Ahold Delhaize’s Karin Chu Talks Moving From AI Possibility To Impact
At Ahold Delhaize USA, artificial intelligence is helping to optimize inventory management, elevating consumer experiences through increased accuracy in e-commerce and buy-online, pickup-in-store transactions.
Karin Chu, VP of AI and data science for the company, said that investing millions of dollars to run promotions to bring more consumers in may be worthwhile, but not if the store runs out of stock. By using AI, Ahold has enabled new warehouse efficiencies, ensuring that products are delivered at the right time and shelves stay full.
Its success has been driven by an in-house platform that drives AI insights through two families of algorithms: hourly item-level forecasts by store and hourly order-level data by store. It’s an effort that has required cross-functional collaboration with the consumer experience at the core.
“Transformational solutions are never built in silos,” said Chu during the recent Analytics Unite keynote presentation, adding that it was a collective effort that made its home-grown Spectrum technology come to life.
The company is continuing to further its progress with AI, looking at pick-path optimizations where associates can use insights to reduce unnecessary aisle traffic, AI batching and Gen AI-powered substitution recommendations.
How To Get There
Deploying AI requires a strategic roadmap that considers scalability, tech stack fit, and team capability, said Chu, who added that the company uses a “start small, scale fast” approach.
“Build with a vision of how you're going to scale it, because it’s expensive, and bake in your active user feedback to make your algorithm better,” she suggested.
It begins by aligning AI goals with business goals. Then comes resource assessment: “Is the data ready? Can we repeat this down the road once it’s live? Is the right technology in place? Do we already have a third-party solution we can upgrade or another layer to add, or do we have to build a custom solution?”
Also: See Mark Anthony’s AI blueprint, which includes leadership buy-in and collaboration
Once created, it’s a living product, not just a formula. Because of this, teams need to identify someone who can upgrade it based on changes in the business environment. Data science teams should also be able to sprinkle a little magic, refining back-end processes without impacting the consumer-facing experience. Behind the API, she said, are hidden algorithms and A/B testing powered by large language models.
“With AI, it doesn’t stop at deployment,” said Chu. “You need machine learning engineers so the code can go live. Then you have front-end engineers to build out the user experiences. Since it’s a product, you need a product owner and also a business stakeholder who is using it. Collaboration fuels success.”
Moving From Possibility To Impact
While proof of concept is cool, said Chu, AI implementations ultimately need to add value.
“The success comes from the impact it makes, not from the data volume you have,” she added. “How are you going to use user data? Without having a clear path forward on how to squeeze value out of that data, once you start gathering it, you’re forming expectations with customers that you may not be able to meet.”
Additionally, having built-in controls is necessary. There should be a backup plan with human intervention because there’s no successful AI without the human element, she said.
AI strategy transitions from theory to practice with the right mindset and culture. AI literacy has to be a part of the core, making shared learning and continuous progress key parts of empowering people to build their confidence and knowledge.
Speaking as a statistician, Chu said the data science and the tech stack are not the challenges. Actually, the algorithms are the easy part. What’s difficult is change management because at the core of these implementations are people.
“If you don’t have that down, it’s not going to work,” said Chu.
