Service
AI in your business: where it takes work off people's hands.
AI is not a project of its own. It is an ingredient in processes you already have: reading documents, sorting incoming requests, drafting recurring replies. I look at where that saves time in your business, and I'll say just as clearly where it doesn't.
Honesty first
Most processes in a business don't get better through AI. They get better through clear rules, a decent form or an interface. That isn't modesty, it's experience: where the answer follows unambiguously from data, conventional software is more accurate, cheaper and easier to follow.
AI gets interesting where language and unstructured text are involved: a document nobody forced into a form; an email that has to be read before it's clear where it belongs. Those are the places I look for, rather than laying one tool over the whole business.
What you don't need for this: an AI strategy, a model of your own, or a data project up front. It starts with one concrete process that costs time today.
Where it applies
Where AI genuinely contributes in a business.
- Reading invoices and delivery notes: amounts, line items and customer data land in the right field instead of being typed in.
- Sorting incoming emails and forms: enquiry, complaint or order, and straight to the right place.
- Drafting recurring replies: your people check and send, instead of writing each one from scratch.
- Making free text from reports and notes analysable, for evaluations nobody could run before.
- Making your own material searchable: finding offers, contracts and documentation without knowing the exact wording.
- Preparing text in two languages, reviewed by your people before anything leaves the building.
How we find out whether it pays off
In the analysis phase we take the processes that eat the most time and test them against one simple question: is there text here that somebody has to read and classify? If so, it can be tried on real examples from your business before anything gets rebuilt.
One thing matters throughout: AI sometimes gets it wrong. So we decide up front what runs automatically and what a person confirms, and how you can tell that a result is uncertain. A suggestion your people review is often the better starting point than an automation nobody trusts.
Your data
The first question in any AI project is not which model gets used, but which data is allowed to leave the building at all. That is settled and documented beforehand, not sorted out afterwards.
Hosting in the EU, GDPR-compliant processing, documented data processing agreements. Where a process can work without external services, it runs without them. In case of doubt, personal data stays where it belongs: with you.
Whether an application falls under the EU AI Act, and what follows from that, is something we look at together before implementation.
Which process eats the most time at your place?
A 30-minute initial consultation. I'll say honestly whether AI contributes anything there or whether a simpler solution gets you further.