Successfully piloting AI is one thing, but deploying artificial intelligence across an organization and creating a data-driven culture is another.
"We aim to bring back methods, processes, and potentially even ready-made software solutions for MLOps to our company," says Mattias Jonhede, Manager Advanced Analytics Engineering at Volvo Group.
In terms of participating partners, this newly launched initiative is AI Sweden's largest project to date.
Machine Learning Operations (MLOps) are a set of methods, tools, and processes that aim to make machine learning development more robust and scalable, with a higher degree of automation and collaboration across different departments and disciplines.
How can an organization truly become data-driven? AI Sweden's new MLOps project, formally titled "Data-driven organizations – Best practices for operationalization of AI in Sweden," aims to answer this question.
Fifteen of AI Sweden's partners from the private sector, public sector, and academia will address challenges related to large-scale deployment of AI solutions and the broad application of AI across organizations. The overarching goal is understanding how organizations should adapt as data and technology evolve. The project aims to develop tools and guidelines that help businesses implement suitable organizational structures and technologies to work in a data-driven manner beyond the prototype and testing stages.
"As we begin deploying AI in production and scaling up the number of models, it quickly becomes apparent that the choice of organizational structure and technical infrastructure are interdependent. Often, the responsibility for a model falls to someone who didn't develop it, sometimes even to individuals with less technical expertise."
Stefan Wedin
Project manager at AI Sweden
A key distinction between most prototypes and deployed AI solutions is that the latter must function sustainably over time. Achieving this requires a deeper understanding of the technology and innovative approaches to organizational structure and processes.
The technical challenges encompass changes in both data and models. Key questions include: Will new data interact with a current model in the same way as existing data? Will an updated model interpret data consistently with its predecessor? How can models be effectively retrained with new data?
"This clearly illustrates the intersection of technology and organizational issues," explains Stefan Wendin. "When evaluating different combinations of data and models to determine the optimal choice, you need not only assessment tools but also a clear vision of the AI application's objectives. Defining 'best' requires aligning technical capabilities with desired outcomes, which extends beyond purely technical considerations."
From an organizational perspective, key considerations include defining necessary roles and responsibilities, establishing processes to guide AI toward appropriate objectives, and determining prerequisites for new projects, data utilization, and technology adoption.
Scheduled to run until December 2025, this initiative represents AI Sweden's largest project in terms of participating partners.
"I attribute the high level of interest to the fact that many Swedish organizations have now completed one or two AI pilots, only to discover that the transition to full implementation is more challenging than anticipated," explains Stefan Wendin.
Ellen Reinhardt, Head of Project at Aixia and co-project manager with Stefan Wendin:
"This initiative enables us, as a supplier, to collaborate closely with both existing and potential clients. By jointly exploring their needs, we can gain deeper insights into the solutions we should be developing and offering."
Fredrik Enqvist, Product manager (AI platform) at Region Västra Götaland:
"A significant hurdle we face is the complexity and cost of processes required for full-scale AI operations, given the technological and competency demands involved. We're hopeful that this project will yield a versatile, platform-agnostic pipeline applicable to our operations."
Daniel Jakobsson, AI Strategist at the Swedish Transport Administration:
"While we have several promising models in our research pipeline ready for implementation, we currently lack the technical and organizational capabilities to do so. To a repeatable fashion put models in operation. This project presents an opportunity to overcome this challenge and find viable solutions."
Mattias Jonhede, Manager Advanced Analytics Engineering at Volvo Group:
"Our plans involve deploying thousands of models to forecast and optimize our aftermarket logistics. A key challenge we anticipate is ensuring robust technical support for governance as models transition from development to implementation and ongoing operation. Clarifying roles and responsibilities throughout this process is crucial."
Project Title: "Data-driven Organizations – Best Practices for AI Operationalization in Sweden"
Objective: To establish a framework for functional and scalable MLOps, facilitating AI implementation in Swedish organizations, encompassing necessary processes and organizational adaptations.
Participating Entities: Aixia, Hewlett Packard Enterprise, Linköping University, NetApp, Proact, RedHat, Region Halland, Sahlgrenska University Hospital, Statistics Sweden, Swedish Tax Agency, Stormgrid, Swedish Transport Administration, Volvo Parts, Region Västra Götaland, and AI Sweden.
Funding: Vinnova and participating partners
Budget Allocation: 35.7 million SEK
Project Duration: April 2024-December 2025