AI for companies
AI automation that takes over real work.
Most AI projects run aground at the same point: an impressive demo that touches none of your real systems. We build AI into the workflow itself, on your data, connected to what you already run.
Where AI earns its keep
AI is good at work that is too varied for a script and too dull for a person. That is a narrower category than the sales pitches suggest, but within it the difference is large.
- Reading documents and turning them into structured data: purchase invoices, packing slips, contracts, supplier quotes.
- Sorting, summarising and routing incoming messages to the right person or the right system.
- Drafting text that a person then approves, rather than sending blind.
- Searching your own documentation and answering with the source attached, so somebody can check it.
- Flagging exceptions in flows that normally run automatically, so a person only looks at the odd cases.
Where it goes wrong
AI disconnected from your systems is a demo. The work is not in the model, it is in the connection: where does the data come from, where does the answer go, and what happens when the model is wrong.
So we always build AI inside a process with a clear boundary. What can be automatic is automatic. What needs judgement lands with a person, with the context attached so they can decide in seconds rather than work it out again.
And we measure it. If an AI step does not demonstrably save time or prevent errors, it should not be there. That is an uncomfortable principle for a supplier, but it saves you an expensive disappointment.
AI agents, without the hype
An agent is nothing mysterious: it is AI allowed to take a series of steps rather than give one answer, with access to your systems and limits on what it may do alone.
That works well for tasks with a clear goal and a checkable outcome. Reading a purchase invoice, finding the supplier, matching the lines against the order and reporting the difference is a fine agent task. Wandering unrestricted through your administration is not.
The skill is not making the agent do as much as possible, but deciding exactly where it should stop and ask.
How we approach it
We do not start with the technology but with the question of what work actually gets done today and how long it takes. Almost always one process comes out that costs the most and returns the least.
We take that one first, put it live, and only then look further. Weeks, not quarters. That way you know after the first release whether this works for your company, before you attach a large budget to it.
Frequently asked questions
- Does this work for a small company?
- Especially. In a small company, one person losing two days a week to manual work is immediately a large share of your capacity. The return is often more visible than at a large company, because you feel the difference directly.
- What happens to our data?
- We decide that up front, not afterwards. Which data the model sees, where it is processed and what is retained is part of the design. For sensitive flows we can set things up so data stays within Europe and is not used to train models.
- Do our systems need to be in order first?
- No, but it helps to know where the data lives. In practice, cleaning up the data flow is often half the gain, quite apart from the AI. If that is the case we say so, rather than layering an AI project on top.
- What does an AI project cost?
- A first scoped process usually sits in the same band as an integration: under 10,000 euro. If it becomes an agent working across several systems with error handling and review, it moves towards 10,000 to 25,000 euro.
Which work would you want gone?
Tell us which process costs the most time right now. We will say honestly whether AI is the answer there, or whether ordinary automation gets it done faster and cheaper.