Kroger’s Yael Cosset on AI Shopping Assistants and the Road to Autonomous Retail | Groceryshop 2026

Hello, everyone.

This is Omnitalk Retail.

I am Ben Miller, host of grocertalk, and I'm coming to you live from Las Vegas from the viewsion podcast studio booth 913 at grocery shop.

And joining me now, fresh off his main stage keynote, is Yael Cosset, EVP and Chief Digital Officer at Kroger.

Yael, welcome to the podcast.

Thanks for having me, Ben.

Okay, come on.

You are Kroger's Chief Digital Officer.

So what's in your remit and what is currently top of your to do list?

Big question.

The title can be misleading.

It's really about the journey.

My responsibility really span everything that we've done around digitizing the customer experience and digitizing the associate experience and our operations.

So if you think about technology at the service of creating and consuming data, data at the service of data science, artificial intelligence, and how we raise the bar for the customer experience or our associates or our operations.

Okay, fantastic.

So one of the things you mentioned there was that you look after data for AI strategy.

So let's kind of, what is the AI strategy?

What are you trying to achieve?

And have you got the data foundations in place to enable you to do it?

Short answer, of course, we have the data foundation all figured out.

In reality, you know, we have a unique asset.

We've been investing in data capability for decades now.

We have petabytes of customer data and operational data, and that's a massive, massive asset and massive advantage.

Now, is that data 100% ready for consumption in a very sophisticated model, whether it's a proprietary model or a frontier model?

Not quite.

And that's part of the investment we've been focusing on over the last year, year and a half, and continue to do so.

But the ambition really is to make that data just like water, consumable by all of our models, to really raise the bar, bring to life a customer experience that is differentiated, bringing both the customer context and the operational context.

So it's not just about personalization and relevance for the customer.

It's also what is the reliability from an operational perspective to fulfill that experience, to support that experience, to drive the.

Back end as well.

Okay, let's come on to both those elements.

Let's start with customer facing.

So you've got an AI shopping agent launched, been out there in July.

Okay, what I know is what are shoppers actually using it for?

Because please don't tell me.

To organize super bowl parties, as every press release says.

What are they actually using you for?

What have you learned so far, Yael?

I mean, no, it's not to organize their super bowl party.

Maybe it will be this, this winter.

No, I think the, the, there's really two stages.

I think there's the curious stage, which is like, oh, I hear about it, or I see you have a shopping agent assistant.

You know, what does it really mean?

I try it.

It could be, I have a photo of a recipe, I have a photo of a meal I just had at a restaurant.

And let the agent, let the assistant really translate that, that inspiration, whether it's my grandma's favorite gnocchi and bolognese sauce recipe or saw online and took a snapshot of so really kind of like removing that friction.

It's kind of curiosity testing.

Okay.

The second phase is like, now that I understand the extent of the capability, let me make that shopping assistant work for me rather than just like discover.

And that's where I think it's been kind of mind boggling what consumers have done, what our shoppers have done.

It ranges from giving us very specific context around health restrictions.

So not health specific, like, you know, health diagnosis, but, you know, allergens.

It's budget constraint, time constraints, family profile, plans for the day, for the week, and really, you know, relying on that assistant to simplify their life.

And I would tell you, far exceeded the expected use.

The scale and the complexity of context that consumers are giving us to help make their life easier.

That's really interesting.

So you've got greater adoption, more depth.

What else has surprised you?

Anything else in the adoption of the tool?

Yeah, I think the part that's counterintuitive or was counterintuitive for me.

Now I'm being educated.

When you're very precise in how you satisfy a customer's expectation or request, I get your context, I create a meal, I'm bridging the inspiration to your shopping cart.

You tend to assume that that shopping cart is going to be very narrow, very precisely and contextually optimized.

Yes.

And therefore not much shopping outside that specific purpose.

And actually I was wrong.

And I think we were wrong.

We underestimated the fact that once you create that connection with the customer, give them some time back, some simplicity back, they actually still want to browse and branch out away from that specific occasion.

So I would say the, the basket composition is not as limited as I would have expected.

So it's become a personalized launch point.

That's correct.

Really interesting.

Really interesting.

Okay, I'm going to move on because we've got time and I could talk to you all day.

I Heard you talk about AI moving from assisted to augmented to autonomous.

So it was super interesting.

Can you just kind of explain what your thinking is there and what that means for Kroger?

Yeah, absolutely.

If you look at the importance of the intersection between the customer context and how we use AI in that, in that specific environment and the operational context we have on both sides, really moved on a journey on a continuum from assisted, which is using the operational example, providing insight to inform an associates who may be making a decision on a merchandising assortment, or an associates who is replenishing a shelf in a store in an optimized way.

So that's the assisted.

You.

You know, I kind of know what I need to do.

Just give me the nudge, the insight that makes me smarter or better.

Yeah.

The second is around augmented, which is I want to create a more complex level of intelligence around how multiple steps in what I do every day can come together and bring to life something that's superior.

And I would say consuming a lot of data signals is hard.

As a human, we tend to be single threaded.

Okay.

And that's where augmented, the augmented stage in artificial intelligence really comes to life.

So if you think about replenishment, I.

Was going to say give me an example.

Well, replenishment is a good example because, you know, I need to know what matters to the customer, how low I am on my store, on my shelf, on my holding power, but also need to know what's available upstream.

Okay, cute.

But I also need to know what is the next campaign, what is the marketing nudge that I'm about to activate in the next day, two days, week?

Because that should also inform what holding power I need to have in the store.

Yeah, it's hard for us as individuals, as human to really optimize.

You end up optimizing functionally for the supply chain, for the labor in the store to really restock that shelf as opposed to for that customer output.

The last stage, really, and I'll make it short, is the autonomous point, which is how do we accelerate the integration of all these points and really act as an enterprise built around the customer and allow a number of decisions or actions to be taken autonomously.

Again, not without the human in the loop.

You know, verifiability, auditability of all these decisions and actions is non negotiable.

Otherwise you lose the trust of the associates or lose the trust of the customer.

I've got that.

I was going to ask about governance.

You've already answered that question.

So the autonomous stage, is that for you hypothetical, are you building the blocks to get there?

Are you already doing some bits of that?

We have areas in our business today that are in the autonomous stage now.

You know, it's specific areas where we feel comfortable.

We're not creating risk either around the integrity of the data or our operations.

There are things that are a lot easier, you know, being wrong on a replenishment cycle order, optimizing the batching of picking in a store, optimizing the picking and creating an autonomous dynamic algo that direct and drives the picking or the punishment process.

Not a lot of risk.

You get wrong, the data learns from it, it optimizes better than.

There are areas where we're a little bit more conservative, specifically around the consumer.

Being wrong and ignoring or making a mistake around allergens is something that can be quite devastating.

Both trust and obviously health for customers.

So we are balancing the two aspects of wanting to leverage the capability but also being responsible in how we do so.

Understood?

Understood.

Look, yeah, there's a wider strategic update from KRoger Coming up 20th of October.

But one thing that you're relatively new.

I don't want to call him new.

You're relatively new.

CEO Greg Foreign has already talked about is the need to simplify the business for associates so they can spend more time with customers.

So what's your role and what's AI's role in that process?

There are hundreds and hundreds of activities of tasks that our associates, whether they are in our offices, in our supply chain facilities and our stores are taking that are, I would say, commodity, mundane tasks.

Nobody wakes up in the morning and say, oh, I am so inspired and I can't wait to work on that spreadsheet so that I make sure this product resonates with the customer.

Nobody does that.

Our associates want to have the space to be creative and to do things that are going to be really impactful for our, for other associates, their colleagues, or for our customers.

Our job with technology feeding into data and data feeding into AI is to remove, automate, simplify these mundane tasks so that we can give time back to our associates to do things that are inspiring, energizing to them, and impactful for both our customers and our associates.

And that productivity loop, that impact is not just something that we can reinvest in the customer experience.

We also reinvest in the associate experience, really helping them grow as we evolve the use of the technology.

Yeah, one last question to get you out of here on 8451 is part of your remit.

Everything we've talked about AI building these data platforms.

How is that Moving into the 8451 part of Kroger that a lot of our listeners engage with regularly?

Yeah, absolutely.

So I think the, the CPGs, the partners we have, who are very keen and very tuned into the insight that we've delivered and made available to them over the years as part of the 8451 portfolio, are also very keen on looking at what's next.

How does their ecosystem, their data signals, potentially further connect into the 8451 and Kroger ecosystem to create that superior experience they're keen on, whether it's for insight or media activation and audience design, how do they bring their data, their data assets and our signals together to create that special experience.

And I think that's we are moving the needle and moving kind of the limit of what we were able to do because of technical and data science constraints in the past.

And I think that's going to be quite transformational too.

Exciting stuff.

Exciting stuff, right?

Yeo, thank you so much for joining us.

Thanks for the teams at Vuzion and Grocery Shop for making the coverage of this conference possible and stay tuned for more.

Awesome.

Thank you for the time.