Where to start with AI
Leadership wants AI. You were handed the goal. Here is how to start with something that works, earns trust, and actually ships.
Most AI projects fail because they start in the wrong place. A flashy demo. A tool bought before anyone agreed on the problem. A pilot no one can measure.
If you are the person asked to bring AI into the business, the pressure is to move fast. The better move is to start small and start right. Here is what to pay attention to.
Start with the problem, then the technology
Start from the work, and find one process that is repetitive, high volume, and eats hours every week. Client onboarding, invoice matching, first-line support, weekly reporting. Something people already complain about.
The best first project is boring on purpose. It is easy to explain, easy to measure, and low risk if it needs a second pass.
Choose something you can measure
Pick a process where you can count two things: the hours it takes today, and what those hours cost. That number is your baseline. When the automation runs, the same number becomes your business case.
With a baseline, every conversation about value rests on numbers, and the case makes itself.
Keep a person in the loop
AI should propose. People should decide, where it matters. Map your process and mark the points where a human needs to sign off: a payment over a threshold, a message to a customer, anything hard to reverse.
Sign-off is how you move fast on the routine work and stay careful on the rest.
Insist on seeing what it did
When you can see what the AI did, you can trust it and defend it. Every step should be logged in plain language: what ran, what changed, and what the AI decided, in words your whole team can read.
It is worth building in from the start. When a customer or an auditor asks why a decision was made, a readable record is the difference between a short answer and a long week.
Know where your data goes
Before anything touches a live system, ask a question: where does our data go, and who can see it? Is it sent to an outside model? Is it used to train that model?
For most businesses the answer should be simple. Your data stays in your instance, in your country, and it is never used for training. A vendor worth choosing says that clearly.
You are ready to automate a decision once you can explain it to a customer or an auditor.
Get ahead of the regulation
The EU AI Act is law. If you use AI in ways it covers, you will need to document how those decisions are made. Retro-fitting that record after the fact is painful and expensive.
The easy path is to log as you go, from the first automation. Build the paper trail into the system itself, where it stays current on its own.
Bring your team with you
Automation lands badly when it is sprung on people. It lands well when it takes the work they already dislike and gives them back time for the work they are good at.
Show the team early. Let them watch it run on real cases. Train the people whose day it changes. The goal is the same people, freed from the busywork.
A short checklist before you start
If you can answer these, you are ready to begin.
- Which one process are we automating first, and why that one?
- How will we measure whether it worked?
- Where does a person sign off?
- Can we see, in plain language, what the AI did?
- Where does our data live, and is it ever used for training?
- What do we need to log for the EU AI Act?
- Whose day does this change, and how do we bring them along?
Not sure which process to start with?
Spend an hour with our team. We map where AI helps in your business, where it falls short, and what to do first. Honest, specific to you, and yours to keep.