Retail is among the industries best positioned to benefit from agentic AI. It runs on large volumes of data; involves constant decision-making across pricing, assortment, and promotions; and relies on highly repeatable workflows. McKinsey analysis suggests the upside is significant: Retail merchants could reclaim up to 40 percent of their time by offloading manual, repetitive tasks to agentic AI and devoting that time to strategic activities instead.
But how does a retailer get there? How can a merchandising organization turn agentic AI’s promise into reality? McKinsey’s Daniel Läubli and Maura Goldrick, two of the authors of the article “Merchants unleashed: How agentic AI transforms retail merchandising,” answered these questions on an episode of the McKinsey on Consumer & Retail podcast, hosted by Monica Toriello. The following is an edited transcript of the conversation. (Subscribe to the podcast.)
Agentic recommendations
Monica Toriello: There’s a lot of excitement about AI, but also a lot of skepticism. When it comes to agentic AI in merchandising, what’s real and what’s hype? What’s the most powerful application of AI in merchandising that you’ve seen at a retailer so far?
Maura Goldrick: It’s a fair point that there is a lot of hype. I’d say most of my retail clients are not really seeing a ton of value from AI just yet. That said, there is a ton of experimentation, and some are starting to see value in exciting places. They’re not brand-new use cases that didn’t exist before AI existed. Rather, it’s about taking proven recipes across pricing, promotion optimization, assortment optimization, or vendor negotiations and bringing AI into those places to make them faster, more granular, and generally more nimble and customizable to merchants’ needs.
There is a lot of excitement and experimentation happening across the retail spectrum. I’ve seen, on a microscale, individual merchants doing unbelievable things with AI. They are building their own agents and automating their own workflows. They’re demonstrating that you can see real-time savings if you spend the time training and building the AI infrastructure.
Daniel Läubli: We talk a lot about the automation potential: the 40 percent efficiency gains. But the quality of the work and the types of things you can do also change, right? An agent consistently does things the same way and works 24/7—and that also goes for things you would probably have deprioritized. In the past, you would have said, “No, that’s too much work,” and would not do it. Now you can do these things. Whereas you would have said, “No, this is too small of a benefit; let’s not go for it,” now you can go for it because the cost of doing certain things is much lower.
Especially in merchandising, a lot of activities are still manual, heavily exception-based, and not standardized. Two things I often discuss with my clients, particularly in grocery retail, are performance management and procurement. These were activities that were done manually in the past. You’d come in on Monday morning, review your dashboards, do your analysis, and have different questions. You would try to gather this information from Excel spreadsheets across five systems.
Now, these tasks have become much easier. The agent does them for you; it gives you the answers directly. You come in on Monday morning no longer looking at dashboards, but at a report detailing what worked and didn’t work last week, along with recommendations—which is a game changer. It was impossible to get that in the past. It’s a similar story in procurement, which was a human-to-human interaction with a lot of manual work. These are just two highlights from past work where we already see lots of benefits in retail.
Monica Toriello: Your article touches on some of this; it describes a day in the life of an AI-empowered merchant. What are the biggest changes you’ve seen AI make in a merchant’s day-to-day life?
Maura Goldrick: Imagine you are a merchant who has 20 of the best junior category managers you’ve ever had. Very often, we see merchants come in on a Monday morning, and they’re constantly tearing down what happened, why it happened, and what they should do about it. Half the time, they’re having to make decisions that are data-light, like “I’m out of inventory. I don’t know why, but I’m just going to throw more inventory at it.” Half the time, it’s the exact opposite: “We’re going to study this for days and pull a bunch of data,” then a week and a half goes by, and “Now it’s not even a problem anymore, and I didn’t do anything because it took me so long to get to the answer.”
What’s going to feel very different is that a lot of those things will be teed up incredibly quickly. They’ll come with recommendations. It’s not just dashboard after dashboard. It’ll focus your attention on the things you should be focusing on. These agents will have been trained to come up with ideas for you (see sidebar, “Merchant AI Accelerator: A suite of merchandising solutions”).
I do think there’s a journey that goes from agentic reporting to agentic recommendations to downstream connection, but it still requires a human in the loop before you get anywhere close to full automation. But I keep going back to the idea of having 20 junior analysts on your team who don’t know everything but are pretty capable—and the more time you spend apprenticing them, the better they get. That’s a big takeaway for merchants: Real apprenticeship of agents has to happen, whether that’s a central team or individual category managers providing feedback and guidance to agents—“this actually was a terrible recommendation, so you shouldn’t do that again”—which will only make the agents better over time. It’s going to take time. Nothing’s going to come out of the box perfect.
Subscribe to the McKinsey on Consumer and Retail podcast
Solving the most pressing business problems
Monica Toriello: In your article, you talk about phased implementation. If I’m a CEO or a chief merchandising officer, how do I start capturing value from agentic AI? How should I determine which business unit to start testing this in?
Daniel Läubli: There are many different elements you can get started with. The most important thing is to identify the most pressing business problem for you at the moment. It’s not just about installing a tool. When you start using agents, it takes real management attention; it requires real change management. You might rethink some processes. You might rethink some roles and team structures in your organization.
You need to start with something that is important to achieving your goals for this year and next. I have a few clients who say, “Let’s start with procurement. That’s where we don’t have enough capacity to have all these discussions with our suppliers, and we would like to increase our capacity through agents.” Or some say, “An agent can help me with performance management, where one of our biggest challenges is getting the recommendations together. That’s a top priority.”
Start with an element or use case that is important to you and drives a business result. Then go for it end to end—not just bringing in a platform where these agents can live or bringing in new tools, but really thinking through what it means, from how people should work to how you can reimagine the process. If you just bring in the agents but leave everything else as is, the impact will be very limited. People need to start working differently. Their roles might change; the skills they need might change. You need to train them.
Starting with a narrow use case and going end to end is how I would think about it. Then you can go use case by use case across the different functions to “agentify” your whole commercial organization.
Maura Goldrick: I completely agree with everything Daniel just said. Building out an impact-oriented road map for how you’re going to bring AI in is exactly the right place to start.
I would add that my new litmus test for whether a retailer is serious about starting a change management journey is, “Do you have a large language model sitting on top of your enterprise data warehouse, with licenses for all your merchants?” Will that alone drive billions of dollars? Absolutely not. But humans take time to learn, get comfortable, and trust. Merchants have massive P&L [profit-and-loss] accountability, and typically they can do things however they want so long as they hit their numbers—but that also means there’s tremendous risk to trying new things. So, generally, I find them to be even more resistant to change and less trusting of new technologies, new recommendations, and anyone coming in from the outside telling them how they should run their category.
Letting merchants spend time with AI to see what it could do for their business goes so much further than any center of excellence trying to tell them what they should be doing. Spending time getting their hands dirty and seeing what’s possible goes a long way toward building that trust over time.
Excitement and expertise
Monica Toriello: Most retailers are already piloting AI in one way or another. In your experience, what are the critical enablers or success factors that allow a company to scale up and truly benefit from agentic AI?
Daniel Läubli: It’s important to see this as a change management exercise, not a tech- or IT-led program. This must be business-led; it must create a P&L perspective. It’s dangerous to start with the agents and just hope they will somehow create value without being sure where that value will come from.
Then you need to ask, “What does this mean for my processes? What does this mean for my strategy? What does this mean for the talent in my organization, and how can I develop that talent?” Those are the right discussions to have.
Yes, it’s also important to organize your data in the right way—to have the knowledge graphs, ontologies, and everything in place—but that’s the last question I would ask. I don’t know if this is true for you, Maura, but in my discussions with retailers, usually the first question I’m asked when we bring a new idea is, “What data do I need for this?” And I say, “The data is really the last problem.” Often, the much harder problem to solve is, “What changes are needed in my organization, and how can I excite the organization to make those changes?”
Maura Goldrick: Retailers should also think about this as a capability-building journey. You fundamentally need to teach your merchants how to gain a new skill and a new tool they can leverage. There is a learning element that will be important beyond just changing people’s hearts. You need to make sure that they have the capabilities, that they have places they can go to ask questions, and that you have early adopters.
When I think about what’s going to be a game changer, it’s whether a retailer has “change champions”—those people who are going to be your tip of the spear. In my experience, you find them in interesting locations. It’s almost never the person whose title is AI engineer or anything like that. It can be a junior person who geeks out about this stuff and loves doing it on the weekends.
Be attentive to where the excitement and the expertise are actually growing in your organization, and let that flourish. I know a chief merchant who meets with a right-out-of-college analyst once a week, and the analyst shows him how to build his own agents for his Monday morning routines. It’s a reality that a lot of the most AI-native people tend to be more junior in the organization. It’s also a great role-modeling exercise to show that senior leaders are spending time upskilling themselves.
Daniel Läubli: You asked earlier, Monica, where a retailer should start. Actually, if you’re a retail leader, you should probably start by experimenting with AI agents yourself. I think we underestimate this. I see this with myself: At McKinsey, we now use thousands of agents for different things, and I try to create time to use them myself.
It’s not hard to build an agent, so just try it. Practice. Get a feel for what it can do and what the limits are. “Agents” sounds very technical, but when you start building them yourself, you realize the process is quite intuitive. Of course, when you start building agents at scale, then all the complexities we discussed come into play—but experimenting with them yourself gives you a good sense of what’s possible and what’s not. That’s also how your employees feel when they start experimenting with them.
What agents can’t do
Monica Toriello: You’ve both mentioned talent and skills. Do retailers now need to acquire new skills beyond traditional retail skills? Do they need to look elsewhere for talent in addition to upskilling their people?
Maura Goldrick: I do not think it requires a wholesale change in talent. It is helpful to start thinking about the “technical translator” skill set as something to add to job descriptions and something to look for more of. These are people who have deep business understanding but also enough understanding of the technical side to bridge that gap.
I would also say it might broaden the backgrounds from which you look for merchant talent. In an agentic world, you might be more willing to pull in someone who’s been on the CPG [consumer packaged goods] side of the house and has deep category expertise but hasn’t actually been a retail merchant—because you can teach them what it takes to get things on the shelves while building on their depth of understanding of the customer, the category, and the brands.
I’m incredibly hopeful that there’s a lot of upskilling possible with the current merchant set, mostly because I think agents will never be very good at deeply understanding the nuance of a category in the art-meets-science way. I do a lot of my work in apparel and specialty categories, and there’s always going to be some art to that. Even in other categories, there’s always innovation. There’s always newness. There are always choices you’ll need to make about how to differentiate yourself from other players in the market.
We talk about saving 40 percent of merchants’ time. That doesn’t mean you’ll need 40 percent fewer merchants; it means you’ll need merchants to spend their time thinking about how to compete with all the other retailers who are also saving 40 percent of merchants’ time. How do you differentiate yourself? How do you innovate? How do you stop spending time entering individual SKUs and doing item setup, and instead move toward more strategic debates about how you need to show up to your customers?
Daniel Läubli: If I worked at a retailer, I would not be scared that my job is at risk. Your job might change. The most cumbersome part of your job might get automated—and you might actually like that. Your role will change, and you might have to learn some new things, but it’s typically the most cumbersome parts of jobs that agents automate best. So I would feel quite positive about what’s coming.
