How to prepare a company for AI: from the goal and the data to a first version

I start an AI implementation from a goal that can be measured and from the data the company already has. Only then do I pick the tool. That’s how I worked on the MAIA assistant for Adamed and on the Berries team’s dashboard.

Published 3 November 2026Based on the project: Adamed MAIA

What an AI implementation can mean

The AI implementations I lead are usually one of four things: a prototype for a board decision, a dashboard on company data, an AI assistant or an automation. Each needs slightly different preparation, but the first steps are the same.

The goal first, then the tool

First I agree with the company what should improve and how we’ll know it worked. At Adamed, it was the decision whether and how to build a tool for medical consultants. At Berries, it was letting the team see sales against the plan in one place.

A measurable goal lets you compare the result with the starting point after launch. Without one, it’s hard to tell whether AI helped at all.

Check what data you already have

AI answers from the data it can reach. MAIA answers in Polish, in the industry’s language, using the data in the app. So before I build anything, I check where the data lives and whether it adds up.

At Berries, sales, ads, newsletters and stock lived in several tools, and each counted results its own way. The first step was cleaning up measurement, so the numbers from Shopify, Google, Meta and Klaviyo could be compared.

Settle access and data rules early

Who will use the tool and what they should see.
How people will sign in. In the Berries dashboard everyone signs in with their Google account and sees the parts they’re allowed to see.
Which data can go into the tool and which stays out.
What data the first version runs on. For MAIA, it was demo data that resembles the real thing.

Start with a version people can try

I build the first working version with AI in a few days. Decision makers can use it instead of judging a presentation. At Adamed, the people reviewing the project asked the assistant about doctor profiles.

Bigger changes go to a test version first, and only then live. That’s how I work on the Berries store and dashboard.

A starting checklist

Set the goal and how you’ll know it worked.
List the data the company already has and check that it adds up.
Pick one person or one team the tool should help first.
Agree on access and data rules.
Build a first version and hand it to the people who decide.
After launch, compare the result with the starting point.

The goal first, then the tool

A measurable goal lets you compare the result with the starting point after launch. Without one, it’s hard to tell whether AI helped at all.

Check what data you already have

Settle access and data rules early

Who will use the tool and what they should see.

How people will sign in. In the Berries dashboard everyone signs in with their Google account and sees the parts they’re allowed to see.

Which data can go into the tool and which stays out.

What data the first version runs on. For MAIA, it was demo data that resembles the real thing.

Start with a version people can try

Bigger changes go to a test version first, and only then live. That’s how I work on the Berries store and dashboard.

A starting checklist

Set the goal and how you’ll know it worked.

List the data the company already has and check that it adds up.

Pick one person or one team the tool should help first.

Agree on access and data rules.

Build a first version and hand it to the people who decide.

After launch, compare the result with the starting point.