For service firms with a digital core
Accountants, consulting and engineering firms: your work is full of text, data and recurring steps. That is exactly where AI pays off fastest — if someone builds it instead of advising about it.
A lot of text, a lot of data, a lot of repetition
At accountancy, consultancy and engineering firms the work is full of the same movements: read a document, summarise what is in it, check whether it is right, and retype it into another system. Exactly the kind of work language models are good at — and exactly the kind of work nobody tracks the volume of.
The trap is that it stays talk, or a pilot on tidy sample data that then stalls. On tidy examples AI always does well; your files are not tidy.
So I build with your own documents, in your own process, with the people who do the work today involved. Where it runs and under what conditions we decide together.
Sound familiar?
A lot of time goes into work that is the same every time: reading, summarising, checking, retyping.
AI has been discussed and maybe piloted, but nothing is running in daily practice yet.
You want to know what it delivers concretely, not what it can do in theory.
What I do about it
01
Work that keeps coming back
The repetitive steps in your files and processes, built into something that runs on your own documents.
02
Knowledge you can actually find
Everything locked up in reports, files and email threads made searchable in a way your colleagues trust.
03
A decision within weeks
Every proof of concept ends with a clear decision: keep building, adjust, or stop.
I work inside your processes, with your data. Not a generic AI tool, but something that fits how you actually work.
Three things that started this way
Text, administration and repetitive work: exactly where your hours go.
Dictation in your own writing style
TokTalk
Speak instead of type, with text that comes out sounding like you write. Runs on European AI with EU data storage; here privacy is not a footnote but the product. Mac and web, paid from €4.50 a month.
toktalk.coInvoices without twenty logins
Invoice Fetcher
Collects invoices from suppliers automatically. One login per site, then batch downloads — and what the system learns about one website benefits the next user.
finvoice-phi.vercel.appAdoption software for animal shelters
Animalz
One popular animal draws thirty replies in a few hours, all answered by hand. Animalz sends a personal message on every status change and pre-sorts replies, so the hours go to the animals instead of the inbox. Open source, free at its core.
animalz-iota.vercel.appHow this works with your data
The question service firms always ask first, so here it is upfront.
- Preferably we work with real data. That is the only way to know what it delivers for you — on sample data everything works.
- Where the proof of concept and the data live we decide together based on your policy. I prefer an isolated environment of my own, so we do not get stuck on internal policy while we are still finding out whether something works.
- A data processing agreement is part of onboarding.
- Which models we use depends on the problem. If you want to limit yourself to European models, that is possible in consultation.
When I am not the right person
Three situations where someone else serves you better.
- You are after certification, a governance programme or an AI policy document. Others are better at that.
- It has to be an off-the-shelf package you only need to switch on. I build on your own process, which is bespoke by definition.
- Nobody is allowed to spend time on it. Without the people who do the work today, you still know nothing at the end.
on tidy sample data everything works →
A working proof of concept — or that phase costs you nothing
And you can stop at any decision point — no obligations
What people ask about this
- Is our data even allowed near an AI model?
- That depends on your policy and on the kind of data. Where the proof of concept and the data live we decide together, a data processing agreement is part of onboarding, and limiting to European models is possible in consultation. What is and is not allowed within your sector is a question for your own officer; I focus on what is technically and practically possible.
- We had a pilot that went nowhere. What is different here?
- Two things: we work with your own data, and the people who do the work today are involved. A pilot on sample data proves the technology works — we knew that. What you want to know is whether it holds up on your files, with your exceptions.
- How long before we see something?
- A working first version is usually there within a week. After that we sharpen it based on what your people think of it, and it ends with a clear decision: keep building, adjust or stop.
in closing
Want to see if this fits?
No sales pitch. One conversation to find out whether there is something worth building — and if there is not, I will say so.
€ 150 per hour · 1 to 3 days a week · you can stop at any time