01 / Good first candidates
Four kinds of work AI handles well.
Documents in, data out
Invoices, delivery notes, forms, contracts. Read them, pull out the fields, put them where they belong.
Requests sorted and routed
The shared inbox, support tickets, orders by email. What is it, how urgent, who handles it.
First drafts a person approves
Quotes, offers, replies, reports. AI writes the draft in seconds; your people check and send.
Checks against rules
Order against invoice, completeness before approval, the same quality check every time.
02 / The five-question test
Is this process worth automating?
- It happens every week, or more often.
- It looks roughly the same each time.
- You can say what goes in and what should come out.
- A mistake gets caught before it costs money.
- One person owns the process today and can judge the result.
Four or five yes: a strong first candidate. Two or fewer: fix the process first, or pick another one.
03 / Not yet
What to leave alone for now.
- Decisions that carry liability, with nobody reviewing them.
- A process that changes every month. Automate it once it settles.
- Work whose information sits on paper or in people's heads. Get it written down first.
- A broken process. Automating it makes it fail faster.
04 / AI or plain automation
Most of a good automation isn't AI.
Fixed steps, like moving a record from one system to another or sending a reminder, are cheaper and more reliable with plain rule-based automation. AI earns its place in the steps where the input varies: reading a document, understanding a request, writing a draft. The best setups are usually plain automation with AI in one or two steps, and a person where a mistake would cost money.
05 / How it runs
One process, live, measured.
Measure it today
How often it runs, how long it takes, what mistakes cost. That's the number the automation has to beat.
Build, buy or skip
Often a tool you already pay for can do it. Sometimes the honest answer is "not worth it".
Run it next to the people
The automation drafts, the people approve. You see where it's right and where it isn't before anyone relies on it.
Hand it over
The team owns it, knows when to override it, and knows who to call when the input changes.
06 / Questions
Questions people ask
What is AI automation?
Using AI to do the steps in a process that need reading, sorting or writing, such as pulling data out of documents, routing requests or drafting replies, usually combined with plain rule-based automation for the fixed steps. A person stays in the loop where a mistake would cost money.
Which processes should a small or mid-sized company automate with AI first?
A process that runs every week, follows the same pattern, has clear inputs and outputs, and costs real hours today. Typical first candidates are documents turned into data, requests sorted and routed, first drafts of quotes or replies that a person approves, and checks against fixed rules.
Will AI automation replace our staff?
Usually it removes the repetitive part of a job, not the job. The people who did the work review the output, handle the exceptions and own the process. Companies that plan it that way get more out of it than those that plan for headcount cuts.
Is AI automation allowed under GDPR?
It can be set up that way. The questions are which personal data reaches the AI provider, where it is processed, whether the provider uses it for training, and whether a data processing agreement is in place. Often the personal data doesn't need to go to the model at all. Check the setup with your data protection officer; this is not legal advice.
Do you only advise, or do you also build the automation?
Both. I can pick the process and set the approach while your team or a vendor builds it, or deliver it with my team and partners, led by me until it runs in daily use.
07 / Read next