Automation studio · EU-hosted
We find the recurring handwork inside your processes, prove what it costs, and replace it with something that runs on its own. Scripts, workflows or language-model agents — whichever the problem actually needs.
Describe your process See how we work
01 — The problem
Exports get copied, numbers get reconciled, mails get retyped. No line item, no owner, no end date. It is simply what Tuesdays are.
Two hours a week from someone hired to think is roughly twelve working days a year — spent on a task that was never designed, only inherited.
The blocker is that nobody has written down what the process actually does, including the three exceptions everyone works around.
02 — How it works
Automating a process nobody has mapped is how you get an expensive version of the same mess. The analysis is not a formality before the real work — it is what makes the build short.
03 — What you get
Every step, who touches it, how long it takes, and which three exceptions break the happy path. Usually the first time it exists in writing.
Week 1–2What to automate, what to delete, what to leave alone — with the reasoning attached, so it survives you leaving the room.
Week 2Scripts, n8n flows or an agent, in your environment or an EU-hosted one. Version-controlled, with the credentials in your name.
Week 3–4Documentation, a runbook for when it fails, and a session with whoever owns it next. If we vanish, it keeps running.
Week 404 — Principles
You own the repository, the credentials and the infrastructure. Not: a platform fee that outlives the project
Data residency and processing basis are decided before the first line of code, not retrofitted after the works council asks. Not: a US API call quietly in the middle
A cron job and forty lines of Python is a valid answer. So is a language model. The process decides, not our licence agreement. Not: a language model because it looks good in the pitch
The point is that capable people work on the questions they were hired for. Not: a business case built on cutting posts
05 — Stack
Chosen because they are self-hostable, inspectable and replaceable.
06 — Selected work
A wholesaler with roughly 40 staff rebuilt the same sales report every Monday from four exports. We mapped it, found two of the four exports were redundant, and replaced the rest with a scheduled flow and one approval step.
Python · n8n · scheduled · EU-hosted
Incoming supplier invoices were typed into the ERP by hand. A locally hosted model extracts, a rule set validates against the order, and anything ambiguous goes to a human queue rather than through it.
Ollama · rules engine · on-premise
07 — Background
Leadership of business intelligence, data science and product management. Accountability for a company-wide AI rollout. Before that, pricing and sales strategy in consulting, and growth in D2C commerce.
We have had to defend automation work internally — against budget, against IT security, against the question of what happens to the hours it frees up. That is a different skill from building the solution. Both are needed.
08 — Contact
No calendar link. Write what the process is, and you will get a real answer rather than a discovery call.
hallo@scriptorix.ai