CASE A — SALES
“A quote takes three days because the knowledge sits in a hundred past projects.”
Custom AI approach
An assistant that drafts reliable quotes from CRM, ERP and past proposals – using your calculation rules, not internet knowledge.
AI Workshop · 2 days · On-site or remote
The AI that moves your company forward is not for sale.
It has to be built. In two days we find out which one it is – and calculate honestly whether it pays off. No Copilot course. No slide marathon. A working format for companies that are serious about custom AI.
Max. 8 participants
Fixed price — known upfront
Results in writing
No follow-up obligation
Cleared up front
There are enough AI workshops. This one is built differently.
What you will not get here
- Copilot or ChatGPT introduction
- Prompt tips for better e-mails
- Slides about “the future of work”
- Use-case bingo without feasibility checks
- Tool sales with commission
What happens instead
- Your processes, your data, your systems as the starting point
- Every idea checked for feasibility by engineers on the spot
- Architecture, data protection and the EU AI Act considered from day one
- Effort and ROI quantified – derived from your numbers, not guessed
- A roadmap your team – or ours – can execute right away
What we work on
Three starting points we see all the time.
No reference-page prose – typical situations from mid-sized companies. Your case is different? That is exactly what day 1 is for.
CASE B — ADMINISTRATION
“Delivery notes, orders and receipts are typed into the ERP by hand.”
Custom AI approach
An extraction pipeline that reads incoming documents, validates them against master data and hands them to the ERP – with a review workflow instead of blind trust.
CASE C — SERVICE & KNOWLEDGE
“Thirty years of knowledge sits in folders, file shares and people’s heads.”
Custom AI approach
An internal assistant built on your own documentation, operated in your cloud environment – answers with sources, data stays in-house.
The agenda
Two days. No agenda item without a result.
Led by the architects and engineers who would build your solution afterwards. Which is why nothing gets promised here that cannot be built.
Day 1 — Understand & Discover 09:00–17:30
- 09:00 Goals & success criteriaHow will you measure, twelve months from now, whether this paid off?
- 09:45 Process deep diveWith the people who do the work – not over their heads.
- 11:30 Data & system landscapeWhat exists, what is missing, what is usable? ERP, CRM, DMS, file shares.
- 13:30 Use cases from daily workCollected from your bottlenecks, not from a catalog.
- 15:30 Feasibility checkOur engineers vet every idea immediately: data, interfaces, models.
- 16:30 PrioritizationBusiness value × feasibility. The day ends with a top candidate.
Day 2 — Decide & Plan 09:00–17:30
- 09:00 The top use case in detailProcess steps, data flows, interfaces, edge cases.
- 10:30 Solution architectureModels, integration, operations – build vs. buy weighed honestly.
- 13:00 Data protection & EU AI ActGDPR, risk class, governance. Operated in your cloud if you prefer.
- 14:30 Effort & ROIWhat does it cost, what does it return, when does it break even?
- 16:00 Roadmap & decision paperMilestones up to the prototype – as a document for management.
- 17:00 Go / no-go recommendationEven if it reads “no-go”. That is part of being honest.
The outcome
This is what ends up on the table.
Four documents your management can decide with – not “impulses”, not “food for thought”.
01 / USE-CASE PORTFOLIO
Every use case, scored by business value, feasibility, data situation and risk. Prioritized, not alphabetical.
02 / TARGET ARCHITECTURE
Models, data flows, interfaces and an operating model for the top use case – including the build-vs-buy assessment.
03 / ROI CALCULATION
Derived from your numbers: time spent, volumes, cost of errors. A basis for a real investment decision.
04 / ROADMAP
Milestones from prototype to operations, with efforts and responsibilities – executable with us or without us.
The toolbox
Technology-agnostic. Actually.
Which building blocks get used is decided by your use case – not by our portfolio. This is the toolbox we assemble your target architecture from in the workshop. Some of it comes from Germany, and much of it can run entirely inside your environment.
Platforms & Cloud
Where models run – in your cloud, on-premises if you prefer.
Azure OpenAI Service AWS Bedrock Google Vertex AI Hugging Face Groq fast inference Together AI model hosting Replicate model hosting
Frameworks & Orchestration
How models become applications.
LangChain LlamaIndex Haystack deepset · DE Semantic Kernel Microsoft CrewAI multi-agent AutoGen multi-agent DSPy prompt optimization Ollama local models vLLM inference serving
RAG & Vector Databases
Making your knowledge searchable – with sources.
Pinecone Weaviate Qdrant Chroma pgvector PostgreSQL Milvus
Evaluation & Observability
Making quality measurable before anything goes to production.
Langfuse Berlin · DE LangSmith Weights & Biases
A selection, not a commitment — the tool follows the use case, never the other way around.
Plain terms
Three promises you can hold us to.
01
Fixed price.
The price is set before the workshop. No day rates that multiply, no hidden costs.
02
An honest result.
If off-the-shelf software is enough or the use case does not pay off, that is exactly what the recommendation will say. We earn our money building things – not presenting slides.
03
No follow-up obligation.
Roadmap, architecture and documents are yours. Implement with us, internally, or with someone else.
Vendor-independent · No tool commissions · GDPR & EU AI Act built into the concept
Frequently Asked Questions
A fixed price you know before booking – depending on whether we work one or two days and whether on-site or remote. You will get it in the free first call. After that there are no further costs, and no obligation to implement with us.
A maximum of eight people: management or a division lead, two or three people with real process knowledge from the business, and someone from IT. That mix matters more than prior knowledge – we do not require any AI expertise.
Standard assistants make existing work somewhat faster, but they do not set you apart – anyone can buy them. Custom AI targets the processes that define your business: your data, your systems, your domain logic. And if the workshop shows that an off-the-shelf tool is enough for your case, that is exactly what our recommendation will say.
Very little. We hold a 30–45 minute call beforehand to agree on participants and focus areas. We do not need system access – it is enough that the right people are in the room and that processes and bottlenecks can be discussed openly.
You receive all documents in writing within one week. Then you decide: build the prototype with our development team (typically within four to eight weeks), implement internally, or park the topic with clear reasoning. Either way, the documents are yours.