Sathish:
“You have been in retail for 30 years and seen many transformations. How is the AI transformation on the retail side different from the others?”
Greg:
This is my 34th NRF show coming up. To tell you how long I have been at this, I was the guy who introduced Wi-Fi into retail point of sale as a product manager some 30-plus years ago. We are seeing the fastest transformation ever in retail, and in society as a whole, with AI. There is such a disparity in where retailers are in their readiness that it is almost scary, because a lot of retailers are being left behind, mainly because they underinvested in getting clean, accurate data for their inventory. That is particularly true when AI meets the store and ecommerce, because roughly 89 percent of all transactions still have a store component, either as an actual walk-in, which is decreasing, or as a fulfillment engine for local delivery, buy online pickup in store, or click and collect. That store-level inventory accuracy and data quality is critical to success. If you put AI on bad data, all you do is get really bad decisions faster.
Sathish:
“Is AI finally connecting ecommerce and the store? Is that actually happening now?”
Greg:
It is starting to happen, but AI is not really a technology in this sense, and it is not the AI that has to catch up. The problem is the quality of the retailer’s data and networks. It is a recipe more than a technology. There are no deficiencies in the AI itself, the deficiencies are in the data. So step one is clean data. Do you have an accurate view of what you have in your stores? Are the descriptions accurate? More importantly now, are the full descriptions on the website ready for agents to shop, not just a human looking for solutions? Do you have all the speeds and feeds in the right format on the website so the agents can read them? I found out that just having FAQs with a list of answers does not necessarily get you found. In the age of agentic commerce you actually have to have it in a JSON format so you are seen by the search engines and the bots.
Sathish:
“With all of this happening, where do you see the biggest opportunity for a retailer to spend and see a real benefit?”
Greg:
It really is to do better forecasting and to optimize the other journeys. Most people have optimized the walk-in store journey, but they have not optimized the journey where the store is a fulfillment engine for online, whether that is picking for buy online pickup in store, click and collect, or local delivery. Do you have the items in stock? Is the layout in the right order? Have you optimized it? Because you are paying extra labor for that order, and increasingly delivery fees are competitive wipeaways, so you are just adding labor. The other piece is optimizing labor throughout the stores during peak walk-in periods. We are also finishing a study on loss prevention: how to leverage AI to reduce shrink, whether theft or honest mistakes. Increasingly merchandise is locked up and consumers leave because they cannot get someone to help them. So AI at the store level can make associates smarter, faster, and more productive, and also lower cost and loss.
Sathish:
“The inventory challenge existed before generative AI, and people were using AI for demand forecasting. Has that efficiency improved?”
Greg:
It is a huge problem. We lost 1.7 trillion dollars last year to not having in stock what the shopper wants to buy, or it not being in the right place, or not knowing where it is. At the store level, retail is a living, breathing organism. You have theft, things get broken, and if you do not have an accurate view of what is actually there, you cannot sell it. If you do not know exactly where it is in the store, you lose all the margin to someone wasting time trying to find it. The typical consumer thinks retailers are in stock about 75 percent of the time, while the retailer thinks they are in stock 90 percent or more, and that delta is the difference between expectation and the actual experience, because things are locked up, in the back room, not on the sales floor, priced wrong, or damaged and not removed. All of that is lost margin from inaccurate data.
Sathish:
“With generative AI now in the mix, has that improved things, or does it depend on the data?”
Greg:
It has zero impact unless you have the data right. If you have the data right, it is incredibly impactful, because now you can position things and see how they are moving. With accurate store data you can initiate promotions to drive traffic, and if you look at the weather and know it is a hot day, those cold items might sell more, so you can adjust price a little or promote them to drive volume. You can add more impulse items for a walk-in shopper, or make better suggestions based on what an online customer has bought before. Amazon does this terrifically: before you check out your groceries, it asks do you need this, do you need this. One hack I use is that I never buy bottled water, but they put a frozen bottle of water in every bag to keep the fresh and frozen items cold, so I now have cases of bottled water just from buying groceries. It costs Amazon about four cents to protect the strawberries instead of a cold pack. Those small, data-driven moves drop right to the bottom line.
Sathish:
“Does a small retailer have the budget for this compared to a large chain?”
Greg:
They will not have the same budget, but the key blocking and tackling is exactly the same, and the large retailer actually has disadvantages the smaller one does not. Organized retail crime targets large retailers because they have the same store formats grouped in regional areas and can be hit fast, causing inventory issues. The smaller retailer often has a much better handle on their inventory, and as long as it is electronically categorized and accurate, they can use the same AI tools for better forecasting, pricing, markdowns, and labor assignment. It comes down to data quality. The recipe is: first, accuracy of the data, which might need RFID or computer vision to reach a more accurate level; second, the network, wired or wireless, to process all that IoT data in the store. Once you have those two, you can go any direction you want with AI. The biggest improvements we see today are personalization and one-more-item, forecast accuracy, and loss prevention, looking for anomalies around peak promotional periods when both employee and consumer theft rise.
Sathish:
“In ecommerce, personalization is easy because you track clicks and behavior. How does personalization work in a physical store?”
Greg:
Usually it is done through couponing and specific offers. There is an opportunity to give real-time offers in the store, but various legal challenges are pushing back on personalized pricing because it is being positioned as a negative, as if your price will be higher because of the color of your skin or your income level rather than your actual sales relationship with the store. So it is generally done in the app before you go into the store: hey, you are a frequent buyer of this, here are extra coupons, a two-for-one, before you go in, rather than trying to entice you in real time in the aisle. That is largely because of government and union pushback around shelf labels and pricing on shelf labels.
Sathish:
“AI agents are assisting customer support for online stores. Should those tools be given to store associates too?”
Greg:
At the customer support level it is a no-brainer. You have the history of all the calls and solutions, and it gets someone new up to expert level really quickly. It does not help your best people, who have heard all the stories, but for new people it is huge. The same is true in stores. If you have a handheld with access to AI tools for configuration, health information, or specific product details, you can serve the consumer better as they walk in. Zebra has a great approach, moving to slim models on the device itself, your own custom fine-tuned models, rather than using ChatGPT or paying for licenses. It is very fine-tuned to the needs of that specific retailer and associate.
Sathish:
“Product discovery has changed online, with people starting in ChatGPT and other tools. How will that impact retailers?”
Greg:
It is changing rapidly. I will tell you how rapidly: I bought the domain Retail Agent Con in March as a joke, for an April Fools that there was a new conference for agents only, no people allowed. But the world has changed so much that we are actually putting on a real conference by that name in two months, because with the rise of agents for shopping, pricing, promotion, and marketing, plus consumers having their own shopping agents, there is a whole new paradigm. Most retailers are not ready. I now have five real people in my organization and close to 30 agents, with a chief of staff that orchestrates the others, and I am learning how to manage that. A good salesperson does not necessarily make a great sales manager, but all of us are going to have to be agent managers. There are people with massive ecommerce businesses who will be invisible because they have not optimized for agents. I asked a room of major fashion retailers how many had optimized their sites for agents, and nobody had. I told them they would be invisible by Christmas. About seven to eight percent of retailers have made that transition so far.
Sathish:
“So retailers, not just ecommerce companies, should make themselves visible in tools like ChatGPT?”
Greg:
They need to at least make their sites agent-optimized. If they want to build additional relationships, that is another story, and I wish I were an expert on that piece. It is a similar challenge to delivery with Instacart and DoorDash: who owns the customer relationship and the data, whether it is DoorDash, Instacart, ChatGPT, Anthropic, Google, or Walmart. Walmart will work its own magic, but does a mid-range retailer have the same pull with these behemoths? At the very least, make sure your stuff is ready. We used to think it was SEO. It has nothing to do with SEO anymore. It is all about AEO and GEO.
Sathish:
“Perplexity has talked about a shopping agent where you just say order me a pizza and it finds the retailer and delivers it. Does that make individual retailers replaceable?”
Greg:
It is going to be fought, and it depends on how much value you are to that particular model. It is like retail media networks. We had the technology for about 15 years through Catalina couponing, which had the relationships with manufacturers and retailers, and they should have managed the whole retail media network. But there was a tug of war where the CPG companies wanted to own it and the retailers wanted to own it, and nobody had the technology to do both, so it took about 15 years before retail media networks played out. I do not think we have that much time to wait when it comes to agentic commerce.
Sathish:
“When online retail arrived, people thought it would replace stores, but 30-plus years later most retail is still physical. Will AI agents change that equation?”
Greg:
The agent is not going to fulfill, so fulfillment is still the question. We continue to see more going to ecommerce in every segment, but for large-format retailers their stores are their distribution advantage. For grocery, health and beauty, mass merchants, and fast food, proximity is key, so the store stays heavily involved in fulfillment. I expect more of the actual sales to keep moving to ecommerce and less walking in, which necessitates rethinking stores: not as big a walk-in area, but a bigger area for staged orders for pickup and delivery in the back. One data point slapped me in the face: 80 percent of all items in a grocery store sell less than one a week. All those spices do not need to take up shelf space. If you reclaim floor space for delivery, optimize picking, maybe use robotics, you lower cost dramatically, and because less is on the floor you also reduce theft.
Sathish:
“You have attended NRF for 34 years. Is robotics filling the shelves going to be a reality soon, or is it still a pipe dream?”
Greg:
It depends on the situation. Where robotics really takes off is inventory accuracy. Add shelf labels for pricing, robotics, and reading through RFID and computer vision, and you free up labor to get shelves stocked. People are still more efficient than robots at stocking, given the variety of merchandise and locations, and it is not cost-effective to automate that yet. But the biggest driver is that we do not have enough people to work the stores. Retail has grown about 30 percent in the US since 2020, yet the number of people working in stores has gone down, including restaurants. The average sit-down restaurant is missing about four people compared to pre-COVID, which is why service is worse and why everything is moving to kiosks at the front end. So you may be forced into robotics for picking and stocking simply because you cannot get enough people. The question becomes how do I open more stores when I cannot staff the ones I have, and the answer is automation and AI. I have to do more with less.
Sathish:
“Should retailers implement AI agents across everything, merchandising, supply chain, customer service, the way big enterprises are?”
Greg:
You have to prioritize the biggest bang for the buck. We look at where you have the cleanest data and the biggest opportunity if you get it right, and often one set of data crosses multiple use cases. If you get customer data clean across channels, you unlock upside on the top line, which is far more powerful than reducing cost, because cost is only about 28 percent of the P&L for the average retailer. Sales-related activities usually come first: personalization, pricing, delivery and promotion options, because they touch more of the income statement. Number two is delivery and cost of goods. It is estimated that over 50 percent of trucks on the road are less than 25 percent full, so using AI to fill return legs drops product cost dramatically. Everyone thinks of ChatGPT doing an invoice or a document faster, but that SG&A benefit is smaller than the others. Get the clean data in the biggest buckets first.
Sathish:
“Should agents with clean data be given full control, or do you always need a human in the loop?”
Greg:
There is no one-size-fits-all answer. The question is, if you need a human in the loop, what is the cost of the error if an agent goes off script. In our case, across about 3,600 different use cases we have run, we average about 15 times faster than a human. I rarely just let agents run and complete things unsupervised, but there are functions where the agent now does it 99 percent well when it used to only do it 40 percent well, because it has done it enough times and I have corrected it enough that it is almost autonomous. There are other cases where it is good but really does not understand what we are doing, and we cannot take its recommendations. Can agents be customer-facing? Absolutely. I misplaced my Sam’s Club card and, using Synchrony Bank’s online agent, had a new card on the way in 45 seconds, validating my transactions, zip code, and last four of my social, something that normally takes 15 or 20 minutes on hold. I do not see how a human does better than that. But there are other situations where you do not want the human out of the way.
Sathish:
“Where do you think retailers are getting AI wrong right now?”
Greg:
The biggest thing is poor-performing retailers who have not done the data work trying to use generative AI as a way to catch up, and they are basically running themselves out of business doing it. You cannot skip the data part, you just cannot. The second mistake is going for the home run first instead of building momentum internally for adoption. The third is not understanding or working with the culture change that has to happen. Some people think AI will just take their job, so they want nothing to do with it, while others think it is the greatest thing to help them. As a leader you have to read the room and overcommunicate that the goal is to do more with the same number of people, not to take anybody’s job. Stress that AI is most beneficial for all the things you wish you could do but cannot because you do not have enough people or time. Offload the mundane stuff and let AI handle it.
Sathish:
“Which retailers do you think are doing this correctly, deploying their data the right way?”
Greg:
There are many: Tapestry, Walmart, Home Depot, Kroger, Target, Amazon clearly. What they did was start around 2018 getting their data right. The way I picture it is like getting from Times Square to LaGuardia Airport. When Amazon went profitable on ecommerce in about 2018, those big retailers realized they were in an IT race and decoupled IT spend from the revenue line. It used to be that if revenue went up four percent your IT budget went up four percent, at about two percent of revenue a year. They said no, we are in an IT race, and started spending more and got their data right. So in the metaphor, those guys are already through the tunnel and on the expressway racing to the airport, while most retailers are stuck bumper to bumper on 42nd Street: get started, bad data, fix it, get started, bad data, fix it. Meanwhile the leaders gobble up market share. The top 12 grew sales about seven percent on average but profits 15 to 20 percent, so profit growth is about 2.3 times sales growth, because they got their data right.
Sathish:
“If you were advising retailers on how to prepare for the future, what would be your first advice?”
Greg:
The core thing is inventory accuracy and clean data. That is number one, and from there you can go any direction. Go through your income statement and ask where you have the cleanest data and the greatest opportunity. Second, make sure your online presence is optimized for agents, because we just passed 57 percent of all internet traffic to ecommerce sites being agents, and we passed that this month. If you are not optimized for that, you will be left out. Third is culture. You need to teach people that regardless of their role they are going to be managers of agents. Sit down, outline your job function by function, and turn each function into an agent. Maybe you no longer do all ten pieces, maybe you do two with a human in the loop, then string them together and reduce the time by 90 percent, freeing you to do other things. You also need a dashboard showing what still needs your approval, because when you are running so many agents you have to know where a decision is waiting on you.
Sathish:
“You said data is the foundation. What about synthetic data, given the tools now available to train systems with it?”
Greg:
I would not worry about synthetic data in retail right now. Retailers have tons of data, they are data rich and wisdom poor. It is more about which data is relevant, and the only way you know is if it is clean. I remember a gentleman from A&W Root Beer years ago telling me he had 20 different combinations of point of sale systems in his stores, so all his data came in differently. It might have a top-line sales number, but below that the combos were all different and not clean or accurate together. So it is about getting to clean data in your core areas and working from there. That generates the foundation. A single version of the truth, a data lake or repository with all your data, and tools like Databricks or Snowflake to extract it into the right process at the right time, available to store associates and managers for better decisions. The results go up significantly.
Sathish:
“You have been with IHL for 30 years. Has the way retailers buy technology changed, or do they still just buy the tool everyone suggests?”
Greg:
There is a lot more general-purpose technology coming in now because of the speed at which AI tools are created. Larger retailers are looking at startups, and in some cases building their own technology and selling it to others, which is vastly different. Lowe’s has an IT house, Walmart has an IT house, and they are selling their technology and logistics solutions to others. Previously it was very hierarchical and took 18 to 20 months to do anything, but 18 to 20 months ago we were pre-GPT-4, so we cannot operate that way. At the same time, those who did not upgrade their Windows platform last year are now in a memory crunch with prices up 30 to 40 percent, so it is a race to lock in prices. Retailers are deploying more technology than ever: point of sale, payments, self-service kiosks, computer vision, RFID, upgraded networks. They are spending roughly 60 to 70 percent more of their revenue on IT than five years ago, because the transformation is becoming an existential threat for those who have not been investing.
Sathish:
“In the next three to five years, where do you see retail heading? People are calling for a slowdown.”
Greg:
The slowdown talk is marketing, promotion, coordinated pushback, and some legitimate concerns mixed together, but it all comes back to recursive self-improvement and AI improving itself. Retail is going to change very fast. In three to five years I believe the job disruption will happen faster than the positives get fully baked in, which will have a negative economic impact first, and I do not like that. The active group pushing against AI and not learning it is actually making that disruption faster. That said, the retailers who are ready will become more efficient and profitable. I expect poorly run retailers going out of business at a level we have not seen since the start of COVID, while the biggest players get bigger because they are past the toll booth and racing ahead, opening more stores and fulfillment centers more cheaply and taking share. Fundamentally retail is retail, there is always a portion where you want to walk in and touch the product, but more keeps going to ecommerce, even groceries now. We also have to fix the theft and lockup problem, because in many cities we have made the in-store experience so poor that you might as well buy online and only go to a store when you need it right now.
