- The next step in artificial intelligence and machine learning deployment at H&M lies in the reusability of models, said Errol Koolmeister, product area lead engineer for AI foundation at H&M, speaking Thursday at the Virtual AI Summit Silicon Valley.
- Machine learning products support H&M operations in areas including design buying, assortment quantification, fashion forecasting and logistics. The company plans to create reusable components, going from use-case specific components "where each use case [is] responsible for the entire end-to-end process," and closer to the creation of component libraries that can be deployed across the organization, said Koolmeister.
- The retailer is building a new data platform called Yggdrasil, which will let teams throughout H&M use central data assets with end-to-end self service capabilities. Koolmeister said the company has plans to launch the platform at the end of year.
In AI implementation, organizations grapple with scaling issues. Advancing investments from the pilot stage into business critical processes is challenging, due to constraints in accessing talent and organizational culture pitfalls.
But given the threats the retail industry faces — consumers pulling back on spending and inventory challenges — digital is imperative. H&M "must act quickly to improve its online proposition globally" as it adjusts to shifting shopping habits everywhere, analyst firm GlobalData said in a research note.
This year, H&M already planned to open fewer stores as it expanded digital operations globally. But because of the pandemic, the company plans to close 250 stores, accelerating the reduction of its retail footprint.
Leveraging data to improve digital operations becomes critical for H&M, as the clothier navigates a changing retail industry contending with supply chain challenges and shifts in consumer behavior.
To democratize AI use throughout the organization, the company will need to follow a series of principles, according to Koolmeister:
- Bring tech closer to the business side, ensuring they have access to products built by the tech organization.
- Become experts in scaling and professionalizing AI capabilities.
- Focus on the value delivery component of projects.
"Doing this means that we need to leverage reusable components," said Koolmeister. "We can't hire enough people to build the solutions that we want to do, so let's build them smarter instead."
By Koolmeister's own admission, the company's machine learning journey "started quite late in the game," with an initial proof of concept in 2016.
"We don't have all the answers yet," said Koolmeister. "But these teams will be able to help the rest of the organization, and also focus more on engines to support the AI-based product teams."