Start with the work, not the tool
If you begin by asking which AI tool you need, you start from the wrong side. A tool is a means. First you need to know which work you want to run by itself.
So look at your own week. Where does someone type data from one screen into another? Where does the planning live in a spreadsheet that only one colleague understands? That is the work you can automate. The decisions stay with you.
The three questions we start every project with
This is how our team works. The order matters, because each question builds on the one before.
- The problem: which work does someone retype every week?
- The workflow: what can AI read and prepare, and what runs on its own?
- The person: who signs off, and what happens when the system is unsure?
What the work looks like when it runs by itself
An example makes it concrete. These are situations we recognise at companies that automate their manual work.
- A document comes in and the system reads it, instead of someone retyping it.
- A form fills itself with data your company already has somewhere.
- A report someone used to put together by hand every Monday assembles itself on Sunday night.
- A queue arrives already sorted, coded and checked. The few truly unclear cases are set aside for a person.
- Data is entered once and then stays correct across your mail, Excel, accounting and CRM.
When you need AI, and when you don't
You need AI where the work requires reading: documents, mails, invoices. The model reads, sorts and prepares. A person checks and decides.
Where the work is only passing data along, you don't need AI. A regular workflow is enough. It is simpler, easier to check, and you don't pay for what you don't use.
When the system is unsure, it doesn't guess. It puts the case in front of a person, with the reason attached. One example is AI that reads and prepares invoices in an existing accounting setup, after which someone signs off. You can find it under our work on AI and automation.
Which tool fits which work
There is no best tool, only a fitting one. Here is a short guide.
- Power Automate belongs to Microsoft 365. It is the obvious choice if your team works in Outlook, Teams, SharePoint and Excel.
- n8n can run on your own server, so your data stays under your own control.
- Zapier and Make have ready-made connections with many web apps and are quick to set up.
- Excel and Power Query fit when the data already lives in Excel and your team works in it.
- KNIME fits recurring data flows that must be fetched, cleaned and combined the same way every time.
- Python is for what falls just outside a standard tool.
- An AI model comes in when the work requires reading, such as documents, mails and invoices.
Build on what you have, or start over
Most of the time we build on the systems you already have. We find the part that still runs by hand and put the automation behind that part. Your tools, your logins and the reports people trust stay. Nobody has to learn something new. That is usually the difference between a project that lands and one that is quietly given up.
Sometimes the old system is the problem itself. The process lives in a spreadsheet, in one person's head, or in software that will never bend. Then we rebuild it as a real application, with AI designed in. That is the more expensive route, and we say so upfront.
What to look for: the bar
These are the demands we hold ourselves to. You can set them for any other supplier too.
- It is faster than what it replaces, measured against your old process and not against a presentation.
- There is a trail that someone can check.
- Someone who doesn't program can run it.