Overcoming obstacles to AI innovation
Industries of all kinds have felt the impact of AI — including the agricultural sector, where individual and commercial gardens alike are prime candidates for automation, better equipment care and personalized customer experiences. At Husqvarna, their mission is to modernize landscaping by tapping into data insights and AI to optimize the performance of their wide variety of outdoor power products — including chainsaws, trimmers, brushcutters, cultivators, garden tractors and autonomous lawn mowers — and to improve the overall customer experience.
As they embarked on their data-driven journey, Husqvarna lacked the resources and data infrastructure needed to not only access and unify their data, but also share and analyze it to support the business solutions they envisioned. With the native AWS solutions they were using, maintaining compute resources was highly complex and resource-intensive. And Jupyter notebooks running off individual laptops weren’t scalable, making it challenging for data teams to explore data at scale, and to collaborate and share insights with the wider team. As a result, decisions were made in silos and on gut feeling rather than by depending on the data.
Additionally, disparate data meant that professionals at Husqvarna were limited in what they could do with it — and the little they could do was expensive to maintain. Often, along with siloed data came siloed skill sets, preventing the type of collaboration and streamlined processes needed to continue seeing success.
“A lot of people felt blind,” explained Linus Wallin, Data Engineer at Husqvarna AI Lab. “When we thought about what to do with a product, the outcome was often opinion based, or what was at least seemingly common sense. It wasn’t a strategic way to operate and was not at all scalable. But we didn’t have the ability to democratize our data for other teams to explore themselves.”
For many reasons, the Husqvarna team knew their data architecture was in need of an overhaul. “Our old system was extremely difficult to maintain,” said Wallin. “Everything took an incredible number of man-hours to do, so managing the cost of our clusters when they got too large, bringing those clusters back up when people needed them in a timely manner, and doing all of that manually — it was exhausting and horrible. We needed a better path forward.”