Enabling MLOps to shorten GfK’s development cycle
Impact
ML6 advised and supported GfK on setting up an MLOps way-of-working, enabling GFK to bring insights to production in weeks instead of months, at much lower costs.
Intro to the customer
GfK connects science and data to help solve their clients’ business questions. As the fourth-largest market research organisation in the world, GfK provides their customers with actionable recommendations for key decisions about consumers, markets, brands and media to drive marketing, sales and organisational effectiveness. Data and advanced analytics is the lifeblood of the GfK business.
GfK is on a mission to revolutionise real-time access to critical knowledge, by transforming their offering from consultants supported by technology to technology supported by consultants. This technology is mainly offered to their clients in the shape of the Newron platform, a single access point that combines market, consumer and brand intelligence supercharged with AI-powered recommendations.
Challenge
To be able to deliver increased value with the Newron platform, it’s critical that GfK is able to bring meaningful insights to production in a timely manner. This means that the development cycle time needs to be brought down from months to weeks, requiring tighter collaboration between the data science and engineering teams in the company.
Solution
In order to propel GfK on their mission to revolutionise real-time access to critical knowledge, and to overcome the current challenges, we collaborated to establish an MLOps-driven way of working at GfK for Newron. The MLOps methodology bridges the gap between data science and engineering using a set of tools and best practices, empowering the organisation to bring AI solutions to production reliably and efficiently.
ML6 and GfK co-developed an MLOps system to bring sales forecasting models to production. We focussed on an initial implementation, with scalability in mind. We started with knowledge sharing sessions on MLOps principles, best practices, and tools, after which we co-designed a solution based on the specific needs of GfK. The solution was implemented with a mixed team of ML6 experts and members of various GfK teams. We leveraged daily standups, design sessions, pair programming and code review to transfer knowledge and build the necessary skills at GfK while developing the system.
The delivered solution covers a comprehensive CI/CD development workflow, and a production-ready pipeline, while accommodating the complex needs of GfK, including data and model versioning, data lineage, monitoring and alerting, data drift, and outlier detection. Thanks to our collaborative approach to building the system, the GfK members know the solution inside out.
By
Results
With the base system set up and the knowledge in their hands, GfK will now expand the system to include more models and expand existing functionality further. ML6 will continue to provide expert support where needed.