Yellow Elk launches dedicated AI hub
Sweden is among the best in Europe at getting started with AI. And among the worst at crossing the finish line.
Mattias Tengblad is launching Yellow Elk AI, a new hub dedicated to taking organizational AI initiatives from pilot to full-scale production. With the data foundation in place, and with ownership that lasts over time.
Sweden is among the best in Europe at getting started with AI. And among the worst at crossing the finish line.
Mattias Tengblad is launching Yellow Elk AI, a new hub dedicated to taking organizational AI initiatives from pilot to full-scale production. With the data foundation in place, and with ownership that lasts over time.
The situation:
Almost everyone has tested something. The pilot worked, the demo impressed, and then nothing happened. The model never accessed real data, no one had decided who would manage it, and when the project ended, the momentum died with it. What remained was an experiment that never left the lab. The numbers tell the same story. In Frends' benchmark study, conducted by Sapio Research in April 2026 among 611 decision-makers in six European countries, 38 percent of Swedish organizations have AI in production, compared to 34 percent in Europe. However, only 21 percent of Swedish projects reach measurable bottom-line results, compared to 25.5 percent in Europe.
We start more projects. We land fewer. AI almost never fails because of the model. It fails because of the data beneath it, because no one has decided who owns the solution, and because no one starts using it.
Why data, not models:
We are not an AI company that learned data. We are a data company that builds AI.
Yellow Elk has been building data platforms and data foundations since long before AI became a buzzword. Over 200 consultants work daily with Snowflake, Databricks, Microsoft Fabric, Azure, AWS, and Google Cloud, often in organizations with high requirements for traceability. This matters for a reason that is easy to miss. AI makes competent people faster at what it is good at, and everyone more confidently wrong at what it is not. The difference isn't visible in the answer. A key metric might be correct on a daily basis but silently wrong on a monthly basis, and the report looks equally convincing in both cases. What is missing is rarely the model. It is someone who recognizes the error.
Who we are:
The hub is launching with six people who together cover business, data, modeling, and what it takes to actually get a solution into production. Mattias Tengblad leads the hub. An entrepreneur with over twenty years of experience as a CEO and business builder, including roles at Universal Music, MTGx and Viaplay, Pop House Group, and the music platform Corite.
Peyman Safari Hesar, Lead AI Engineering. Built the data science function at Sambla Group from the ground up, including the data platform, pipelines, and BI, and then put predictive models into production in a regulated financial business.
Carl Rygart, AI Platform Engineer. Twelve years of experience in security-sensitive and regulated systems, most recently working with RAG solutions for searching contracts and settlement documentation.
Amanda Carp, AI Business Developer. Holds an M.Sc. in Machine Learning from KTH. Drove the AI agenda at Vattenfall Customer Service Nordic, where she took an LLM solution from concept to production with measurable cost savings.
Rasmus Mattsson, AI Engineer. Builds production-ready AI pipelines and agent-based systems, including work for an international consultancy firm.
Andres Vourakis, Senior Data Scientist. Eight years in analytics and AI. Builds production ML models and analytical workflows that allow users to query data in plain language against a governed semantic layer.
Quote:
"Most AI projects don't fail because of the technology. They fail because of the underlying data, and because no one has decided who owns the solution once the project ends. That’s why it’s not enough for our consultants to know the tech. They have to understand the business and the end-users, otherwise, the implementation becomes someone else's problem."
"We start with what you want to achieve, not the model you just heard about. That assessment determines which use cases are worth building, and which ones we recommend you skip."
— Mattias Tengblad
🫎 About the company
Yellow Elk AI works with AI strategy, machine learning, generative AI, data platforms, and management for organizations looking to take AI from experiment to production. Founded in 2026. Part of Yellow Elk AB. Based in Stockholm.
Devoted to data
Contact