Scott Jennings 0:00:09.9
This is an expert roundtable around AI in the supply chain. What I really want to get across to everyone is the business outcomes and the business value of investing in your data, AI, and planning capabilities to impact your supply chain. What I wanted to start out with today was just a quick table setting on AI and where we see AI from a supply chain point of view, and then we will begin our lively discussion around our expert roundtable with our friends from AWS and BCG. Now, what I first wanted to get into was AI has permeated our lives in so many different ways since November of 2023, when the OpenAI model came out, and it's not just in the supply chain, it's not just in our business life, it's also in our personal lives. These two kids up here happen to be my kids. This is Hunter and that is Logan, and what you're seeing there is - just show of hands, did everyone, when they were a child, they had a picture taken with their soccer team or their baseball team, and it was always really awkward. There were parents up there as coaches, standing in an awkward way, and then somebody, your parents probably, put something on their desk for a couple of decades with your sports team.
Scott Jennings 0:01:30.0
My kids, no different, last year the exact same picture. This year, when they showed up to get their baseball pictures done, their teammates weren't there. The only thing was there was this panel of wood with feet marks on it where you had to go stand, and the kids came in, they stood there, they took pictures. Hunter did, then Logan did, and then a whole bunch of other kids. They never actually saw their teammates, and that's because this year there were AI pictures taken. They stitched together that awkward picture that we all have in an AI-friendly way and individually, here's what it looked like. They painted the blue sky and eventually they painted in all their teammates. I didn't have licensing rights to their teammates, just my own kids, so I can't show them, but this is what they look like, and this is how pictures are being done. So AI has permeated our lives in very, very different ways, and sometimes it gets it right and sometimes it really gets it right and let me show you an example of it really getting it right. This is my AI headshot from a supply chain conference called GS1 Connect, the global language of supply chain, that I had taken, and it is an AI headshot, and I think they absolutely nailed it. [Laughter]
Scott Jennings 0:02:40.4
They just got me, I'm going to update my LinkedIn profile, and I'm going to confuse a lot of marketers out there once I show up. [Laughs] AI is a good thing. In the supply chain, it is said you cannot spell supply chain without AI, and in fact, it's actually true. You can't spell supply chain without AI, and no matter where you go, there are different types and stripes of AI that you hear. Whether it's our friends from BCG talking about generative AI and reimagining your supply chain, or our friends from AWS talking about optimizing your supply chain processes into things like agentic AI in the supply chain. I know that Adam spoke around several use cases around agentic AI this morning, but just to define it for the group. If generative AI gives you a recommendation, agentic AI executes the recommendations with those very same large language models. So it's the orchestration and execution of the recommendations across your supply chain, which is really, really interesting, and one of the topics that we're going to discuss in our presentation today, because that's where the value is. It's not just knowing, hey, here's an idea. It's actually execution of that idea in a very, very slick fashion, and that's why you hear [?Gartner/Gardener 0:03:55.7] talking about the business outcomes in supply chain, because that is where the business outcomes are going to occur. It's with that agentic AI, if you're looking at where the puck is going as opposed to where the puck is.
Scott Jennings 0:04:08.8
Now, it's not just me saying that. It's the Association for Supply Chain Management. Very large organization of global supply chain professionals from a variety of different companies. Every single year they pull their membership for what are the top use cases, trends that we're hearing across supply chain, across the globe, and at the very top of the list is artificial intelligence. What's interesting about that is if you look at how they rank, it's the likelihood and the impact. So what they're basically saying is, we believe, our membership believes, that the impact is not only real but is likely to happen coming out of artificial intelligence, and I think a lot of that is linked to the promise of being able to execute all of those recommendations across the supply chain. Now, one of the things, one of the reasons I think AI is so popular is because supply chain is so hard. It is so much more complicated than other business units. Now, I'm not going to throw rocks at my friends at HR and finance, but maybe I will. It's really, really complicated. If you think about all of the components around risk, around pricing, around policy, around things that are outside of your control, that happen in supply chain, and then if you look at this example supply chain, which was really well drawn. That's Dell's supply chain. They drew out their supply chain and that's how simple their supply chain is. That's one company.
Scott Jennings 0:05:36.4
So imagine if you took the whole consumer electronics space and mapped it up on here. There wouldn't be room, and you wouldn't see anything. It would just be lines everywhere. But if you think about it from an AI point of view, what if you could decode that, take out all of the monotonous work that is repetitive and could be very easily removed by agents within the supply chain - that's the promise. That's what gets everyone so excited, is because it can help decode some of this complexity that we have seen for years and years and years in the supply chain, because your data doesn't live within the four walls of your organization. Your supply chain doesn't exclusively live within the four walls of your organization and never will, and that's what makes it so complicated, but also what makes it such a rich target for supply chain agents. Now, digging into some of that association for supply chain rankings, they also made a nice little heat map here - and just by show of hands, who's heard of the SCOR model from ASCM, used to be the APICS model? Supply chain operational reference model. It used to be plan, source, make, deliver, and return were the big processes. Now they have the digital version of that. So now it's - and I actually have to look at it now - orchestrate, plan, order, source, transform, fulfil, and return.
Scott Jennings 0:06:49.5
They built a heat map for their trends on where they're most relevant, and if you look at the plan piece, just focusing in on that, it's deep green for AI and it makes complete sense because a lot of AI is forward-looking, just like a lot of planning is forward-looking. So there's a lot of really good alignment between the trends and where their association and members see the value and what we're hearing across the ecosystem, whether that's on the Anaplan side of the house, whether that's AWS, whether it's BCG, or any of our other customers that are in the room as well, we hear the same things. AI is very relevant in that planning space as well as supply chain as a whole, and supply chain as a whole is where some of our partners play. Amar, who represents AWS, will be able to talk you through a lot of the tangent use cases to planning, because if you look across the way that AWS is looking at use cases, you can see the planning pieces. Demand forecasting, risk management, bottleneck identification, emissions reports and ESG, and of course, decision excellence around things like S&OP as well as IBP.
Scott Jennings 0:07:59.5
But in supply chain, unlike almost any other business unit, everything is related. You can begin a product development and end at customer service. It's all interconnected within that supply chain, which is some of the promise of AI helping decode some of the complexity of that - and at Anaplan, I know this is a very wordy slide, it's a little bit against my dogma to have words on slides, but I felt like there's so much top talk around agentic AI, I thought I would define it a little bit, put some words behind it, so that we can make it real. Adam walked you through that really good example around anomalies, and almost every one of these examples is there's an anomaly, and whenever you hear anomaly, the synonym for that is also, there's something in your data that's trending off a certain way, and the system has recognized that as an anomaly. In the agentic world, it's, okay, well, we have a shipment delay, what are we going to do about that? How do we go solve that problem? Oh, we have an overstock. We have an out of stock. How do we orchestrate a new order going out or perhaps an internal transfer so that we can avoid that out of stock? How do we do that? How do we kick off an agent in order to do that?
Scott Jennings 0:09:08.4
On the order quantity side of the house, detecting unexpected order changes, I think in the supply chain world, the default or delivered in full on time, it's like chasing Sasquatch. Everyone wants to do it, but we've never actually found that perfect order. But having the ability to have agents out there monitoring for that perfect order, that's potentially possible, we might actually find Bigfoot and maybe he's not so scary. On the pricing side of the house, I think everyone heard Charlie's example this morning around being able to reprice regularly to capture more profit. What if you could do that and had that ability to very quickly and easily set off a set of agents or AI processes that could capture the value of being able to reprice at more granular intervals, as opposed to once or twice a quarter, and doing that across the board in every single market. What if you could just isolate specific markets based on changing market conditions? That's the promise of what we're looking at from an AI and agentic AI perspective.
Scott Jennings 0:10:11.0
Then finally, there are some people that know a little bit about AI from Nvidia, also a great Anaplan customer. They recently released, 2025, this is for retail and CPG, some of the outcomes from improved supply chain operations, because there's no need to invest in all of this AI and great technology if you're not going to get the business outcomes. I was at a conference recently, and I heard the CTO of a major CPG company that said, if you can't find the ROI, skip it, and that's where a lot of these AI projects become so impactful is if you can attach it to real ROI, and if you look at what Nvidia is saying in their most recent 2025 study, what's relevant to supply chain? Boy, you look at the enhanced decision making or decision excellence, optimized inventory and supply chain management, improved forecasting, predictive analytics. That's right in the wheelhouse of what we're going to be talking about today, and to me, this is the promise of all of the fancy widgets behind the scenes, is that ability to capture actual business value. Are we excited yet? [Cheers] All right. That's what I like to hear.
Scott Jennings 0:11:17.0
So, with that, what I'd like to do is I'd like to kick off our expert panel, and I'm really excited today because we have some real experts in the space coming from different angles that have seen different things that are here to share their experience on how AI is impacting supply chain, not just from a technology perspective, but from a technology people and process perspective. So I'd like to invite to the stage Amar Sanghera - Sanghera, excuse me - and Shetty from BCG. Thank you, gentlemen. [Applause] So first thing, let's go ahead and do some introductions and we can start with you, Amar, if you could introduce yourself and then on to Shetty.
Amar Sanghera 0:11:59.5
All right. So, good afternoon, everyone. I'm Amar Sanghera. I represent AWS. I've been with AWS for close to four years now. I lead the AWS supply chain solutions go-to market and strategy. Very lengthy title. Amazon AWS like lengthy titles, which tend to describe what we do. So, my day job essentially is working with customers in helping identify the way they can leverage AWS capabilities, and many of our partner capabilities in implementing some of the changes. Essentially could be around improving their supply chain operations, could be around planning efficiencies, and also adopting core digital transformation and supply chain. I live in Michigan. In a prior life, I spent a lot of time implementing business applications as well as doing consulting advisory, and I used to be a supply chain planner myself. I used to work for Johnson & Johnson. I saw a board out there that Johnson & Johnson is here. So, hello to everyone who is from Johnson & Johnson. I used to be a planner at J&J Medical in Asia Pacific. I used to do planning for Ethicon Endo-Surgery. Shetty.
Shetty Abhijeet 0:13:10.8
Amar, thank you. Hi there, Shetty. I'm a managing director and partner with BCG, 20 years doing operations both as an operator as well as a consultant. For the last ten years, I've done supply chain transformations with big data analytics and an AI lens to it. It's interesting, ten years back, AI was not even a buzzword. We used to call it advanced analytics, and big data was the term for it. It slowly changed to AI, machine learning, now agentic. I've been on the frontier of all of that, 100 per cent of my time and effort is now spent on planning as a topic, and you can't do planning now without machine learning, without AI - and it's nice, Scott, that you teed up that page with the circle around it and a dark green in planning, because planning is where a lot of our clients, as BCG, are spending and investing their time and effort, trying to bring machine learning into their organization. Super excited to talk about what's happening at the rock face and what works - and more importantly, what does not work. All ten fingers are burnt, scars on the back. I know exactly what can go wrong in an AI implementation and where you should start, but also importantly, where you should stop.
Scott Jennings 0:14:34.5
Excellent. We're going to have a good time, and I would like to apologize. My headshot is incorrect here. I think we saw it earlier, the real one, that's the real Scott Jennings. But nonetheless, we'll kick off, and what I'd like to kick off - well, our first question is, what do we think are the biggest trends in AI and supply chain management that you see across the market? Both of you are working with a lot of different types of companies, coming at it from a little bit different angle. Amar, if you could kick off here around some of the things you're seeing, like the big picture. We showed those, that big, broad swath of AWS use cases. There's a lot that AWS is doing in the supply chain.
Amar Sanghera 0:15:12.6
Yes. So, one thing when we think of AI is, people assume it's the hairy scary thing which some nerdy kids are doing and somebody's using them. I think one big realization that we are seeing in the last five years is that people have now gradually moved from experimentation into using those things at scale. What was being done in pockets, in areas as experiments, is now becoming more mainstream, and it's down to two things which I look at. One, I think there's a fundamental realization that there is there is a profound - I would say - differentiation people can bring into their business processes by using AI as a supplement into the human decision-making process and into human efforts, into managing processes and outcomes. So that takes us back to the thing, which is one big event thing I can think of, like the AI being used is, now there are more use cases with ROIs because they are now being seen with a process and an outcome in mind of a business problem and a challenge of how they enhance the human decision-making process.
Amar Sanghera 0:16:31.3
The second big area around AI and machine learning, which I'm seeing, is that the vastness of the data, which is in the bigger ecosystem, it is now becoming far more easier to connect the data. Now, again, oxymoron. You're at an event where you say one of the biggest challenges is where's my data? I can't handle my data - but that's not it. The challenge is that there are capabilities to consume the data, the capabilities to manage and understand the data. What really is needed, from an AI perspective, is having a stronger data foundation of managing the data and looking at the data as your own investment. Being a business user doesn't only mean that you are responsible for managing outcomes, it also means that you're responsible for the underlying engine or the fuel for your operations and efficiency improvement, which is the data you are generating. I see a big trend now with the AI initiatives taking off, that the business owners and IT are now working together and not seeing IT as the people who are responsible for the data, they need to fix it. Business people are seeing more and more ownership. That's a big trend I'm seeing now of the leaders who are adopting AI at scale is bringing those two things together.
Shetty Abhijeet 0:17:43.5
Awesome. That's a good segue, Amar, into what I was going to say. When we started teeing up AI to clients about seven or eight years ago, it was all about experimentation. We are well past the experimentation stage. We are now seeing clients become extremely impatient about delivering value from AI and machine learning. They've had a thousand flowers bloom. Lots of experiments across supply chain, marketing, sales go-to market effectiveness. Now the language has shifted from proof of concepts to what's working at scale, what is delivering value, and Scott, as you said, no ROI, skip it. CFOs are getting involved in writing big checks, and it is no longer a blank check. Clients are also getting very, very smart and savvy about what use cases to invest in. The big use cases that we are seeing clients invest in, in the core and at scale, planning is definitely one of them. We are also seeing a lot of value getting unlocked in sourcing and procurement. Interestingly, gen AI is being used to scan raw invoices, go back in history, come back and tell you how is your rate varying over time and so on and so forth. So real hard dollars and value being unlocked from AI. We are also seeing a lot of partners and clients show interest in putting the entire value at risk and say, how can we deliver value for you, but also get part of the value that we unlock for you and start-ups are also becoming fairly smart about it. So that's in the core and the scale.
Shetty Abhijeet 0:19:22.1
At the frontier of AI, we are seeing a lot of reinforcement learning and agentic stuff happening, right? So agentic AI is becoming now one of those experimentation things that clients were doing with machine learning five or six years ago. People are talking already about autonomous supply chains. Scott, you and I were talking before this conversation. Our sense is 90-95 per cent of clients are still not ready to have completely hands off the steering wheel, agentic autonomous supply chain or operations. Humans still want to be involved. So what we are working on is agentic use cases that do a strong decision recommendation, but don't actually decide and don't actually execute on the decisions. We are also seeing a lot of interesting humanoid robotic stuff with start-ups happening in the reinforcement learning phase. Each one of you, I'm sure as a supply chain expert, would have seen Boston Dynamics, Figure, Apptronik. Their videos showing robots working fairly fluidly with human movements. Those are generic movements. They have not yet been transformed for operations, and so therefore we are seeing a lot of reinforcement learning - experimentation, again, happening in the humanoid robotics space.
Scott Jennings 0:20:40.8
Those are fascinating answers. You guys have both seen very broad swaths of the market. What I find most interesting is ten years ago, I was working with a brand that every single person in this room knows, and we were working with a supply chain team, and we were building an analytics project for them - I won't say who or what or what we did - but the only thing they told us is, 'We want to work with you, but you cannot tell IT,' and that has changed lock, stock, and barrel. Lock, stock, and barrel. I don't think we walk into a single company now where there isn't strong alignment between supply chain, between finance and IT, and I think that has been a marked change in the last ten years, and I don't even think there's any way that you could engage around this concept of AI, let alone agentic AI, unless there is that tight alignment between the two. So those are really good, interesting insights, and we're going to build on that a little bit with question number two, which is around, most supply chain organizations have experimented., and we touched on this a little bit with AI and agentic AI. What allows an organization to scale and actually prove value rather than skip it? So, let's start with you, Shetty.
Shetty Abhijeet 0:21:50.1
So let me address the second bit of the question, which is proving value. Like I said, it's very important to be laser focused and clear about what use case you're going to invest on. There is enough and more real experience-based data out there across organizations for you to know which use cases to invest big money in. This is no longer about you burning your dollars, experimenting within your organization. There's enough publicly available data out there, and Scott, he showed the use case map. I think those are the use cases that are delivering value. So, proving value is about identifying the right use case. We are all, again, supply chain gurus. The number one discipline that we have been drilled from day one as a supply chain operator is segment. Segment your business for the use case. Pick a part of the business where you can actually prove value that is large and material that unlocks dollars, that you can then fund the journey further on. So phase one, prove value in a large enough business unit but also contained. Don't go end to end; don't bite off more than you can chew. Prove value there, unlock it, and then go to phase two, which is a larger scaling effort.
Shetty Abhijeet 0:23:05.2
That then brings us to point number two which is scaling, right? Find a business unit and find a set of operators and commercial owners that are fully owning and sponsoring the effort. As supply chain, at the end of the day, the value is resident within supply chain but actually gets destroyed elsewhere, right? So involving a business owner is absolutely critical. Having a senior C-level client sponsor is crucial to the success of any effort. In fact, when we walk into client conversations, the first question we ask is, who's your sponsor? How senior are they, and are they from business? Not supply chain, but from business. Once that sponsor sees value, you've got somebody who beats the drum, inspires people, and then rolls it out to the rest of the organization.
Scott Jennings 0:23:54.5
Amar, I can only imagine how much experimentation has gone on within the AWS around AI and supply chain.
Amar Sanghera 0:24:00.7
So I'm going to respond to this from an AWS lens, but also share - because Amazon, right? AWS exists because Amazon wanted to scale, build capabilities to scale. That's how AWS originally was founded, is internally to serve up Amazon's need of high amount of compute, and also second, what you can think of is, it's publicly available, Amazon's secret sauce of innovation, which is agility and having the concept of single threaded leaders, creating smaller [?two pizza 0:24:30.9] teams, smaller teams which allow Amazon to really innovate at scale - that's something which AWS was founded on. So when we think of AI, I think experiments were not something which… Amazon has been working on this internally for a long, long time now. Using machine learning in our ecosystem has been an integral part, part of our robotic ecosystem, part of the way we distribute, plan, we forecast things. Now where it has now grown towards is now using these things at scale. So the whole concept of using single threaded leader, and also the concept of what we call as data stewards or data product owners. So converting your business leaders into product owners of an outcome driving capability is a single, I would say, foundational capability, which you are going to enable to make your AI investment successful.
Amar Sanghera 0:25:22.4
Don't think of AI and data as add-ons into your process. Think of those as foundational capabilities which need to be embraced by your teams which are driving those changes. So as we think of proof of value and the use cases, there needs to be a single threaded leader who is going to impact the change and be empowered to take decisions on technology, on AI investment, on investing in data, on bringing in partners. I think they should be enabled to assemble their own [?two pizza 0:25:52.7] team, if I may use the same term, say who they want to be as a part of enabler. Not having, here's my work I've done, I've identified an ROI case and now IT, please go execute, right? That that is not a recipe for success. The second thing, which is from a scale perspective, is also making sure that the barriers to scale are broken. So the barriers to scale are usually twofold, right? One barrier to scale is generally, you have experimented in one space, but the use case identified was so limiting and so specific to that business scenario, that its generalization is now difficult. So that takes us back to saying when you identify value, don't look at the value which is immediate, but also look at the value which can be generated over a period of time, which could mean sometimes not working on very flashy things, but working on things which can really scale across the organization.
Amar Sanghera 0:26:44.4
That's again a decision principle we use at Amazon of saying, when am I going to invest in something which can be multiplied 100 times very, very quickly if I want to replicate that? The second thing from a scale perspective, is also enabling and investing in openness, which means while there is a product owner, it doesn't mean that they are the only team which is going to be execute and scale. Enabling your teams to be embracing that whole culture, that if innovation is done in another team and has been adopted, how to take it forward and use it again, which means - again, I'm going a little technical now. Let's say, we think of the whole concept of composable architectures, of building capabilities once and reusing multiple times. How can we enable our business teams to adopt that process, that innovation happening in other teams and in other places, can be adopted quickly, not only from a business culture perspective, but also from technology. How open can your data be that it can be used again and again? How can I build applications which can be reused in multiple geographies, again, from a scale perspective, not build walled gardens and one use cases which look flashy but cannot scale on the broader scale as well.
Scott Jennings 0:27:57.6
Yes, that's fascinating, and it kind of leads into our next question. Before we do, I'm going to do a very informal survey. Ready to raise our hands? All right. How many of you today, your organizations, are building out AI solutions, capabilities that you're involved with today in the supply chain? Let's see. We've got one, two, three. Okay, so it's a piece of the room. How many of you are thinking about it? Okay. How many of you are not thinking about it? So somebody is not raising their hand, but nonetheless - [laughter] - the idea here is that most people are thinking, most companies we talk to are thinking about some are engaging in it. Shetty, over to you on the people, process, and technology point of view. Are most supply chain organizations ready to take advantage of the AI solution in the market? Before you answer, I'm going to put up some of the things that you've learned in the marketplace up here, just to seed the question a little bit.
Shetty Abhijeet 0:29:01.4
Yes, that's an interesting one. So I think the readiness is there. The challenge is with capability and then post the pilot, one of acceptance, and it's not in the technical functions. It's not in supply chain. It's not in IT. The acceptance problem is in the business side of things, which is why I started off by saying, if you don't have a business leader that's leading the transformation, the transformation is going to fail. It's not a probabilistic function. It is deterministic. The transformation is going to fail. So a few things that we've realized. I think leading with value is important. Proving to the CFO and the CEO and the COO that we are not - pardon my language - pissing away dollars here on just experimentative stuff, but realizing true value is super important. Which then brings us to the second point of how fast can you genuinely unlock value and put runs on the board? In our experience, six months at the most in a business unit should be enough to start delivering real value from an AI use case, if chosen well and if scoped well and if executed well. There's a lot of ifs. It's not easy, but it's possible, and once you hit that six-month mark, you've proven value, you can then rightfully earn, again, the license to invest larger in more parts of the organization and in more geographies.
Shetty Abhijeet 0:30:48.1
The last - you know, as we spoke about before this and as Amar spoke about as well, building both a scalable data approach, data governance approach, data management approach, having the right data fabric built in with the right data, so there's a big governance element to it, but also having a scalable architecture is super important. One of the big choices we make at the start of a programme is to say, what should your end state architecture look like? Should all data reside on one platform? Is that a planning platform? Should data reside on a data fabric? Amar and I were going back and forth, along with Scott before this, right? Who should build the connectors? Should the connectors be owned by the data platform? Should it be owned by a third party? Should it be owned by the APS platform that you select for yourself in the case of planning? These are big choices and these are one-way doors. You cannot really go back on your choice once you've made that choice, and that is where, candidly, clients call in a consulting firm like BCG because these are not necessarily intuitive choices to make within your organization, and so making the right choices at the start that allow you to scale and thinking from a scalability point of view, is what will then truly unlock the hockey stick curve of the value that you're really looking for, and many of these transformations are front loaded. You have a lot of investment at the start, but you need to be certain that you're going to realize a return on your investment, and the only way to do that is to set it up for scalability.
Scott Jennings
Before we have an answer on the next one, I'm going to move to the next one because I think it's right in your wheelhouse - and that's the role of data management. This morning, we heard about Anaplan data orchestrator, and I think a big value point behind that is typically when the data arrives at your beachhead, in a perfect world - in a perfect world, it's perfect. You don't have to do anything. I have yet to live in that perfect world. When that data arrives at your beachhead, it's like, whoa, what do we do with this? And that was the whole concept of Anaplan data orchestrator. When you get into the AI world, a lot of very smart people would tell you, you can't have AI without proper data management, and so that's what I'd like to tee up, Amar, because I know you see a lot of that in the AWS world.
Amar Sanghera 0:33:08.9
Yes, I mean definitely. So when we think of the data, supply chain is an interesting piece. So think of it this way. The data generated by customers, let's say if you're a retail or consumer goods, by customers browsing the shelves is important for your decisions. The data which is generated by weather patterns is important for your decisions. The data which is generated by socioeconomic or political activities is very important. The data of shipping containers. In fact, 80-85 per cent of the data actually is outside your organization. Okay? So now we are talking about problem statement of not only managing data, which is in your organization. I think the semantic I want to use here is, there is two parts of data management. There's the data management in terms of understanding the data which is in your organization, your bill of materials, your processes, the system generated data, and then there is the data which is very important for the decisions. Each of those are not necessarily the same. A lot of times when people start thinking of AI and others and start limiting the scope of things they want to do, and that's where I call and say, when you think machine learning and when people say AI, people use it interchangeably many times, confuse many of the use cases.
Amar Sanghera 0:34:20.5
Machine learning is, I give it data, I identify the patterns, and I come back with an answer. Artificial intelligence is, actually in the truest sense is, if it could also think and reason and use the logic, almost like a human could use intelligence of discerning between the data and patterns and make decisions out of it. That's the AI component part of it, and the AI typically needs far more data than what the machine learning models will need. Machine learning models could work, and that was the limiting factor to say, when I'm doing forecasting using machine learning models, I could use the data which is in my historical data and use that. Where the AI starts helping us is, let's say there's no history of the data. What if I could bring in the data from a lot more sources and bring in all of that dark data which was sitting out there and bring it in? So that brings me to the second element of the data management part, is how to bring the data in and make sense of the data. That takes us to the whole part of saying, investing in that data fabric or the data layer, which allows you to understand the supply chain data in the context.
Amar Sanghera 0:35:28.1
So that is where I would say we see our biggest set of challenges and opportunities when we work with customers, as AWS, is helping customers build those large lift of data in a much easier way, using the whole partner ecosystem like Anaplan and many other players who help us integrate the data and bring it together quickly, put a governance in place so that the bad data and the wrong data can be corrected before it hits your processes, and then also making sure that as the data is being used and generated, it is also created in a purposeful and a thoughtful way. You know, case example being a forecast is a forecast, but a forecast could mean multiple things for multiple people. What ends up happening is different teams start generating multiple forecasts, and I'm going to go back to Amazon example. Amazon generates forecast which is used in multiple different places. Is it the same forecast? No. We generate multiple forecasts which are purpose-built for multiple scenarios and reasons, but the data is being identified and people are very quickly able to identify based on the use which forecast to use where and when.
Amar Sanghera 0:36:36.6
So that is the whole part of data management is also governance and creating that data stewardship of saying, there is an owner of somebody who's generating a data, who is then the custodian and the people who are using it can go back to that custodian and work with them as a supply chain owner. So let's say you're building a supply chain risk application or a supply chain risk management application. Why should it only be focused for one use case? You could actually have an ownership mindset where you define what type of data consumers will need and how do I build a governance so that everybody can access it. I think that is where I see the future of agentic AI and others, because it's only maybe three years down the line or five years down the line, AI talking to AI. It is not very far off. So when that happens is when this governance construct, if they're not appropriately set, we will see a large amount of hallucinations and wrong decisions and things popping up. If you make those investments and then have a patchwork of now humans doing something and some agent running off rogue and doing decisions - which obviously, you've let take decisions with imperfect inputs, and then a mish mash of people then trying to rectify it. That's a nightmare scenario, a dystopian scenario.
Amar Sanghera 0:37:55.3
I think the right approach is to build the right set of governance around data, the governance around how we see the outcomes in terms of the AI being used responsibly and then making sure that there are no, what I would say, misuse of the data which is being generated by creating the data steward role in the organization as well.
Scott Jennings 0:38:18.4
One pivot on this. We're going to do a [unclear word 0:38:20.6] on the next one. We're going to take this data management piece for you, Shetty, and we're also going to combine it with the supply chain planning piece. So think about data and supply chain planning, putting those two together.
Shetty Abhijeet 0:38:33.6
Got it, okay. Super interesting. I was smiling throughout as Amar was talking about data because, I mean, every single client that I've ever walked into has said, we've got a data lake, and every single client who's data lake I've looked at, I've realized there's a data dump or a data swamp. So let me give you three golden rules on data that I personally follow and we follow. Firstly, the data lake needs to be a true data lake for the data that matters to your organization, starting with the use case. Even on forecasting as a simple use case on machine learning that now everybody has worked with for the last seven, eight years, there are 35 different data sets that any typical organization can leverage to forecast something, from temperature to day of the week to blah this, blah that, everything, customer data, etcetera, etcetera. Typically, we find 85-90 per cent of the predictive power comes from not more than seven to eight data sets, and those are the ones that you have to hunt for. So depending on which use case you are setting up and looking to scale within your organization at that point in time, identify the data sets and then follow rule number two, which is drill, drill, drill into the data set to understand the single source of truth for that data set. I in eight out of ten situations find that even something as simple as shipment data belongs in three or four different databases, different snapshots, different granularities, and there is no single source of truth.
Shetty Abhijeet 0:40:21.0
So identifying and really being almost anal about it to say, is this truly the single source of truth for this data set, is a critical question to ask. It's not a dumb question at all. It's a very smart question to ask, saying, is this truly the data that we are going to use for this use case? The third that Amar spoke about is having governance. This is not a once-and-done scenario. You need to have a physical nose to touch, shoulder to pat, or throat to choke within the organization for each data set and identified domain owners. If there is marketing data that is important for your use case, have somebody in marketing responsible for that data. If there is a consumer inside data, like IRI or Nielsen market data, have somebody from your consumer insight organization own that data, which means that when three quarters of data are suddenly missing, that person gets toned down or gets told to, or when data is completely full and sufficient, that person gets a pat on the back. Without that, again, you create errors that compound over time and you don't realize and then that goes on everywhere. So we spoke about three golden rules. The greatest opportunity for AI-enabled supply chain planning is for planning to become a competitive advantage for organizations.
Shetty Abhijeet 0:41:49.1
We haven't really spoken about what's affecting supply chains on the overall. There is tariff and trade wars. There is the complete disruption of supply chain post COVID. There is the incredible demand volatility where consumers are shifting preferences on everything from nutrition to the clothes that they wear and the brands that they use, and then there is a growing sense of start-up, disruption, and innovation that's happening in every single space. So supply chain is going to be distressed for many, many years to come. This is not and this wasn't just a post-COVID situation. This is now going to continue for the next many years, at least half a decade in in our sense. So then planning becomes important. How fast you make decisions becomes extremely important. We have walked into clients and have seen their planning cycles be eight weeks, 12 weeks. We've cut it down effectively to three weeks, four weeks. So a complete reimagination of your process and operations is the greatest opportunity in planning. It's not just planning as a use case. It's how you change the heartbeat and the operating model of your organization. You can come in and say, machine learning allows me to generate a forecast on day three of the month. I spend 15 days after that taking decisions, and then by week three, my plan is locked, loaded, decisions have been taken, and we are not doing reconciliation meetings anymore.
Shetty Abhijeet 0:43:18.6
If any of you have broken your head against a wall in a reconciliation meeting, we have done away with reconciliation meetings all together at clients. No reconciliation meeting at all, no touching the forecast unless it is truly crucial. Eighty per cent of human value add on a forecast is value destructive - and hard data to prove that. I stake my reputation on it. Eighty per cent of value add is value destructive. Let the machine run the forecast. Walk in, take decisions. This is fundamentally about change. So when you talk about AI, Scott, you have to as humans then say, where am I adding value, whereas what value the machine adds. I add value as a human in taking decisions after somebody has recommended a decision to me ,and so therefore AI truly is a competitive advantage because it allows you to change the culture, the talent set, and take better, faster decisions.
Scott Jennings 0:44:24.3
Excellent. Well, we're going to have to land the plane there. This has been a fascinating conversation. I think that everyone really understands the power of where we're going from a supply chain planning perspective, the value AI can add, some of the big use cases it can address relative to supply chain planning, and I think…
Amar Sanghera 0:44:41.2
I'll just add one thing. The human value add part, right? So just as an anecdotal input here, I work very closely with Amazon's internal teams which use AWS significantly as well. We generate close to 400 million SKUs are forecasted on a weekly basis and sometimes on a daily basis - zero touch. So from a forecast to a PO generation or a replenishment decision, 99 per cent flows through without a human touch and happening on a daily rhythm. I know it causes chaos for many people who have worked with Amazon because they're like, what is happening? Suddenly, there's a spike and then there's, oh! We are letting the machine play out because what the machine is able to calculate and build what I call as a continuous and a systems thinking approach, the biggest opportunity I think with AI is allowing the full system view of managing and replicating your whole supply chain ecosystem as a part of your decision-making process. I'm not saying that humans can't do it. Humans in those reconciliation meetings are supposed to do that, we're supposed to do that, but there are many factors which implicate that. People not being aware of data, incomplete data, and personality issues, being able to work together. Allowing machines with the agentic AI and generative AI, we can now have agents which are goal seeking, which can really automate and drive goal seeking behavior, which can improve the overall system of the ecosystem, and that's where daisy chaining many of these agents, what we call as multi-agent orchestration, which in Amazon, we are doing at multiple customers of ours, we are doing in planning is allowing them to say, here's a forecast, here's a supply plan, here's a risk plan, run me a simulation and generate this goal seeking using this multi-agent approach, which can give you those answers of what you're seeking for the decisions you were looking to build.
Amar Sanghera 0:46:35.9
I think that's the future. That's the one single biggest opportunity is, allowing the true utopia of systems thinking and supply chain. Even though it will be chaotic to start with, but after a point you say, yes, there's a system running here, and I can clearly see that in Amazon. It seems chaotic from outside. How do you manage it? But there's a system, there's a rhythm, there's a beat you can sense from the moment you walk into a Amazon warehouse or operations, you can feel the rhythm, and a lot of that is being powered by the AI and machine learning we use.
Scott Jennings 0:47:06.4
Well, this has been a fascinating panel. I really enjoyed the discussion. I'm glad that everyone was able to attend today. Remember, this is the best session, so if there's any feedback forms, you loved this session, let's do another one. Thank you very much, Shetty. Thank you very much, Amar, Boston Consulting Group, AWS, and thank you for attending.