Most organizations know AI is important. Fewer know where to start – and even fewer do it in a way that actually lasts over time.
It's not because they lack will. It's because they lack a structured model that takes them from thought to action, without losing control along the way.
Mats and I talked about the journey and how to move forward in a good way where no one runs too far and everyone can keep up.
Why? AI is changing everything — whether you want it to or not
AI is not new, but it has become accessible to everyone. It is compared to the smartphone revolution: what feels revolutionary today will be as obvious as BankID in a few years. Organizations that do not act now risk being left behind — and employees who do not understand the basics risk being deceived or making the wrong decisions.
Who? All employees — not just technicians
It's not programmers who will build everyday agents — it's regular employees. The hairdresser will still cut hair, the chef will still cook. What's changing is how the work is done. But that requires respecting that people are at different points in their skill development journey.
Key message about change management:
You can't hire yourself out of the problem — AI expertise is not available in sufficient quantities, and even if you find it, the person has to learn it in the business. The fastest and best way is for employees who already know what they're doing to learn AI. That requires meeting people where they are.
What? Process documentation and bounded agents
Most common mistake: rolling out AI on incorrect, outdated or undocumented data. AI does not value information — it uses what it is given. Basic requirements for success:
Documented processes
Not a complete process mapping — but enough for the AI to understand the steps. Write down how you do things, in plain English.
Bounded agents
One agent = one task. ”Write a report” is one task. ”Do all our external monitoring” is several. The narrower the definition, the better the quality.
Information ownership
The process owner is the one who cancels the vacation in the event of an incident. If you can't find ownership — start where it's clear, and build from there.
When? Now — but with the right framework
The best time to plant a tree was ten years ago. The next best time is today. You don't need to have the entire infrastructure in place — you tighten the framework according to what you have. If you start small (internal agents, known data, volunteer pilots), you also don't need heavy experts from the start.
What is critical in terms of time is not to wait for perfection. The concept — how to think with and collaborate with AI — does not change even if the models are replaced.
How? Champions program with respect for the skills journey
The structure for responsible AI adoption is based on taking care of your employees — not pushing out licenses and hoping for the best.
1. Send out a survey
Map out curiosity, concerns, and existing conditions (processes, licenses, information classification). All answers must be addressed — ignoring a tough question is the worst possible change management. Prepare your answers and ”We don’t know yet” or ”No” are the right answers if they are true. Here is a draft of a survey that you should adapt to the tone and language of your business.
2. Kickoff for everyone — not just the enthusiastic ones
Invite the whole group. Address concerns explicitly. Show a simple, safe example in the tone of the business. This is where you turn ”I’m going to lose my job” into ”this could actually be fun.”.
3. Champions program with workshops
Instructions, pilot missions, joint display of what has been built. The point of showing half-finished agents that sometimes go wrong is central — it de-dramatizes and builds psychological security. The program should NOT be run by a zealot but be a training program the way HR runs programs. Structured and long-term.
4. Iterate — never ”done”
Build agents, test, make them a little better. Agent KPIs aren't about token consumption — they're about whether it actually works for the people who will use it.
The essence of change management: you can't expect people to come to you and ask. We have to go out and meet the business where it is.
Use the Maturity Model as support
Think of the model as a project model. There are templates, support, and a lot of smart people who have thought of it. Use it.
Business Process needs at least 200
You don't need to have complete process mappings — but you do need to understand the steps in what you want to automate. AI can't read body language or fingertip feel. What isn't written down doesn't exist for AI. Start simple: write down how you do something, step by step, in plain English.
Key: process owner — the one who cancels the vacation when there's a fire.
Governance, Risk & Compliance at least 200
You don’t need to have GDPR-perfect control over everything — but you can’t run blindly. 200 means you know what applies to the specific initiative. Tighten the boundaries: run internal agents, use data that is known and harmless. Toilet paper is not sensitive. Patient records are. Start with what you know is safe.
Everyone runs out of the room when GDPR is mentioned — but you don't have to solve everything before you start.
Infrastructure at least 200
You need to have the licenses in place, a team that owns this, and a technical environment that encapsulates the work. A WhatsApp group with Meta AI won't do. The simpler the pilot, the less infrastructure is needed — but something fundamental has to be in place. Run it internally, in a confined environment, until you know what you're doing.
Management of Content at least 200
AI bases its answers on what it has access to — and it doesn’t value the information. An old quote and a valid price list are treated the same. It’s enough that you can point out: this information is true, current, and can be used. You don’t need to clean up the entire file structure — just tighten the framework for the pilot.
Most common error in AI rollouts: ROT data (Redundant, Outdated, Trivial).
Staff & Training must be 300
The one area that needs to be at 300 from the start, but not for everything in the entire business – for this skills development. There needs to be someone running the champions program, who can provide answers — or take the question further and come back with an answer. You can’t shoot from the hip here. Employees who are afraid, insecure or feel ignored will not use the tools. It’s change management, not technology.
You can't get hired out of this. Those who already know what's done in the business are the ones who will learn AI.
Don't miss the entire conversation — watch the webinar (a link will be provided here when it's up on YouTube) in its entirety and download the presentation as PDF to bring with you the most important things.
