Shelley E. Kohan (00:03)

Hi everybody and thanks for joining our weekly podcast. I’m Shelley Cohen and I’m very excited to welcome two of Deloitte’s most sought after thought leaders in the space of AI and retail capabilities. So Roland, you are the principal and retail consumer products, AI and engineering expert. Welcome.

Roland E (00:23.016)
Thank you.

Shelley E. Kohan (00:25.013)
And Sarub, you are Deloitte’s US retail AI strategy and engineering lead. So welcome.

Saurabh Vijayvergia (Deloitte) (00:33.166)
Thank you, Shelley. Thanks for having us.

Shelley E. Kohan (00:35.412)
it’s great. we’re talking about a super hot topic today, which is agentic commerce, agentic AI, and our listeners cannot get enough of this topic. So if any of you have wondered how the world’s biggest retailers are using AI to stay ahead of the curve today, we have two guests that really are going to help us understand what’s behind the curtain and what’s making that happen. And both of you have spent

I know Deloitte Consulting has spent two decades really transforming how global retailers think and operate in the space of AI. A lot of us are just coming upon this in the past few years, but you’ve actually worked a long time trying to get AI and retail integration to work for major retailers. And you guys are on the forefront, the intersection of retail and artificial intelligence. So you have spent most of your career turning messy complex data

into razor sharp business decisions and you have saved companies millions. So we’re so excited to have you on today. So welcome.

Roland E (01:42.524)
excited to be here.

Shelley E. Kohan (01:44.748)
So we’re gonna, thank you. So we’re gonna start by first talking about agentic commerce. And my understanding based on Deloitte’s recent reports is that 25 % of global e-commerce would be agent enabled by 2030, which is like four years away, less than four years away. So let’s kind of start with the conversation about agentic commerce and what’s happening structurally.

Saurabh Vijayvergia (Deloitte) (01:44.92)
channel.

Roland E (02:09.826)
Sure, thank you.

I’ll start by saying, know, agentic commerce is not just another AI use case, right? It’s a new commerce channel. It’s a new operating model. And this is where agents help, you know, customers discover, compare, purchase, and fulfill demand, right? So retail, which has always been shaped by that interface between the customer and the merchants, is experiencing quite a bit of change. So stores, e-commerce, mobile marketplaces, social commerce,

know, change how demand was created and captured in the past. But with agent to commerce, we’re looking at the next generation of engagement, right? And at Deloitte, we describe this as a structural change. And we also look at this as an opportunity for intelligent agents to act on behalf of customers, consumers like you and I, engaging directly with retailer systems. I’ll pause for a second and let Saurabh chime in and I have a few more notes to make.

around that. Sorry.

Saurabh Vijayvergia (Deloitte) (03:13.71)
Yeah, thank you. Shelly, here’s the number that actually stopped me in my tracks. Between January and July of last year, AI-driven traffic to the top, say, 1,000 US retail websites, that rose to 4,700 % percent year over year. Now…

Most retail executives would have noticed this, even if this number, if this percentage was 47%, but in this case, it wasn’t. It was a whopping 4,700%. And by 2030, you rightly said analysts are projecting a quarter of global e-commerce will be enabled by AI agents. So when people ask me about whether agent e-commerce is real or it’s a structural chain or a step change, what I

What I begin to think about is the demand and supply side. Demand side is already there. Consumers are craving for it. They are using universal LLM models to be shopping. What’s not ready in our experience is the supply side. It’s getting there, but the retailers need to basically just ramp up their efforts.

Shelley E. Kohan (04:30.062)
So let me ask you a question, because that’s really fascinating. So if we go back to last year, you said January through July, correct? So retailers during that time, so all this is happening with agent at commerce, but retailers don’t have visibility into it, because they are looking at their websites, and they may be seeing perhaps a drop in website traffic, but they’re not seeing this traffic right now, is that right?

Saurabh Vijayvergia (Deloitte) (04:35.982)
Mm-hmm.

Saurabh Vijayvergia (Deloitte) (04:57.944)
They are. In fact, they are. The way I explain this to CEOs is think, you got to think about from a consumer perspective, think about the last time you did like a shopping task, you did not want to, planning a dinner for eight people, half vegetarian under a hundred bucks, or if you’re buying a gift for someone else whose taste you barely understand, or for that matter, figuring out what you…

really need for your kids’ school project. And that’s a true story, by the way. Those are the moments where a consumer can actually start to delegate to an agent. The retailers have begun to track and to be able to see those numbers. January through July, it came almost as a surprise as the technology was outpacing the adoption from a retail perspective. Now, when that

Shelley E. Kohan (05:31.583)
Sure.

Saurabh Vijayvergia (Deloitte) (05:56.566)
report was published is when most retailers started to track it and then started to at least think about responding to it. How well they’re responding is, you know, the jury’s still out.

Roland E (06:10.036)
And I’ll quickly pile onto that, even though I’d hate to do so, by saying some of our clients have already built agents and chatbots, and we’ve enabled some of that as well, that engage with the customer to do exactly what Saurav just described. So that customer journey has already been influenced and extended or been enabled rather by agents. So it’s happening. It’s been happening. So they are recognizing that traffic.

is that, you know, for the executives, the traffic, the conversion, and the loyalty economics are still changing. It’s dynamic, right? So that’s providing a sense that the next battleground is not just being top of a search ranking, right? It’s not just app engagement, but it’s really going to be down to the question of whether your products, your inventory, the prices and policies or fulfillment promises are understandable and executable by these agents.

Because we just talked about the retailer side of the agent, but also the customer side of the agent that’s going to be making those requests and engaging with your agent as well. Because this is truly agent to agent communication, literally almost hands off.

Shelley E. Kohan (07:26.402)
Yeah, that’s amazing. Yeah, go ahead.

Saurabh Vijayvergia (Deloitte) (07:27.022)
No. Shelley, would say, I want to add one more and it’s a quick story. I was talking to one of the CIOs and about agent commerce just a couple of weeks ago and the conversation was more around, more focused on AI. Now, most retail C-suite executives understand the AI part of the conversation. You know, they’ve heard about it. They’ve got a generative AI or an AI strategy slide for it.

irrespective of where they got the slide from or they built it internally. What they haven’t internalized is that this isn’t just an AI story. It really is an infrastructure story. And that’s the structural change. That’s the step change. The question is not, are we using AI? The question is, can an agent right now, today, read your product catalog, confirm your inventory, complete a checkout?

and trust your return policy. Because for most retailers, the honest answer is no. And that’s not really an AI gap. That’s a plumbing issue, if you will. And the plumbing takes longer to fix than the PowerPoint makes it look like.

Shelley E. Kohan (08:45.186)
Yeah, so, yeah, so to build on that, so you talked about the product catalog and I’m sure you’re talking about product attributions. And so let’s say buyers and planners who in the past have product attributions, maybe a couple hundred per product. But now if consumers are going into agentic search fields and they’re putting things in, tell me the best…

Saurabh Vijayvergia (Deloitte) (08:54.05)
Mm-hmm.

Roland E (08:54.366)
Correct.

Shelley E. Kohan (09:10.466)
you know, the greatest hoodie ever, or what’s the best wallet, or where can I find the most, you know, the best waterproof jacket? A lot of those attributes actually aren’t right now today in the product attributions. And so if AI agents are going through and reading all this, they could be missing out. So now it’s not just about the traffic, this whole issue of Gentic Commerce impacts buyers, planners, marketers.

Roland E (09:41.054)
literally everyone, if you don’t mind me jumping in. I think the example I’ll give or the thought I’ll offer is, if your product detail page is only designed for a human shopper, then you’re definitely under-optimized for an agent being able to engage and decide what to recommend. So to me, while the brand still matters, how it’s interpreted

changes.

Right. And in that agent minded or mediated environment, that brand becomes, you know, trust, the availability, the clarity around what your policy is. And I think to the point, sorry about sharing early on, what’s the quality of your data? Right. And what’s the reliability of your fulfillment? Right. So those structural things that need to be in place to enable it are critical. So to me, and when I talk to my clients at the C-suite, the big question they need to be answering is, are we discoverable?

Are we comparable? And are we selectable by machines in that journey to fulfill the customer’s request or demand? And that’s the big ask. And that’s where the conversation goes. And I’m planning a lab for a client in the next couple of weeks. And this is some of raging topics they need to get past. And I just want to reemphasize what Saurabh shared, where while we’re figuring out how to progress,

The decisions we need to make are mostly around readiness, an honest, introspective look as to what will enable this to happen, and this being what I just described.

Shelley E. Kohan (11:21.828)
That’s interesting. So Roland, you talk about readiness and I know both of you are thinking about readiness in my mind coming from operations and merchandising. In the back of my mind, I’m thinking how are you going to prepare all the people, the merchandise, the planners, the buyers? Like that’s a ton of, that’s a completely different mindset and approach to the business than we’ve had in our industry ever.

Saurabh Vijayvergia (Deloitte) (11:52.493)
Yeah. mean, look, the aspect of visibility is really important. You have buyers, you have planners, and you have marketers. If you’re a buyer, you’ve spent your career thinking about how a product looks on a shelf. Now, that shelf could be physical, it could be digital. In an agent-mediated world, the shelf is purely going to be algorithmic. The shelf is going to be

you know, a JSON payload, if you will. Now, your assortment strategy has to account for what’s really machine readable to what Ro was saying, right? What has clean attributes, has, what can be assembled into a regimen by an agent that’s trying to solve for a customer problem, not for a keyword, but the overall context. And if you’re a marketer, brand emotion has always, always been your leverage.

Agents don’t have emotions. They don’t feel emotions. They read structured data. That doesn’t mean brand is dead though. It means brand now has to almost show up as a verifiable claim. It has to show up as provenance, as certification, as reviews that are machine readable. You’re building, you’re almost rebuilding brand into the catalog and not…

into the erstwhile campaigns. And honestly, if you’re a planner, your inventory signals have to be real time. It cannot be out of stock because the trust gets eroded, not end of day, not batch. And if the agent cannot confirm stock in that moment, it moves to the next retailer, to your competitor, and will likely not even come back because the trust factor has gone out of the way. Now, that’s literally how

Shelley E. Kohan (13:28.235)
and

Saurabh Vijayvergia (Deloitte) (13:47.758)
agents are built to behave, how they are engineered to behave. And once an agent moves to the next retailer, like I was saying, trust level for that agent for your website goes down. So it just doesn’t impact the current transaction. It starts to impact the future ones irrespective of the kind of product or the kind of bundle that the agent is looking to buy in the next or the one after that journey.

Roland E (14:14.356)
And let me just add that our AI sourcing perspective argues that the researchers need to rethink what work gets done and how, not just simply who does it or where it’s done. And we keep in mind that the AI serves as a digital worker that actually reasons, adapts, and acts, orchestrating between human judgment and autonomous execution. So it starts describing.

like the power is in the hand of the agent that’s constructing what needs to happen. Which then means, for example, that as a merchant, I really start to perform more from the lens of, say, asking the agent to identify margin leakage or compare promotional performance or check inventory exposure or even recommend markdown actions. It changes how you’re

engagement happens across all of the actors that typically would be in that workflow. And this is where that discussion needs to move. It’s moving from purely productivity conversation to decision velocity, right? Because you don’t want to lose not just the customer but also the agent because like Saurabh described, that agent will move on because it’s sourcing on behalf of the customer, right? So to me, the retailers who win are those who will be able to

enable better decisioning faster across multiple dimensions. An AI or the agentic AI is purely just a connective tissue between those insights and the action it needs to take, which are all guided by the intent and guardrails set by the human or the customer. It’s a completely different mind shift.

Shelley E. Kohan (16:04.12)
So, my gosh, that’s so interesting and absolutely a different perspective. I wanna go back to something you said about branding. So you said branding, so branding the experience on the website, branding the app experience, branding the in-store experience, all the retailers are on board with that. But how do you brand machine logic?

Like how can you brand to that if you have agents that are out there looking for things? What does branding mean in the case of an agent?

Saurabh Vijayvergia (Deloitte) (16:42.702)
Well, I actually think you do not brand the logic itself. You brand what the logic can trust. In this world, brand shows up as structured proof, consistent attributes, verified reviews, like I was saying, provenance and certifications, clear policies, reliable inventory, dependable fulfillment of the order.

So instead of asking, does my website feel premium to a human? Does that human experience still matter? You need to ask, does my data make my brand legible? Does it make my brand trustworthy to a machine? To me, it’s really brand as machine verifiable trust. Because if the agent doesn’t trust your website and goes away,

then the brand, not just the brand, but overall revenue also starts to erode.

Shelley E. Kohan (17:48.612)
That’s amazing. yeah, so the other thing that is really fascinating about, we kind of talked about this previously, and this is this idea towards multi-agent systems. And so if you maybe could help our listeners who may not be as advanced as the two of you in terms of RPA and AI agents, and what is even a multi-agent system? What’s it mean? What’s it look like?

Saurabh Vijayvergia (Deloitte) (18:15.438)
Yeah, I’ll give your listeners another stat that I was reading in a report about a few weeks ago. I think it was Gartner, but probably it was something else, but just bear with me. the data point was 40 % of enterprise applications will have task-specific AI agents embedded by end of this year, not next year, this year. And that is up less than 5 % a year ago.

And that’s an eight-fold jump in just 12 months, Shelley. And a lot of that is not customer-facing. It’s actually inside the four walls of a retail enterprise. And here’s what I find interesting. Most retailers already spent the 2016 to about 2021 window building out robotic process automations, the RPA, the very rigid sort of automations.

the bots that could handle structured, repetitive, rule-based work. Now, what’s happening now is that RPA layer is getting an AI brain on top of it. The old bot could execute a process. The new combination can decide what process to execute. And that’s the jump. From automation that follows rules to automation that exercises judgment and then executes it.

Right? You know anything to add there?

Shelley E. Kohan (19:45.048)
love that.

Roland E (19:46.408)
Now, I think you’ve covered that pretty nicely, Saurabh, and Shelley, I say is, very pointedly is, the multi-agent universe is here now, right? Like this is in a distant ideology and at Deloitte, we’ve developed a framework for that and that is where we’re really leaning into advising our clients on how best to orchestrate that. What I want to underpin that is, for the multi-agent,

environment to function effectively, not operationally, is more so based on trust and that trust being on having the right levels of security or the ability for decisions that are going to be made between agents, right? And that’s the crux of it. We’ll get into a little bit more of that, but that’s just something I’d like to highlight as we explain what it is.

Saurabh Vijayvergia (Deloitte) (20:40.59)
Yeah, and as we’re explaining it, Shelley, it may just sound like it’s a technology problem. That’s not the hard part. The hard part for most retailers is not the technology. It’s really the operating model. If an agent is making decisions that used to sit with, let’s say, a director or a senior director, who owns that outcome now? Who audits it? Who retrains it when the business changes? Look, it’s very well-documented fact that most retailers can run a pilot.

We’ve had several reports all goodbye and ugly, but pilots don’t deliver value. Very few have figured out who owns an agent on the org chart. That’s not an AI or that’s not a technology question. That’s a, it’s an operating model question. It’s a governance question. It’s the one question that separates the theater from value. And quite honestly, I’d rather see clients spend the next six months

Roland E (21:28.699)
and sit on this bus.

Saurabh Vijayvergia (Deloitte) (21:39.02)
building that governance muscle, then rolling out 10 more pilots. The pilots are easy. The operating model is where things get real.

Shelley E. Kohan (21:47.158)
Is that what you mean by transparent governance?

Saurabh Vijayvergia (Deloitte) (21:51.374)
Transparency and governance basically come together, but the trust aspect, in order to build that, it should not just be driven by technology initiatives. It is a holistic capabilities of an organization coming together to make sure it’s an operating model and a mindset shift than just technology aspect of it.

Shelley E. Kohan (22:13.86)
So what advice do you give retailers and brands in where should they be investing to get ready for this agent data infrastructure that is going to be required and this governance? mean, to me, it seems like a daunting task for massive corporations, but even for medium-sized or small companies, what’s your recommendations?

Roland E (22:35.572)
I believe, Shelley, that, like you said, it’s going to require investments and not just uncontrolled experimentation.

Deloitte’s recommending that retailers invest in agent-ready data infrastructure in the right APIs that will allow the right level of connectivity, interoperability, branded agents, performance measurements, trust, and governance frameworks. Those are the core areas to focus on. And like Sarb was mentioning, the differentiation here is in the trust and governance. And those things go hand in hand. And that valuation perspective

is hinged on the fact that as AI stands in most boardrooms as a key topic, the financial impact may not always be visible yet in the P &Ls, but we’re still going to see more push to get this right and get the operating discipline in place, right? Because if you don’t have that in place, and I like what Saurabh said where, have accountability, right? Executive accountability.

and measurements for performance success, right? You’re most likely going to be in not the very best position, right? You want to make sure that you’re building those right foundations. I’ll stress them again. Having the right data quality, right? You don’t have to boil the ocean, but for what you’re trying to accomplish, get the right data quality in place. That feeds into governance. Make sure security is in place. Understand what’s needed from a customer consent perspective. Have the right fraud controls in place. That’s for

both parties who are engaging, both the customer and the retailer themselves. And all of these things require a huge amount of operational, operating model changes and shifts. If you get those things right, again, the underpinning is trust. The customer trusts you in that interaction or transaction, can trust their agent in what it goes to go perform. They’ve set the intent and boundaries and you, the retailer, can prove

Roland E (24:46.166)
Provide that brand trust by ensuring that you have the right audit and guardrails in place and engaging in the meaningful with the client. I feel like you would lose client share, I believe you would, the client, sorry, customer share, if the customer does not trust you. And if you don’t provide the right information to the agent such that you retain engagement. You’re no longer just interfacing with the customer, you’re also interfacing with the customer’s tool.

So it’s a different mindset, it’s completely different paradigm. It’s super exciting.

Saurabh Vijayvergia (Deloitte) (25:22.082)
There’s a line that I keep coming back to in all my conversations with my clients as well. The agent commerce catch up curve is steepening by the day, right? The reason is compounding. A retailer that invests in clean product data, real time inventory APIs, structured policies, signed agent identity, it becomes more visible to agents, which means

more agent-driven traffic, which means more signal about what agents want, which means their data gets better, faster. It’s really a flywheel. So the retailer that waits 18 months right now is not just 18 months behind. They are two or three flywheel rotations behind. And that gap doesn’t close by writing a bigger check later. So you better start now, because again,

that agent e-commerce catch-up curve is steepening.

Shelley E. Kohan (26:25.058)
love that. That’s a great end to our conversation. If that doesn’t make retailers hop out and get started on this, think this what you’re saying basically is if you wait 18 months, you’re to be six years, 10 years behind, right? And you just can’t catch up. So thank you, Sarb and Roland for being here. Loved having your conversation. I’m sure our listeners learned a lot from you. Do you have any last minute closing thoughts that you want to share?

Roland E (26:30.888)
Yeah.

Roland E (26:54.106)
I think for me, Shelly, is again, thanks for providing us the opportunity to have this conversation. It’s a super exciting and pertinent discussion that’s happening today. I would say, part seriously and also maybe jokingly, our customers, our clients, don’t necessarily have to feel alone in that journey. We’re poised and ready to advise in that journey and to save you from being two cycles behind on a flywheel. I just got that from Sara.

But no fear, we’re here to help. And we’ve been on this journey with a handful of clients already, and that’s being humble about the number of conversations we’re having in the marketplace. But we’re positioned to help our clients and advise them to figure this out. And by the way, we’re in the heart of it. We’re in that industry firmly, and as you can see today, super passionate about the topic.

Shelley E. Kohan (27:52.057)
I love it.

Saurabh Vijayvergia (Deloitte) (27:53.614)
Shahin, thank you again from my end as well. One of the provocative version of what I tell our clients is agent e-commerce will be won by retailers who stop asking how do we use AI and start asking how do we make sure AI chooses us? Those are completely different questions. And the second one is the only one that’s going to matter in 2027 and beyond.

And if you think that question, the second question resonates with you and you’re thinking about it, you’re losing sleep, don’t do that. Call us because we are here to help.

Shelley E. Kohan (28:31.224)
I love it. Well, thank you both for joining me. It’s been a great conversation and thank our listeners too for listening in.

Saurabh Vijayvergia (Deloitte) (28:38.67)
Thank you, Shalik.

Roland E (28:39.656)
Thank you.