Out of a Toronto office, a team of some 700 Loblaw Digital staffers has been experimenting with the grocery giant’s trove of data, applying AI-powered personalization and targeted ads to increase the loyalty of shoppers and encourage them to spend more.

The group creates and runs all the digital products responsible for bringing in Loblaw’s billions of dollars in annual e-commerce sales, and recently added the grocer’s media unit to its purview. In doing so, it has become a kind of hive mind for the technologies the company hopes will carry it forward, especially generative artificial intelligence. 

The unit’s growth arc has been steep. What started in 2012 as about a half-dozen people trying to prove e-commerce has a role in the future of Canada’s largest grocer has become a critical part of the business. It has come to the fore, though, amid high food prices and concerns that the nation’s major grocers operate as an oligopoly. Like all parts of Loblaw, it will face growing scrutiny.

The leader of Loblaw Digital is Lauren Steinberg, a former e-commerce manager with The Body Shop who joined the grocer in 2013 and took over the digital unit’s leadership in 2020.

In a recent interview with The Logic, Steinberg gave a glimpse into the scale of Loblaw’s digital ambitions, from the value of “personalization” to the role generative AI will play in the grocery industry of the future. 

This interview has been edited for length and clarity.

In what ways has your team, which is responsible for AI development across Loblaw, experimented with AI and implemented AI tools? 

Loblaw has used AI for years. Generative AI has allowed us to experiment.

We built this reverse meal planner where you can tell the chatbot what you have in your fridge and we’ll recommend recipes based on your implicit or explicit preferences. Then you can add the items that you don’t have to your basket. That’s something customers told us many times over they wanted. With gen AI, we were finally able to build it. We launched it as a pilot to test against things like hallucinations [incorrect information].

We built Garfielld, which is almost a large-language gateway internally for all colleagues at Loblaw, a safe environment for them to play with conversational gen AI models.

For our supply chain team, we built an app in just a couple of months that fully automates inbounding and outbounding of our fleet. There’s probably thousands of emails a day folks had to get through to be able to schedule loads. We’ve actually been able to scale it almost fully. We started with one per cent of the emails and I think by now it’s probably at 100 per cent.

I understand personalization is quite a revenue driver. How are you using generative AI to match product offers to customers better, and how do you envision expanding that?

How much time do you have? I’m the most passionate about that right now.

We built something called p13n. Thirteen being the 13 letters between P and N in the word personalization. This is effectively a fully baked platform to allow us to build some of the best personalization in retail.

In order to create content to service a one-to-one experience for customers digitally, you have to create billions of pieces of content. Even if I had every creative person in Canada working at Loblaw Digital, I still wouldn’t have enough capacity to build the assets needed to do that.

That’s where our content engine comes in—a gen AI-powered module within p13n that can in real time write the headline and copy, generate an image, determine the right call to action and deploy it.

You mean if I have the PC Optimum app and I go online to place an order, the engine you’re talking about registers ‘OK, Aleksandra is logged in. This is what she wants right now,’ and autofills that page for me?

Bingo.

You’ve said customers presented with these personalized offers spend $8 more per transaction. Has that metric changed at all?

We launched our first proof of concept in December. It’s been as high as $8 difference between p13n experience and our control group. On average, it’s probably a little lower.

I think what’s important to take away is customers are interacting with pieces of content in the p13n version of their experience at three times the rate they are in non-personalized.

We’re only live on two grocery banners, but we’re about to deploy it into a whole bunch of other parts of our digital business: Joe Fresh, Shoppers Drug Mart and PC Optimum.

What do you see as the biggest opportunity and biggest challenge for generative AI in the grocery industry?

The biggest challenge is everyone thinks it’s the next big thing and it’s going to solve everyone’s problems. The reality is it’s still so flawed. How do we balance staying ahead over investing in things that are rapidly changing?

The biggest opportunity is being able to deliver more value for the customer. I think just keeping our finger on the pulse but also not putting too much stock in it.

The company recently added Advance powered by Loblaw, formerly known as Loblaw Media, to your responsibilities. Where does that business stand today? 

I just took over that portfolio seven weeks ago, but I’ve been working with that team for many, many years. It was the Loblaw Digital team that built the retail media platform, which allows advertisers to buy on our advertising units across our digital platform.

It’s seeing great momentum. There’s a ton of untapped opportunities.

What data does Advance rely on to sell ads? What mix of PC Optimum, PC Financial and Health?

It is purely PC Optimum. We don’t use any financial or health data.

What types of ads does Advance sell on digital experiences? Is it embedded, affiliate ads, or ads separated from search results, like a banner? 

Kind of both.

We’re testing different types of ad products. Like, what do advertisers want that doesn’t conflict with what customers are trying to do?

We measure search relevancy, which means customers would have added or clicked a product that falls into the first eight search results. Our sponsored search ads seem to be no different. They follow an ad relevancy ranking. You might want to see a different brand of egg but other people might have an affinity to their eggs only. So we use inferences to help make that determination on the fly and we test.

There’s sponsored product ads throughout the experience and the idea is you don’t even realize it. We’ve had sponsored products on our sites for a couple of years now and nobody seems to notice. I mean in a good way. They’re shopping for them. Advertisers are continuing to invest in that product because it’s a great way for them to get in front of new customers.

Do you only sell ads to brands that sell their products through Loblaw?

Only products within our ecosystem. You’re there to shop for groceries. We want you to be able to see groceries. 

That doesn’t mean we won’t experiment in the future.

The Loblaw Digital website calls the loyalty app a superapp—one that provides the ability to do a lot of different things, like shop for groceries and bank online, in one place. In my understanding of the definition, it’s not quite there yet. Is a superapp something you’re working towards building?

No. We tested a few things a couple of years ago and what we learned was customers go to their place to do their thing. They don’t need everything in one place.

What tests led you to that conclusion?

We allowed customers to shop our marketplace, for example, in the PC Optimum app without having to leave. People are curious, but ultimately, that’s not what they were in the mood for.

I think we need to make sure it’s easy to go from point A to point B, versus ram all the point Bs into point A.

Our customers don’t necessarily want to do that. I think if that changes, we have the infrastructure to pivot.