AI · the alchemy

An AI agency in Berlin that does the maths first.

We build AI agents and document AI for mid-sized companies, with a person signing off. If plain automation fits your process better, that is what you get.

Not an oracle. A tool.Assessment · Agents · Automation

01 - What we do with AI

Four paths, one yardstick: value.

Every project starts with the same question: how many hours does it save - and what does running it cost?

Find use cases

AI consulting the foresight

We go through your processes and rate every use case by value, effort and risk. The result: a prioritised list - including the ideas you should drop.

  • Process analysis with your team
  • A business case per use case
  • A roadmap with priorities
Ask about consulting
Get tasks done

AI agents the helpers

Agents that understand tasks, plan steps and use your systems - CRM, ERP, email. With clear tool boundaries and a human who has the last word.

  • Access to your systems
  • Approval loops instead of flying blind
  • Traceable logs
Discuss agents
Language models in your product

Chatbots & LLM the voice

Assistants built on your company knowledge, support with access to order and product data, LLM features right inside your platform.

  • RAG on your own documents
  • Support automation
  • Claude, OpenAI or specialised models
Plan an assistant
One day, a clear roadmap

AI workshops the map

One day with your team: fundamentals without the hype, your processes put to the test, and at the end a roadmap you can defend yourselves.

  • On site or remote
  • For specialists and managers
  • A roadmap as the result
Ask about a workshop
02 - The honest maths

Does your agent pay off?

Not a matter of faith, just arithmetic: hours saved times hourly rate, minus running costs. Four inputs, a first direction.

Can the process be described exactly?
Is there clean data and a person responsible?
Our assessment An agent pays off.

Enough volume, enough judgement calls: this is where an agent plays to its strengths - with a human signing off.

Time saved per year ≈ 331 h
Gross savings ≈ €18,200
Running costs, agent − €6,000
Net per year ≈ €12,200
Run the real numbers

Rough figure based on assumptions: 46 working weeks, 45 to 80 % automatable depending on the process, running costs of €200 to €500 a month. Not advice - but an honest start.

03 - From production · entsorgo

AI invoice processing: from inbox to approval

A document pipeline reads and classifies business documents, checks them against approved rules and prepares the bookkeeping. People review and approve - nothing is booked automatically. Our guide to automated invoice processing describes what such a workflow looks like in detail.

  1. 01 AI reads and classifies
  2. 02 Rules check deterministically
  3. 03 A human approves
View the AI accounting case
Our stance

AI is not an oracle, it is a tool. We work out every use case in hours and euros - and say openly where it does not pay off yet, before budget goes up in smoke.

04 - Guardrails

Magic with railings.

01

A human has the last word

AI prepares, people approve. No architecture without an approval loop.

02

Rules first, AI second

Anything unambiguous runs deterministically - the language model only helps with the ambiguous cases.

03

Data protection by need

Model, hosting and storage follow your process - not the hype.

04

EU AI Act in view

We document risks and obligations as we go - before anyone asks.

05 - Frequent questions

What to know before your first prompt.

Do we even need AI?

Maybe not. If your process can be described exactly, classic automation is cheaper and more reliable. That is exactly what we tell you in the first call - before you burn budget.

Transparent pricing: our rate is 95 € per hour (plus VAT). An AI potential analysis with roadmap starts at 950 €, and a clearly scoped pilot for production use starts at 4,900 €. Assistants or agents with multiple integrations and broader operational requirements are quoted separately; they can fall into the mid five-figure range. The quoted project prices are starting prices; you receive a binding fixed-price quote after the first call.

The model follows the case: Anthropic Claude, OpenAI or specialised models - chosen by quality, cost and how sensitive your data is.

We choose hosting and storage by how sensitive your data is and document which data flows where. Sensitive content only leaves your environment when that is properly settled, contractually and technically.

A testable pilot with real data is usually ready after 2 to 4 weeks. From then on you decide on numbers, not slides.

06 - Plain talk on AI

Read this before it gets expensive.

All articles
07 - Quick inquiry

One step, 60 seconds.

Reply within 24 hours - with a clear assessment of your AI use case.

  1. We read every inquiry ourselves - no ticket system.
  2. Free first consultation, 30 minutes, no sales pressure.
  3. No newsletter, no spam.
Prefer to talk? +49 173 6618059

Tell us about your project

Budget range

We use your details to handle your inquiry. More in our privacy policy.

AI first call · 30 minutes

Got a use case in mind? Let us do the maths.

Your numbers, a clear answer - even if it is "don't".