How I work with data and AI
I always start by understanding what the solution needs to do. Then I build in small steps, so you can see along the way that it works, and we can correct course as we go.
Approach
Every successful data and AI initiative starts with a thorough business analysis. Done properly, by a technologist who understands the business, the technical work is a breeze.
My formula
The assignments differ, but the formula stays the same: six steps from the first conversation to a solution people use.
- 01
What does it need to do?
I nearly always start by understanding and agreeing on what the solution needs to do, or "hva’ ska’ den ku’", as we say in Jutland. Which decision should get better, and which KPI should move?
An advertising agency wanted to work less and earn more. They sold hours and projects, so it came down to the average hourly rate. A sharp overview of time and project finances showed where the money was made, and a year later the hourly rate was 15–20% higher.
- 02
Model the business
From there, I move into business modelling or process mapping using proven methods. I think in the core objects of the business, but in a way the whole business can take part in. It is about aligning concepts, terms and how the KPIs are understood, because people often see reality differently. At Bestseller, I ran and facilitated that process across all 14 brand houses.
The modelling can stay high-level or go deep enough to specify the data governance requirements too.
- 03
Draw the architecture
With the model as the starting point, I lay out the technical architecture: how the business becomes data and reports, or how skills and agents should be organised in an AI project.
This is where many go wrong with AI right now. Departments vibe-code their own solutions, and you end up with shadow IT and data that don’t fit together. Architecture and governance are what hold it together.
- 04
Build in small steps
Implementation happens in close collaboration with the business, in iterative deliveries where we make sure every business feature is included. I like to build a working version 1 with features people can use, and expand from there.
I much prefer small cycles that show progress and fix mistakes along the way to big projects where five developers go into a room, burn 6,000 hours and come out with a large bill and something that isn’t quite what you thought you had ordered.
- 05
Launch it properly
Then comes the rollout. Technically, it is straightforward, but I have strong views on observability: you should be able to see that the solution is running and that the numbers are right.
- 06
Get it into use
Once everything is built, you can shake hands and say "great, it's running". But getting the business to adopt something new, perhaps in place of something people have used for years, is a huge task. It can easily take more than building it.
It takes stakeholder management, change management and stepping into people's habits. And then we follow up: Did the KPI move? If not, we change the solution.
