Only a couple of people know how your company really works.
And a lot of it is not written down.
Sound familiar?
Clear first, then clever.
What AI can actually do, once your business is written down.
Most companies that start with AI start with a tool. They buy a subscription to an AI service. Someone tries it for two weeks. Then it quietly stops.
The technology is rarely the problem. There was just nothing for it to work on.
AI can't do anything with knowledge that was never written down. If the way you work lives in four heads, a chat group and a spreadsheet nobody dares touch, every tool you put on top of it gives you the same thing. An answer that sounds convincing and is wrong.
So I don't start by building. I start by finding out what actually happens in your company, and I write it down.
Only then is it worth asking which parts should be made clever. And which parts should deliberately stay exactly as they are.
The part most people skip
A good part of what I do is about what not to do.
Every engagement ends with a written list of what I advise against. One of the four buckets exists purely for work that should stay with a person.
That's not caution. It's the list most advisors leave out — what not to build, and why.
- Your company is not too small. What matters is how much runs through how few people, not the headcount. The gain is in work nobody wrote down.
- The owner is in the room. Without you this doesn't work — not because I need your signature, but because half of what I'm looking for is only in your head.
- What gets written down is yours. Everything I deliver you can keep using, editing and extending. Including if you never work with me again.
- Not everything should be clever. Part of your work belongs with a person. That is a choice, not a backlog, and I say it out loud.
Every process goes into one of four buckets.
That's the whole model. Simple enough to explain to your team. Strict enough to stop you investing in the wrong thing.
This work does not need to exist. A report nobody reads, an approval that is always yes, a list kept twice. Costs nothing to remove, returns time immediately.
The work stays, the way it is done changes. Fewer handoffs, one place instead of four. What changes here is the shape of the work, not the tools.
Happens constantly, has a fixed shape, and you can tell within a minute whether the answer is right. This is where building pays off.
Judgement, relationships, exceptions, or real damage if it goes wrong. Deliberately not automated. A choice, not a backlog.
How I sort
Every process gets scored against five things: how often it happens, whether the knowledge is written down, whether the input is clean, whether you can check the result in a minute, and what happens if it goes wrong and nobody notices.
Four steps. Each one stands on its own.
Each step can be bought separately and delivers something on its own. You are never locked into the next one.
1. The Scan
I come for a day. I talk to you, and then to the people who actually do the work. I walk the floor, look at screens, and at the end of the day I draw how the business runs on the wall.
- The process map of your business on one sheet. As it is, not as it was meant to be.
- Every process sorted into the four buckets, and three things I would do first.
2. The Core
Here I take what your business knows and write it down. Why you exist, what you do and don't do, how you sound, who your customers are, your rules, your exceptions, your escalations. Then I get one thing running on it — usually something that answers your most-asked question, or clears the task you repeat most often.
- Your business written out, in your language, ready for use by people and by machines.
- One thing that actually runs on it, working on the day I leave.
3. The Build
I build one thing that works. Not five half things. Invoices that process themselves and land in the right place. Quotes that roll out of the core. An assistant that answers customer questions with your rules. Reports that are simply there on the first of the month.
- One working component, in use by your own team, with your people trained on it.
- A measurement: how many hours a month this returns. Before and after.
4. Maintenance
A core that isn't maintained goes stale. Rules change, prices change, your team changes. I keep it current, feed back what comes out of daily use, and build the next component when you're ready.
Many clients take it over themselves after six months, and that is exactly the intention.
No custom quote before you know the number.
All amounts excluding VAT. Prices apply to companies of up to roughly a hundred people. Larger than that takes more conversation, and I price it separately. If a job needs travel and an overnight, I say so before we book, not after.
Three free scans
No cost, no catch. Ends when three are booked, or on 31 December 2026, whichever comes first. I am early, and this is what being early is worth to you.
Founding Core price
The first three Core projects go out at €4,500 rather than €8,500. In exchange: permission to describe the result publicly, your name as a reference, and half an hour on the phone with someone weighing up the same decision. Ends at three, or on 31 December 2026.
What I don't do
- Resell software on commission. That makes my advice unreliable.
- Place people with you.
- Build something that only keeps running while I'm being paid.
- Promise percentages of time saved before anything has been measured.
- Take on work where the owner isn't in the room.
Taktältarna, Åre
Taktältarna sells rooftop tents from its own showrooms in Duved and Västerhaninge, with partners across Sweden and Denmark. I spent a day there, asking the owner how the work actually moves and where it gets stuck.
I didn't recommend a tool. I recommended what to fix first — a few things that needed nobody outside the company, before anything technical.
Then came the sort. Cut, simplify, hand to AI, or leave exactly as it is, with a person.
One finding stood out. The newest part of the business — content and the ambassador programme — turned out to be the best organised part of the whole company, built in-house by someone teaching themselves the tools. The hard part of this method was already proven inside the company before the scan even started.
This method was not invented on a whiteboard.
It was worked out on a real business.
Rental Valley rents out homes in Portugal and the Netherlands. It now runs on an AI core — invoices process themselves, owners get answers, prices are advised, reports simply arrive.
Every part of this method broke there first, and got fixed there first.
Tim
Process thinker, and the person who runs brainsandbuckets. Sees in a day where the work gets stuck, and can draw it in a way your team recognises and reacts to. That last part is the craft: a process map nobody responds to is a nice picture with no value.
Tim asks the questions you stopped asking because it was always done this way. He comes himself, he draws it himself, he builds it himself, and he is your only point of contact from the first call to delivery.
Niels Marcelis
Entrepreneur. Built Rental Valley, a rental business with homes in Portugal and the Netherlands, and turned it into an organisation running on an AI core.
Call or email and we'll book 30 minutes.
I'll ask a few specific questions: where it gets stuck, what costs the most time right now, and what you'd like gone by tomorrow. After that call you'll know two things — whether a Scan makes sense, and whether I'm the right person. If the answer to either is no, I'll say so.
Book the introduction