeks
Feasibility and data check
We check whether your data holds up: volume, quality, access, interfaces. The output is a defensible statement about the achievable effect and the cost, before development, not after.
03
4–6 weeks
Prototype on the real process
The prototype runs on your real data and inside the real workflow, alongside the existing process. You see the hit rate on your cases, not on a demo.
04
ongoing
Production
Integration into ERP, DMS or line-of-business application, with monitoring, an escalation path and tests. After that you run it or we do. The source code is yours.
↑ Exit points after 01 and after 02.
Difference
Why an engineering firm rather than an AI consultancy
AI consultancies are plentiful. What is rare is someone who then integrates the result into a grown ERP, and answers for it when it does not run at seven on a Monday morning.
Built for operation
We have been building enterprise software that runs in production for more than 15 years, with tests, monitoring and load behaviour treated as part of the work rather than an afterthought.
Into the systems you already have
Your ERP stays. Your DMS stays. We build components that fit into existing systems instead of rebuilding processes around a new tool.
One team for the model and the software
ML engineers, backend developers, database architects and DevOps under one roof. There is no handover gap between the model and the production system.
Traceable rather than impressive
Systems whose decisions can be audited: confidence scores, stated reasons, and a defined path for a human to step in. In regulated industries that is not optional.
Cloud is optional
If your data cannot leave the building, the models run on your hardware. We have optimised models down to microcontrollers.
Evidence
What came out of it
9×
Response time & throughput
Performance audit of an existing web application.
Medical Protection · Medicine & insurance
48 h
Data migration
Automated migration of XML content, implemented with PLINQ.
California Southern University · Education
AI
Manufacturing control
Detection of production defects during the manufacturing process.
German mechanical engineering company
R&S
Project management platform
Domain-driven design, full test coverage, performance across every layer.
Rohde & Schwarz · Electrical engineering
“We were advised very competently and in detail on the technical possibilities and their implementation strategies.”
Dr. Ann-Kristin Baum · Immotech Austria GmbH
Questions
What you are probably wondering
Our data is not clean enough for AI. Is it even worth it?
That is the most common concern and usually the wrong one. Current document and text processing handles inconsistent data far better than the previous generation: an invoice layout does not have to be known in advance to be read. Whether your data holds up is settled in step 02, before development costs start. If it does not hold up, we say so.
Does our data have to go to the cloud?
No. We run models entirely inside your infrastructure or in an EU data centre. Which option makes sense depends on the process, for personal or competitively sensitive data we normally plan without external cloud services.
What does the EU AI Act mean for us?
Since 2 August 2026 the transparency obligations under Article 50 apply: AI systems people interact with directly must be recognisable as AI, and AI-generated content must be labelled. Obligations for standalone high-risk systems under Annex III (including recruitment, credit scoring and biometrics) were postponed to 2 December 2027. Most back-office automation falls into neither category. Classifying your specific case is part of step 02. We provide technical classification, not legal advice.
How long until something runs in production?
Six to eight weeks from screening to a prototype on the real process. Time to production depends on how accessible your target systems are. In our experience that, not the model, is what determines the schedule.
What happens when the AI gets something wrong?
That is a design question, not a residual-risk question. Every automation we build has a defined boundary: below a confidence threshold it does not decide, it escalates. The human stays at the point in the process where their judgement matters, and only there.
Do we have to replace our ERP?
No. We build components that attach to existing systems. Replacing a grown ERP is a project in its own right, not a side effect of automation.
What does it cost?
The screening is free. The feasibility check is commissioned as a bounded package. For development we quote a price only after seeing your data and your target systems. Any number before that would be a guess.
What if we end the engagement?
The source code is yours, and so is the documentation. We hand over to your team or to a third party. We do not build systems only we can operate.
Next step
One day to find out whether it is worth it
Name two or three processes that bother you. We will look at them and tell you where automation helps, and where it does not. The screening costs you nothing but the meeting.