AI Traction
DE/EN Book 30 minutes

Every business has important work it keeps not doing.

We make it easier and faster, with AI automation and systems that guide people through the work.

Book 30 minutes

First 30 minutes are on us.

Be honest

You probably have at least one thing in mind right away. What is it?

It stays in your browser. If you book a call, it goes into the booking notes, so we can start right there.

Finally done.

Every one of them, and every decision in there was still made by a person.

What it can look like

Three builds, so you can picture it.

One is being built right now. Two are what it looks like at your size. In every one, the part that matters stays with a person.

As a mid-size company

The offer that took two weeks

A machine builder with sixty people gets an enquiry. The offer takes two weeks. Nobody is slow. The person writing it has to dig out the last five similar offers, check the price list and ask production about lead times. Only then does the actual thinking start. Meanwhile the customer has two other offers on the table.

What it looks like instead: the enquiry comes in, and the next morning a draft is waiting. It's built from the enquiry and the closest previous offers, with prices from the company's own list. The open questions are marked. The person who used to spend two days assembling now spends an hour checking. The offer goes out in two days.

What stays humanThe price, the terms, and the sentence at the top that shows the customer they were understood.

As an enterprise

The brand that forty agencies work for

A consumer brand with a few hundred people has forty agencies and subcontractors producing material in its name. The brand manual runs to eighty pages. Honestly, nobody outside the building has read it. So before every campaign someone in marketing spends days checking logos, colours and legal lines by eye. Things still slip through, and the ones that slip through are the ones that get noticed.

What it looks like instead: everything produced in the brand's name goes through one check before release. The system has read the manual, so nobody else has to. It flags what's off and shows the rule it's off against, down to the paragraph. Marketing reviews a list of flags in an hour instead of a stack of material in days. And the agencies get the same answer every time.

What stays humanWhat the brand stands for, and every judgement call the manual doesn't cover.

As a coach

The newsletter that finally goes out

She runs a small coaching business and writes a newsletter her readers actually like. It goes out every two or three months instead of every few weeks. The ideas come on long drives and in the shower, and by the time she sits down to write, they're gone. On one thirteen-hour drive through France she had ten ideas for posts. Her phone's assistant couldn't cope with her mix of German and English. Not one of them survived the trip.

What we're building for her: she speaks an idea into her phone, and it lands sorted in her Notion, in her own words. When she has an hour to write, the system lays out her last twenty ideas and asks four or five questions about the one she picks. It doesn't write anything for her. What comes out is her outline and her sentences. The newsletter goes out on the date she names.

What stays humanEvery word a reader sees. The system sorts and asks. It never writes a line that leaves the house.

How we work

First we analyse it. Then we build it. Then we keep improving it.

The order matters. Without the first step, every estimate for the second one is a guess. And the third step is probably why it's still running a year later.

1

Analyse

It starts with a conversation. Thirty minutes on where the work falls behind and whether it's worth touching at all. If it is, we go through your business and its processes properly. Then we come back with what to build, what it takes, and what we'd leave alone.

The analysis is credited in full against the build. And if we find nothing worth doing, you don't pay for the looking.

2

Build

On your infrastructure, with the people who actually do the work involved before anything gets built. A first version running in weeks, not quarters.

3

Improve

Systems change, inputs change, models change. And the first version is never the best one. So we keep what's running running, and we make it a little better every month. Usually the second process comes out of that. Then the third.

About us

Who's behind it.

One founder, and a circle of AI freelancers we bring in when a build needs more hands than one.

What we believe

The work worth automating is the work someone already does by hand. Badly, reluctantly, not often enough. That reluctance is the proof. Nobody hates a task that doesn't matter. So we don't go looking for things AI could do. We go looking for the things you already know you should be doing and keep not doing.

The part that matters stays with a person. We've seen what happens when it doesn't. Someone's personal messages turned into machine-written welcome posts. A brand's voice flattened into everybody's voice. So in every build there is a line the machine doesn't cross, and we draw it before we write a line of code. The system carries the work to a person. It doesn't replace them.

It runs on your ground. Your server, your accounts, your AI subscription. If we disappeared tomorrow, everything would keep running and you'd have the keys. That's also why we can say honestly what leaves your building and what doesn't.

Small first, and we say the number early. The thing is, most of these projects don't fail on the technology. They fail on the size of the first step. So the first version is deliberately small. The price comes up in the first conversation, not in a proposal two weeks later. And before we build something big, we look together at whether that's the place where it pays off most.

So

What important work often doesn't get done in your business?

Thirty minutes, and you know whether it's worth touching. First 30 minutes are on us.

Book 30 minutes
Not ready to talk? Write to us and we'll send you the 12 questions to ask before you build anything with AI.

Fair questions

What people usually ask first.

“We tried AI already. Nobody used it.”
We hear this a lot, and it's almost never the technology. Usually the thing was built for a process nobody had actually agreed on. Or the people doing the work saw it for the first time at launch. So we start with those people, before anything gets built. And we keep the first version small enough that changing it is cheap. If it turns out nobody wants it, we find that out in week two.
“Will the AI write in our name?”
No. AI is to your voice what the messenger is to the message. It carries it, it doesn't replace it. The systems sort, draft, ask and route. Whatever a customer, a reader or a colleague actually sees still comes from a person. That's a design rule in every build, not a setting someone can flip.
“What about our data?”
Whatever mustn't leave your building doesn't. The systems run on infrastructure you own, connected to your own accounts. Where an external model is fine and where it isn't, we settle in writing before anything gets built. If your case doesn't allow it, you'll hear that on the call.
“We'd have to change our systems.”
No. A first build attaches to what you already run. Replacing systems is a different project, probably a worse one, and for sure not the place to start.
“We're too small for this.”
Small is actually the easier case. Fewer people to align, shorter decisions, and the work that falls behind is easier to spot. Most of what we build lands in companies between ten and a hundred people.
“How long until something runs?”
Usually two to three weeks for a first build, once the analysis is done. The first version is deliberately small. It's easier to make a small thing bigger than to make a big thing used.
“What does it cost to run afterwards?”
Your own server and your own AI account. Typically that's somewhere between a few euros and a few tens of euros a month. You pay those directly, nothing runs through us. The improvement work itself we size on the call, once we know what's running.