[00:00:06] Host: Welcome to the first episode of Built For Billions, a podcast by Flipkart Commerce Cloud. Where we share retail insights from the architects that built India's largest market place. Today we want to take a closer look at one of the most important shifts happening in retail right now, which is AI.

[00:00:27] Now AI has become the default strategic priority for most retail businesses today. Leaders are under constant pressure of showing movement on that front. Budgets have been allocated, pilots have been launched, assistance copilots are being taken live, and yet when you strip away all the headlines, what you are left with is,

[00:00:51] a difficulty in mapping the economic incentive or the economic impact of AI, in retail. Alright.

[00:00:59] Recent industry surveys show that most of the decision makers that use AI are not able to tie back AI initiatives to EBITDA lift. Merchandising teams spend as much as 40% of their time in low value manual work and also merchants have gone on record to say that AI tools have had little to no impact on their businesses so far.

[00:01:29] So the question isn't whether AI is important, it clearly is. The question is, where is AI moving the needle inside the retail organization? To unpack that we are joined with Sunil Jagadeesh today, Head of Engineering at Flipkart Commerce Cloud. Sunil works at the intersection of commerce infrastructure, search intelligence, pricing systems, and retail scale AI.

[00:01:56] He's also closely involved with building systems that power complex retail decision making here at FCC. Sunil, it's a pleasure to have you here, how are you doing today?

[00:02:09] Sunil: I'm doing wonderful. I hope you are doing

[00:02:13] well as well.

[00:02:14] Yeah and thank you for having me here and very excited to be part of the very first episode.

[00:02:19] Host: Absolutely. So let me start there, Sunil. So as I mentioned, there's a lot of visible AI activity across the retail space today. But from where you sit, what is the real gap between, the AI ambition and the AI impact? Or in other words, are we overestimating the amount of AI that is used in commercial decision making today?

[00:02:48] Sunil: The impact of AI is real firstly. The way I see it, retail has always had great enthusiasm

[00:02:58] for AI. Where

[00:02:59] I think we have lacked has been the operational penetration. And the reason for that is a lot of what we see today in AI is, what I would call as Layered AI. AI that is sitting on top of existing workflows.

[00:03:16] These are workflows that, probably haven't changed in decades. So we hear a lot of noise hype about copilots, assistants, chat assistants, shopping assistants. But if you look at it behind the scenes, important things like pricing is still rule-based, merchandising is still trapped in the bevy of spreadsheets and inventory positioning is still very reactive,

[00:03:46] and

[00:03:48] this fundamentally exists because of the fact that AI has traditionally been treated as a feature, it's a nice to have add-on, not necessarily something that's fundamental and embedded into the core execution system where the actual ROI and revenue are being generated.

[00:04:07] Host: So yeah, that is an important distinction. So what I'm hearing you say is that AI exists, the AI is present in most places, but not necessarily in places where decisions are being made.

[00:04:23] So that brings up an important point. As you mentioned, we see a lot of buzz around conversational AI,

[00:04:30] generative catalog content, AI assistants and the works. Sure, these are important and meaningful improvements and they can be quick to deploy, but are we somewhere confusing Visible AI with Transformative AI?

[00:04:51] Sunil: My perspective is exactly, there is confusion, but I'm very clear about this one.

[00:04:57] There's a fundamental difference, I believe between AI that improves interactions versus AI that governs a decision. Again, you talked about Gen AI. Gen AI has gotten to the stage where creating content has become fairly straightforward, easy does a great job actually. You have shopping assistants that can help guide customers, but look at, what it means for, core retail.

[00:05:26] Neither of these kinds of activities help improve your margin thresholds, nor do they move your pricing elasticity, for instance. They sit on the top of the

[00:05:37] workflow

[00:05:38] They're not in the economic logic, which is sitting underneath, which is why the AI is not able to make a difference. So if you want AI to move the needle, we need AI to be part of the decision loop.

[00:05:52] The core decision loop. Taking pricing as the example we were discussing, an operational pricing loop, typically will integrate competitor signals, it'll include your elasticity model and also guardrails that you might have imposed in terms of margins. So you don't want a run away system that, optimizes for things that you're not prepared for.

[00:06:16] It integrates all of these things into your

[00:06:18] execution system.

[00:06:20] See, when AI is in this layer in your loop, it becomes extremely powerful. This is when pricing is truly dynamic and it is responsive in real time. That's the difference, and we should we go beyond AI just being an intelligence and informing things to something that executes thing.

[00:06:40] Now you're talking agentic. So that's the difference.

[00:06:49] Host: So the difference is not with the intelligence itself, but if the intelligence is sitting within a closed decision loop. So that sort of brings me to my next point, which I was thinking about that often retailers treat these different systems like pricing, search, inventory management etc, as separate optimization systems.

[00:07:13] So they often work in silos, even if there is AI within these systems. So what would happen if this intelligence were to start speaking with these disparate systems and connect these systems across the retail value chain?

[00:07:32] Sunil: Magic! That's

[00:07:35] where

[00:07:36] I see the magic happening.

[00:07:39] So having these systems be siloed, which is the reality in today's world.

[00:07:44] It means that they're working cross purpose. Your pricing system reduces margins. You want to move volume, but it's unaware that there's a lot of demand for that particular product for which you're reducing margins.

[00:07:56] They're not connected. They're not talking to the same thing. Similarly, your search, for instance, it might focus on converting a specific, product when the reality is the inventory system is unable to fulfill it, so now how do you address these problems? You have to connect the intelligence.

[00:08:18] So when you connect this intelligence, then the objectives start aligning better. You have search and ads, for instance, the intent detection can feed both organic search as well as paid ads.

[00:08:32] Your pricing decisions are going to keep track of stock levels. Not do something in isolation. Finally, your recommendation systems. The value of recommendation systems comes from conversion of your recommended products. One of the things has to be that it has to be available for you.

[00:08:54] I'm in Bellandur [an area in Bengaluru]

[00:08:58] If you recommend me a product in Mumbai, how is that gonna help me?

[00:09:02] So that's the way you think about it. So the recommendation system then becomes smart in understanding and recommending products that are near your customer's fulfillment center.

[00:09:14] So all of these things requires what you would call as a unified intelligence layer, a common intelligence layer, it may sound simple on the surface, but the reality is that this is quite complex and this is where retailers should really have a strong tech strategy, tech architecture, and probably the right kind of tech partner,

[00:09:38] Plug FCC here.

[00:09:42] So essentially you want your business to be singing the same song.

[00:09:48] Host: Makes sense. So this would, this is a sort of a naive question, that it is clear that embedding AI into these systems whether it be merchandising, search or inventory management etc, that is where the real leverage seems to be.

[00:10:07] So why isn't the industry further alone? Is the bottleneck, the current technology that we have? Or is it something else at play?

[00:10:20] Sunil: Yeah I think, it's a mixture of things, it's not one thing. There are problems across the board in terms of architecture data, and there is some amount of process inertia as well.

[00:10:35] So in terms of architecture the reality, again, it's hard to blame anyone here. A lot of existing systems that were built that have been incorporated across most retailers and most enterprise systems to be honest, be it your ERPs of the world, CRMs, OMS, WMS, of the world they were built to not necessarily be interoperable, not necessarily talk to each other.

[00:11:00] That wasn't the game. That wasn't the game earlier, of course it has changed now. What this means is that, the contracts between the system, the understanding of what a product is in ERP versus what a product is in WMS, the identifiers fundamentally are different. AI thrives in having a common nomenclature, having this ability to get feedback continuously.

[00:11:27] When you have such disparate contracts, it breaks. So architecture is one of the fundamental problems where this falls on its face.

[00:11:35] Data.

[00:11:36] I think like data is just fundamental for any AI, ML model to work. You need to have data discipline. AI actually amplifies data discipline but if you have a poor discipline in terms of your existing data, it cannot fix it.

[00:11:53] You gotta fix that first. An elasticity model is only as good as your cost data. If your cost data is poor, then your model will predict gibberish.

[00:12:03] Again, similarly taxonomy - If it's inconsistent, then your assortment intelligence will fail. The final one is also important. It's just, a lot of organizations have gotten into this mode of changes through weekly reviews.

[00:12:19] I'll make changes to prices, assortments in specified intervals - weekly, monthly schedule. But this is not how AI works. AI thrives in shorter intervals, decision cycles needs to shorten. So to win you have to move from this weekly meetings to continuous automated guardrails.

[00:12:45] You need to let the system make these decisions for you. Recommend these decisions for you. With the guardrails, you define the guardrails, stay within that, but you need to let it on it. That's the difference. But the more important part of this entire thing, which is also a reason why a number of retailers are maybe still not as mature yet, is the lack of trust in tech partners in other parties that are helping enabling it because data

[00:13:18] is obviously very important, the sanctity of data, not just yours but retailers being your customers and those, these things have traditionally been coming in the way.

[00:13:33] Host: Yeah. This reminds me of, this saying that in God we trust for everything else, bring data. So yeah, that's the sort of situation here.

[00:13:46] Let's try to make this a little practical. Let's say I'm a retailer listening to this conversation, and I want to understand, so this operational AI, where would it, sit inside my stack? What would it be connected to? Who would it speak to?

[00:14:07] Sunil: I think this should resonate very well with our listeners. I'll take three examples, first, pricing is a very good example, because it's so impactful, if done right. So pricing as a closed loop. If you look at, an operational system using AI, it needs to be an engine.

[00:14:36] It's not just a report like, ML giving go, AI giving a report saying these are the things you need it to be real time. So here's where ML based product matching to look at competitive data, then feeding that data into your own, elastic model so you can figure out should I increase?

[00:14:57] Should I decrease, what do I need to do with those? Then applying game theory. Because competitors are not sitting idle, they're doing the same thing. They're looking at your data. So you need to understand how competitors are going to react to changes. So you apply game theory there, understand competitor reactions, and once it passes your guardrails, make it live automatically.

[00:15:20] No need for human intervention. The impact very real, direct on your business in the core decision making loop. Second is discovery and intent intelligence. This is about making your relevance and monetizations work together. We are all aware of multimodal search nowadays.

[00:15:46] Text and image. Just combine this with inventory aware recommendations. Your search and recommendation systems is taking care of your inventory. The system does not learn only from organic behavior, which is searches that you might be doing, but also from retail media ads.

[00:16:05] To finally ensure that, customer sees exactly what they're looking for, but also what you're able to deliver. You have to intersect the two together. The last one I wanna talk about is retail media optimizations. Traditionally ads has advertising has been something that is considered as a bolt-on.

[00:16:24] A separate thing will come, some other system will be brought on later. Not the most important priority. This is the way traditionally retailers have looked at it. But it doesn't need to be an external plugin, it should not be an external plugin. You should be trained on your first party commerce data as well.

[00:16:41] By balancing the price of the build, predicted conversions and the customer experience. You are ensuring that AI gives you the maximum ad revenue possible but without ruining the shoppers share, So that's what I would say.

[00:17:00] Host: So everything that you've described from pricing optimization to intent driven search, to retail media ranking, all of these systems in order to work, as you mentioned, they need to speak to each other and the way they speak to each other is via integrations. So they are stitched together using integrations, and with it comes its own set of problems, which is,

[00:17:25] a lot of integration fatigue and complexity.

[00:17:28] So as a retailer, when do I stop looking at AI as a feature? At what point? And perhaps start looking at it as a platform architecture decision.

[00:17:43] Sunil: So Integration sprawl is the term I would like to use.

[00:17:46] So integration platforms have, thrived in the industry. There are organizations that are built on integration platforms. My own background, my previous organization, we used to do this as well. The reality however is that there is a threshold, that you as an organization have to make a call on.

[00:18:05] It gets to a point where your integrations stop being effective because the ROI of maintaining and managing and updating them supersedes the ROI that you are getting in terms of your business value of connecting these disparate systems. This is when it is imperative that you start looking at the foundation.

[00:18:28] It's no longer about connecting systems through, such integration, you have to look back at your foundation and see, what needs to be done at that level, that's when it becomes an architectural decision. So if you are laser focusing on retail, for instance, if your search AI is one system, your media AI is another system, the pricing AI is another system,

[00:18:54] They will conflict with each other, because each of them has a different purpose. Search wants conversion. Media wants yield. But pricing wants margin. What are you optimizing for? If you do not have a shared platform, you are essentially paying for multiple systems that are in arguing with each other fundamentally.

[00:19:15] So, by making this a platform decision, an important thing that you get is proximity to execution. It's important to stay connected closer to the data and your buy decision. Like where is that buy? So you wanna reduce latency and you wanna ensure that every model in your system is learning from the high fidelity first party data.

[00:19:39] Host: So this is the other side of the coin. So integration brings complexity. The way to remove complexity is by embedding AI. But that in a way means that you would have to rip and replace the existing system, which most retailers as you know, wouldn't want.

[00:19:58] Because, the architectures have a lot of legacy built in and legacy fatigue as well into the picture.

[00:20:09] How do you then marry these two things? How do you modernize intelligence without having to fully rebuild the existing systems?

[00:20:18] Sunil: Yeah, you don't have to necessarily rip and replace. Yes there is a lot of CXOs, retailers that are concerned that this exercise of modernizing essentially means replacing everything right.

[00:20:34] It doesn't have to be that. There are modern patterns that can help you here. Patterns like modular adapters or compostable architectures. Essentially the way to think about it is modular adapters - these are the way, these are your gateways to your legacy systems. So you don't have to replace your existing legacy systems.

[00:20:56] You can keep them, you can use these adapters and you can continue to build on top of the APIs that are existing. Compostable architecture comes into play because this is your opportunity to rip out your legacy pricing system, bring in a pricing AI system without necessarily having to replace your inventory system.

[00:21:16] Which is how our rip and replace complete model would work. So from that angle, API still remains your foundation. You build on top of your API foundation, you separate out your business logic. Your business logic and the core logic traditionally has all been put together.

[00:21:40] It's all sitting in the same place. You don't want to replace the entire body. You want the ability for your brain to evolve, the intelligence to evolve continuously. So you segregate the brain, the business logic outside of your core systems. It serves two purposes, your core systems need to be stable.

[00:22:00] This is the one that is the core infrastructure, the serving logic, high availability, all those things. You need that to be stable, strong. But you want your business intelligence to keep evolving. So putting the business intelligence layer on top of no code systems, low code systems, this helps not just you as a core team member

[00:22:24] being able to make changes, but also enables other personas, your business users and other folks also being able to contribute to the business logic. When you put these two together, essentially you get to the stage where without needing to completely ripping and replacing the entire system, you are incrementally moving your system forward and brain continuous, stable body can remain stable.

[00:22:51] Host: It's not just about, let's use more AI. It's rather about centralizing decision intelligence across your pricing, search, merchandising, inventory etc. So let's go back to our hypothetical where I'm a retailer listening to this and let's look at it three years into the future. Three years into the future I see two things happening. So there is one group of retailers who are still experimenting with AI tools and other group of retailers, perhaps based on a conversation, have started embedding AI across the retail value chain. So what would structurally separate these two groups?

[00:23:39] Sunil: Yeah I think, it comes down to decision maturity ultimately. The way I see it is that the more mature winners, I expect them to make three shifts. One is, moving from this, we alluded to this earlier, moving from the

[00:24:02] periodic model - weekly meetings, quarterly assortment checks to a continuous model. That is going to be a foundationally important shift that I expect that these winner will make. Again with guardrails in place but the system AI helping them there.

[00:24:25] The other thing is also very important, which is what is your intent? What's your economic intent? What are we trying to optimize for? Is it margin? Is it growth or it's just inventory health? It's important to answer this question very clearly because AI needs clear target.

[00:24:58] If your goal is optimizing multiple dimensions, then you end up in that silo situation we were discussing earlier where you probably end up locally optimizing, but not necessarily your entire business operations, so that's important as well.

[00:25:28] The final one is the data part that I was discussing. I cannot overemphasize the importance of data. Mature retailers would treat data as a shared asset and they would work on it. They would optimize it, they would bring about consistency, and then look at all of the system build on top of it - pricing, search, media everything that's core part of retail, all of them feeding into one another.

[00:25:54] This would be the differentiator between the two cohorts in some sense. The retail retailers who stay in experimentation will treat AI as a tool, continue to treat AI as a tool for a specific task, very localized. While retailers who move the needle will fundamentally look at AI as the operating system that governs their entire commercial strategy.

[00:26:20] So that's a big difference that's the shift that I see, which will segregate the two set of people.

[00:26:02] Host: Sunil this has been an incredible conversation. Thank you so much for enlightening me and all of our listeners today. It was a pleasure to have you.

[00:26:12] So, I think what's changed with me is the reframe.

[00:26:15] Where the question isn't, are we using AI? Everyone is using some form of AI. The questions that we want to answer would be: Is my AI actually touching economic levers?

[00:26:30] Is it connected to decision making systems? Is it compressing my decision making cycles or is it just generating better reports? And also, most importantly, is it reflecting in my PnL? These are the questions that if you are a retailer today and you're thinking about AI, these are the questions that merit a bit of thinking at your end, and you can take these back to your organization.

[00:26:59] I hope this was a clearer view into what kind of AI actually moves the needle in retail versus AI that just seems like it does. My name is Shrikant. This is Built for Billions, a podcast by Flipkart Commerce Cloud. We will be back next month with a lot more conversations about how to build retail at scale.

[00:27:25] So please stay connected. Thank you.

---