From pilot to practice

We help organisations identify where AI makes the most difference, and then build the solution so that it works in practice. We also tell you which use cases won't hold up. A use case nobody can put numbers on is an experiment, not an investment.

We work across the full range: classic predictive analytics and machine learning, generative AI and large language models, and AI agents that solve concrete business problems. Often in the same engagement, the choice of method follows the problem, not the other way around.

The job doesn't end at delivery. It extends to the solution actually being used, maintained and owned by you. Someone needs to own it on the Monday after we've gone home, and that should be decided before we start.

‍

Why data, not models

We're not an AI company that learned data. We're a data company that built AI.

A model is only as good as the data it works with. If the data is scattered across systems, lacks quality control or can't be traced, it doesn't matter how good the model is. AI becomes an experiment that never leaves the lab.

That's why our AI engagements rarely start with the model. They start with what data exists, what it means, and who is responsible for it. Often that work is already done, and when it isn't, that's where we begin.

Who builds it

AI isn't a department here.

It's easy to put together an AI team on the side of the business. It produces a demo fast and a solution rarely, because the team lacks what decides the outcome: knowing what the data actually means. Here, AI is built by the same people who built the data platforms and the reports, in businesses they have known for a long time.

That matters for a reason that's easy to miss. AI makes competent people faster at what it's good at, and makes everyone more confidently wrong at what it isn't. The difference doesn't show in the answer. A KPI can be right at day level and quietly wrong at month level, and the report looks just as convincing either way. What's missing is rarely the model. It's someone who recognises the error.

We also build with AI ourselves, every week. Internal tools for content, motion graphics, consultant profiles and engagement proposals, built by us, for us, in daily use. That's where we see what holds up and what only looks good in a demo.

Technologies we work with

We build where the data already lives, rather than moving it. Databricks when data and AI should live in the same place, Azure AI Foundry for putting language models into production with governance and traceability, Snowflake ML when the models belong close to the warehouse. Python throughout.

We also build RAG solutions, where the model answers from your own documents rather than from whatever it happens to remember, with citations back to the source document, so the answer can be checked. And AI agents that carry out tasks in your systems, not just describe them.

We choose models and vendors so they can be replaced. The connections between data, functions and systems are built so that your material and business-critical parts stay with you. That way you can change model later without rebuilding the business.

We recommend what actually solves the problem, not what happens to be closest at hand.

‍

Pilots that worked but never became anything. Models built on data nobody trusts. Solutions nobody maintains once the project ends. Organisations that bought an AI tool and discovered it doesn't know their own processes, contracts and previous work.

And most often: an ambition that's bigger than the foundations. Lots of ideas, high pace, and a data foundation that can't carry the weight. We start with what you want to achieve, not with the model you just heard about. That review decides which use cases are worth building, and which ones we recommend you skip.

That's where we make the biggest difference, when AI stops being a project and becomes a way of working.

Working in legal? See Legal AI for law firms and in-house legal teams.

Let's talk data.We'll help you find the right way forward, wherever you are today. Contact Mattias Tengblad.

‍