AI in Supply Chain - Lumi AI Webinar

AI in Supply Chain Webinar

In this webinar recap, industry experts from Lumi AI,  and leading supply chain thought leaders explore how Generative AI is transforming supply chain operations. The conversation includes demand forecasting and inventory optimization to procurement and transportation management. This discussion dives into real-world use cases, measurable ROI, and the evolving role of cross-functional collaboration.

Key speakers include Ibrahim Ashqar (CEO, Lumi AI), Kunal Thakker (Walmart, Newegg, UPS, SEAIR Global), and Colin Kessinger (End-to-End Analytics, Accenture, Stanford Lecturer), who share practical insights and cautionary tales about adopting GenAI at scale. They discuss why many companies still struggle to realize bottom-line impact - what McKinsey calls the "GenAI paradox" - and how to move beyond dashboards to action-oriented AI solutions.

Whether you're just getting started or looking to scale your AI initiatives, this discussion outlines a clear, actionable framework to unlock value with GenAI in complex, data-heavy supply chains.

AI in Supply Chain Webinar Transcript

Louisa Burwood-Taylor:
Welcome, everyone, and thank you so much for for joining us. My name's Louisa Taylor. I'm the head of media and researcher at Profunda.

Louisa Burwood-Taylor:
We're a venture capital firm investing in deep tech to unlock bottlenecks in food and agriculture and planetary health. And today, we're co-hosting this webinar with one of our portfolio companies, Lumi AI, to dig into how AI is being used in supply chains. But more importantly, what is actually working. No doubt many of you feel quite overwhelmed with the rapid advancements we've seen over the last 18 months, and we know from various conversations that many organizations don't really know where to start, let alone get ahead of the curve and stay competitive using AI.

Louisa Burwood-Taylor:
So together with Liam, we've got, to awesome guests with us here today, who, as you'll hear, I've got a ton of experience and insights to share on this topic. We have Kunal Thakur, who is a global supply chain and e-commerce executive with 20 years experience around heading strategy and omnichannel operations for organizations at all levels, from startups to fortune 100 companies like Walmart.

Louisa Burwood-Taylor:
He's currently serving as the CEO of C, a global a logistics solutions and freight forwarding company. Prior to this, he served as the president for MTC Trade and the chief supply chain of said Thracian. And then we also have Colin Kessinger, who is a supply chain analytics leader, a teacher, a researcher and a practitioner. And over the last 30 years, he's worked with leading institutions and companies of all sizes, helping them to develop, deploy and adopt analytics solutions, working both on the technology and the processes necessary to make them successful.

Louisa Burwood-Taylor:
And Ibra, would you like to introduce yourself? My co-host.

Ibrahim Ashqar:
For sure. Yeah. Luisa, thanks for that. Ibrahim Asghar here is CEO and co-founder of Lumi AI. Prior to founding, Lumi spent my entire career in data analytics, primarily in the realms of supply chain data sets. I was the director of data science at stored, a unicorn supply chain technology and fulfillment company. And before that I was in consulting within Deloitte's Artificial intelligence practice building enterprise, create data products for, fortune 500 companies, primarily in retail and supply chain.

Ibrahim Ashqar:
So that's been my my bread and butter and really excited to, be co-hosting this here with you today. Luisa.

Louisa Burwood-Taylor:
Fantastic. So, as you can see, we are in great hands. And before I kind of kick things off, I just want to let you know that we will have a Q&A portion at the end of this event, but please do pop your questions in the channel. And then later when we open up the Q&A, you can raise your hand if you want to ask your question live.

Louisa Burwood-Taylor:
We can bring you up on the screen. So to do a bit of, scene setting, we thought it might be helpful to share some stats around how corporations are investing in AI today. And as you can see from this first slide, it has grown significantly in, in the last less than a year. And particularly among the larger businesses that have more than $5 billion in revenue, there's been a over 400% jump in spending, in less than a year.

Louisa Burwood-Taylor:
So I think the takeaway is, you know, if you're not investing in AI as a business, you're probably being left behind. And then more specific to supply chain companies. Here you can see this is a McKinsey survey. The global supply chain lead a survey that was conducted, last year. You can see where exactly the interest lies, in applying AI to supply chains and so I think this is where I'll come to you now, to start, if I may, if from your experience, you know, where do you think supply chains could benefit most from AI?

Louisa Burwood-Taylor:
And does that reflect what we're seeing in this chart?

Kunal Thakker:
Absolutely. Thank you. Luis. And Ibrahim, first of all, for hosting this. I think it's going to be, super informative here. So glad to be, joining the team. You know, this chart seems very much aligned in terms of, my background, an experience that I've worked whether at Walmart or at Rossio or Newegg or more recently at Sia, where, you know, our investment in AI, as well as our applications for AI in supply chain have been more aligned towards some of these bigger buckets.

Kunal Thakker:
When you talk about real world use cases where we've, you know, leveraged AI, one of the biggest has been, inventory forecasting and demand planning. Hands down, taking both the external and the internal factors, whether it's making sure that we understand what our sales have been, what the trends have been, what the market has reacted to.

Kunal Thakker:
At the same time, taking into consideration any type of promotional activities, etc.. Plugging that all as clean data and then leveraging generative AI to help us, create, and make smart decisions, with leveraged in part in coordination, with machine learning models do with integrated sales velocity, seasonality and promotional lift. We've also enabled skew rationalization and automated replenishment decisions.

Kunal Thakker:
To have help better in-stock rates versus being out of stock, at the marketplace or, on our website. With that being said, one of the, two case examples I can share is about, most recent experience at Rossio, where we had a ton and ton of inventory, that we had built up, post Covid, and had to optimize it to make sure that, you know, our working capital and cash flow is in line.

Kunal Thakker:
And it helped us save over $100 million, in supply chain cost savings when it came to inventory management. With the right demand forecasting, and planning. So that's just one of the many examples. But, typically the way, we've also used, I is think about it in terms of a flow of information, funds and goods.

Kunal Thakker:
So when you think about the flow of goods, I personally leveraged it, for transportation optimization, saving over 8 to $10 million with, you know, where we had, well, 100, $150 million transportation spend, for forest, middle and last mile, using AI powered, transportation management tools to analyze real time freight data, automate on carrier selections.

Kunal Thakker:
You know, reduced detention, demurrage and last mile costs at the port, by being more, proactive versus reactive in terms of shipment coming in, scheduling for appointments, etc.. On the that's on the flow of goods and the flow of information, leveraged it with vendor scorecards and predictive alerts, to improve on time and follow or, or if, if you will, rates to be over 95%, developed AI driven dashboards to flag underperforming SKUs, underperforming vendors.

Kunal Thakker:
So we can then make a very proactive decision on whom we are using and how we are leveraging those vendors. And of course, enabling more productive or proactive risk mitigation and better supplier collaboration, which is again mentioned on the chart here as well, in terms of risk assessment and simulation, being under risk and transparency chart.

Kunal Thakker:
Last but not the least, the flow of funds application has as a selective example I can share real time is on the returns reduction and profitability insights. So reducing returns, by around 15%. And improving, customer lifetime value of evangelize customer return patterns to understand what's happening, why is it happening. And then at scale going into taking action items like refining the packaging or product listings or policies on behavioral insights and analytics, coming from, generative AI.

Kunal Thakker:
So those are just some of the very high level real life examples for us, across the board, whether it's a CEO new at Walmart or even now at Seer, including implications on contract management. And vendor management.

Louisa Burwood-Taylor:
Right. Thank you so much. And so, Colin, keeping things a little bit more high level for now, but what is your point of view on, you know, supply chain challenges and the use cases and benefits of AI and also perhaps why it isn't useful.

​Colin Kessinger:
Yeah. It's interesting. You look at a list like this and I'm largely convinced that it hasn't changed in 15 years, too. But as CIOs over and over and over again, where they were investing and and what were the things they cared about, these were the, the likely candidates. I think the interesting thing about it and the important thing to remember is that, I think I see the same failure modes ahead of us, right.

​Colin Kessinger:
Which is, you still need the right set of users. So technology's made a lot of things better. It's going to automate a lot of things. It's going to make a lot of things easier. Some of the insights will be even better, but probably a few more solvable problems. But at the end of the day, you have to invest in the people along with the technology.

​Colin Kessinger:
So like, we're forever sold and every hype cycle that, hey, the technology will do more so much of this that it no longer matters who's doing it. I don't think we're there yet. So, one, I think it's important to have a reasonable, set of expectations associated with the technology. The second thing is, and this, you know, speaks to a lot of technology we see is there's still a huge premium on asking the right questions.

​Colin Kessinger:
Right. So there's a whole field now called prompt engineering, which is a glorified way of saying ask the right questions. Right. That's all it is, is asking the right questions. And you're going to see, as you engage with these technologies, that you need the expertise in-house to be able to ask the right questions. And so this is another way to talk to you about the training.

​Colin Kessinger:
And the skill set that you need to invest in, in, in addition to, to the solutions that you may, pursue to it. That being said, you know, this list currently, as it did with Canal resonates this with like my current clients, we ask before we engage in a, across the board in all these areas, and, I like the tool set that we're seeing.

​Colin Kessinger:
There's every reason to be optimistic. I think the expectations are higher, for all the functions. Right. So we live in a world of continuous improvement. Even if you have something today, with this technology, it will be better, there is no doubt about it. And you just have to get it in your mindset, in your investment mindset that you're going to be continually investing in these things.

​Colin Kessinger:
It's not going to be a one off, endeavor at all.

Ibrahim Ashqar:
Yeah, yeah, I think I completely agree here. I think there's it's not a surprising list in my opinion. It's the same set of actors that we kind of hear and heard about for, for years. And it's like, okay, these are the depressing topics that supply chain leaders, executives in general really care about. I mean, if you look at the ones which get the most, visibility into and supply chain visibility, obviously it's forever going to be on top there, and it's always one of these topics that just sounds easier than what it actually is.

Ibrahim Ashqar:
Especially if you're in a company that has many disparate systems that don't really talk to one another. There's just like this harmonization that, that you kind of need to do before, so that's always a bit of a challenge. You know, for example, you want to calculate a certain metric doc to stock, for example, you want to see when something arrives and when the inventory is put away, one of the warehouse management systems, may be collecting these timestamps, but the other may not.

Ibrahim Ashqar:
And so it's would be impossible to get this holistic view of that metric across the entire organization without having some sort of changes to your receiving or put away processes in general. And, you know, it's like some of the bigger challenges with supply chain visibility to begin with, demand forecasting, of course. It's probably the bread and butter or the Holy Grail or something companies always want to get better at.

Ibrahim Ashqar:
There's all sorts of techniques starting from very simple moving averages to really more complicated machine learning algorithms and, and things like that. And I think it's actually a good point to note here in that the term AI is quite broad and maybe some level setting on, on what we mean here, because I do kind of see a bit of like, a bifurcation of, of this term.

Ibrahim Ashqar:
Right there is what we considered, in my opinion, this narrow traditional view of AI, or machine learning, rather, where it's like a system that's really good at doing one job. One specific job. Really? Well, like, you know, predict how many units of this one skew are we going to sell next week? So a very demand forecasting type of use case.

Ibrahim Ashqar:
And you can't do anything else counting or a picture of a car that can tell you a Joel kind of how to pull, can do all of these things. Whereas there's these there's this other subset, this growing subset of generative AI, which is kind of these, you know, the, the, the ChatGPT is of the world powered by much larger like large language models.

Ibrahim Ashqar:
And they're capable of you know, a much broader set of capabilities. And so I think for for the sake of this conversation in this webinar, we kind of want to anchor the discussion around, you know, generative AI rather than traditional machine learning type algorithms for supply chain, because, you know, that that could be covered in another topic, as a whole.

Ibrahim Ashqar:
AI pillars, you can go deep into those topics. But yeah, for this conversation, I think it's generative AI in and supply chain that use cases. And if we actually skip to the next slide we can we can see. And the adoption here has been pretty pretty massive. I've I've personally spent time on in both sides of these camps.

Ibrahim Ashqar:
Right. Building initially start off my career building machine learning models to improve decision making for supply chain. So that was like more traditional use of AI. And you can kind of see it was, you know, adoption kind of was stalled over the decades. And, you know, you know, more, more increasing more recently. But if you look at the adoption curve of AI, that's, that's been growing quite rapidly.

Ibrahim Ashqar:
Right. And now with Lumion K, that's kind of what we do. Right. At the heart of our technology is generative AI and how to use generative AI to streamline the process of extracting supply chain insights from large ERPs. So, you know, that's kind of me spending time on on the generative AI side. But adoption here is unlike anything I've seen.

Ibrahim Ashqar:
And it's it's really such a, a cool thing to witness. It's like a new phase shift in technology, so to speak. Kind of like the internet kind of born again. But, you know, just some something I think is going to continue to grow. And I think if anything, in a few years from time, I think majority adoption will be more from Jenny.

Ibrahim Ashqar:
If anything, it's not saying traditionally AI is going to go anywhere. It still has its very important use cases. But yeah, the the Jenny piece, especially as it's getting more embedded across different functions, is going to be quite, quite huge here.

Louisa Burwood-Taylor:
Yeah, absolutely. And so if we look at this next, next slide, we can, we can see the impact of adopting Jenny has had on various different business units. Between between years. And you know, it's interesting to see supply chain is, is that, you know, at the top as, as one of the functions that has reported the largest decrease in costs through the use of Jenny.

Louisa Burwood-Taylor:
But at the same time, you know, at this McKinsey report is just come out this month has indicated that, you know, the bottom line impact is not really being seen at the enterprise wide level. These are the nearly 8 in 10 companies report using Jenny I yeah. As many will report no significant bottom line impact. And they're calling that the Jenny AI paradox.

Louisa Burwood-Taylor:
What are we going to say. Thank you Brian. No okay. Yeah. I mean, just interested, you know, in in what you guys think around that and how how we can see this kind of shifting and how companies could start to see more of a bottom line impact from from using Jenny across supply chains.

Kunal Thakker:
You know, and, is a is a is a good call out here. And I think that as Colin also mentioned, that when you think about Jenny is going to work, is going to become a part of the day to day operations. In my opinion, I it doesn't replace supply chain fundamentals. It scales them. Right? I mean, when we aligned with the core business, it is a lot going to be, you know, an execution engine versus becoming our strategy.

Kunal Thakker:
But with that being said, if we take an approach and I'm just going back to the fundamentals here, that, if we start with a business problem and not an algorithm, we will see results. If we build the data foundation before trying to create a very sophisticated model, it'll work. Prioritizing use cases, that will help improve speed or service levels or margins is going to showcase a good ROI for us.

Kunal Thakker:
Right? It's going to show as an impact. You know, I in my, again, opinion, should augment operator decision making, not bypass it. Right. So when you start looking at how it works as a catalyst more than anything else, we will see the impact both in the tangible and intangible way. A lot of times the expectations are that, hey, if I'm leveraging AI, are investing in AI, I should be seeing results overnight.

Kunal Thakker:
That doesn't happen. It probably maybe is making the operator more efficient and it's bringing in more operational excellence. So, I would say there are certain use cases where it doesn't give us the impact that we are looking for. And I'll give you an example from one of the past, work places is, you know, thinking about, hey, it generative AI is going to help me predict and pivot, right?

Kunal Thakker:
It can give you data to help make a smarter decision. Right. But, you know, blockchain has been similar like that, that the expectations were so much it became a buzzword for awhile. But at the end of the day, you know, if you really want to see an impact, start small as a business. That's my recommendation. Start thinking about putting in a service or a cost or a capacity outcome that we want out of it.

Kunal Thakker:
And then, leverage it as an application to allow, you know, scale the business or is thinking that is going to make the strategy and the decision for you.

Ibrahim Ashqar:
Yeah, I think I agree. Cornell's so, so much with that point on like starting small and that that may be and I'd love to get your perspective, but I wonder if that may be kind of why we may be seeing this Jennie paradox in a way. In the sense that a lot of these like, start small, prove the value scale up, is still happening.

Ibrahim Ashqar:
We're still in that kind of that, that, that, that journey or that phase in that while we may be seeing some ROI on, on smaller initiatives, they're just not materially been scaled across organization to impact the bottom line as much. You know, I think I've got other thoughts as well here, but Kyle, I'm curious to get your take on on this as well.

​Colin Kessinger:
Yeah. I think, you know, one of the interesting things for me as I watch these journeys is, if you think about your bifurcation on that, the prior slide about. So I very specifically and, but I'd call, decision, making capability versus Jenny I which is content creating. Right. You can think about the nice thing about Jenny is it opens the door to a great number of applications that we didn't have before.

​Colin Kessinger:
Right. And so you start with the basic thing of even self-driving cars. And it's pretty amazing that the thing can drive itself. On the other hand, you know, even the dumbest people we know can drive a car, right? It's actually not that great of an accomplishment in the measurement of human endeavor. When it's all said and done.

​Colin Kessinger:
So, it might be other reasons why you can't drive a car, but it really comes down to intelligence. Right? And so and now if you put that in the context of, of corporate activity, there are a bunch of things that Gen I can take on, starting with the simplest version of chat bots, but working on its way up to dealing with contracts and workflows and workflow automation and things of that nature.

​Colin Kessinger:
So I also think it's important, as you engage in these activities, to find the easy problems to solve, the less controversial problems to solve, like there's whole aspects of every supply chain full of tasks that nobody really wants to do. And we can talk about decision support and optimize decisions. Often those are more contentious, and they may be a little more sensitive to some data aspects.

​Colin Kessinger:
So I think that's a useful lens for evaluating when and how when it starts, but also measuring your your return, appropriately. You know, where I think a solution. And I know this isn't an ad for Lumi, but you know right now. So cursor is a productivity tool for software engineering. So it's a great application. Can I wildly successful deserves all the hype it gets.

​Colin Kessinger:
And yet it's still not good for everything. Right. There's there's certain programing tasks it's great for and and other tasks that it's not. And as you start layering in, it comes back to my earlier point where the technology does much better when it has expertise embedded into it. And so you're going to see targeted AI applications that solve specific tasks.

​Colin Kessinger:
And that's how you move up from the equivalent of being, hey, can you drive my car to the store and back to, hey, could this be a delivery van? It has to make weird stops and do other things. How do you get that? You have expert, expert knowledge on all the special task. The delivery van has to do, in addition to just generic driving.

​Colin Kessinger:
I think you'll see those parallels. I think Lumi is a good example of it. I think you'll see other technologies where that knowledge base and knowledge expertise is there to complement the genie AI. The genie AI doesn't have to divine it all by itself, because it won't get there fast enough, and you'll start seeing increasingly advanced applications of this.

​Colin Kessinger:
But I do think my version of Start Simple is as much around what is the cost of success or failure in the application as it is. You know how many nodes or or any other measure of traditional measures, of of, complexity, like being a decision we can all rally around easily, I think is a great criteria to apply to what am I going to trust this machine to go do for me?

​Colin Kessinger:
And I just do think there's a lot of ready, readily accessible, opportunities to engage it.

Ibrahim Ashqar:
Yeah, I, I love that. Sorry, Louisa. Just, just, just I don't I think, I think, you know, the example of the driving car and versus the delivery van that's making stops and doing all these complicated, things is, is a is a great analogy. I think over here, like, the applications, the Genie applications that had the most widespread, I guess implementations across enterprise are definitely more of those horizontal.

Ibrahim Ashqar:
I, chat bots, which were, I think, harder to quantify benefits. It's truly like a productivity booster and time savings. And I think the more vertical journey I deployments, the more specific ones that you're talking about. Those are coming. Lumi is definitely an example of those. I think that's that's exactly right. I mean, you know, when we when we kind of work with our clients, we try to tie success to like a metric moving metric in the right direction.

Ibrahim Ashqar:
So we're trying to, you know, anchor our success to some sort of bottom line impact. But, yeah, I mean, those deployments are have not hit, you know, think of like, the likes of a fortune ten retailer. One of our clients, while we were delivered some success with 1 or 2 functions, business functions, it's not across all the functions so that you can see a tangible outcome on the bottom line.

Ibrahim Ashqar:
Right. So we're having some success with within the business function. And you know, so that's why you may be seeing that a first half of the year of 2024. You know, according to the slide, versus the second half of 2024, cost reductions have increased. So that makes sense on an initiative level, for vertical applications. But to impact the actual bottom line of the company in the earnings call, it needs to be more widely, deployed or implemented.

Ibrahim Ashqar:
And so, you know, I think it's just a matter of time when we get there, but it's a matter of starting small, proving the value and kind of scaling across.

​Colin Kessinger:
But and sorry, I just wanted to point because you said something that, reminded me of this today, the difference between a generic solution and a vertically, specific solution may be measured. For example, in a percent of the time you get the answer that's going to work for you. Right? And if you think about organizational efficiency and job function efficiency, if a solution is, you know, it's really easy to do a proof of concept in a pilot and say, hey, wow, that's give me a great answer.

​Colin Kessinger:
If that's only happening 70% of the time, the adoption rate is going to, decline precipitous like the overlay of the vertical expertise is what gets you to a better answer 90% of the time, or 92% of time or 93. And I think that's the other way to gauge us, is you have to believe that you're going to get to a point where the answer is high quality enough frequently enough that it is a bonafide win.

​Colin Kessinger:
So, you know, there's a lot of things you can put into ChatGPT today, which is great for a bunch of things, but in fairness, it's probably about accurate somewhere between 40 and 70% of time. That's not going to make it into a a workflow, productivity boost, in most functions. Right. I think that's the other piece that you have to be very mindful of is it looks really attractive, real easily in the POC level.

​Colin Kessinger:
But the devil and triple clicking on some of that stuff to make sure that it's a real win is, I think, essential, given where we are in the stage of development.

Kunal Thakker:
Totally agree.

Louisa Burwood-Taylor:
I'd love to get to some, some more specific, use cases. And as we move on to this next chart here, I know you wanted to talk a bit about this and how this sort of shows that, you know, supply chain, has performed above expectations in the use of Jenny. I'd love to hear your thoughts.

Louisa Burwood-Taylor:
And then and to the others, you can you share some specific use cases on where you, you know, had your expectations? You know, work were great, you know, were better than you thought on the ROI.

Ibrahim Ashqar:
Yeah, for sure, I think I think, Lisa, I really like this slide, as a follow on from the previous one because previous ones like, hey, you know, the Jenny AI paradox, we're not seeing the impact on the on the bottom line just yet on an enterprise wide level. But we are seeing some successes at at a micro, level within the organization when not really micro, it's still substantial enough that we're expecting that we're seeing tangible ROI and, and this, this chart kind of shows that and shows how supply chain is, you know, if you look at it more often than not, it is above ROI expectations when it's implementing AI.

Ibrahim Ashqar:
Jenny AI solutions in supply chain. And as you know personally from Lumi, there's there's a couple of good ones, that come to mind, especially, for clients who we've helped kind of, provide recommendations on what to buy, when to buy and how much. So, to inform procurement decisions or, you know, recommended PPOs and things like that.

Ibrahim Ashqar:
These were for like more mid-sized type of, CPG retail clients as they have a massive supply chain arm, and they're really getting hurt by inventory stock outs. And how they actually. What prompted this whole thing? You know, there's, the fact that they were losing sales on the fact that they weren't able to service their clients, because they didn't have the inventory in time.

Ibrahim Ashqar:
And so using Lumi, we were able to kind of connect to their ERP, teach Lumi about, the intricacies of their supply chain, their business terminology, and, be able to kind of allow their, their, their supply chain planning teams and their procurement teams the ability to get like instant access to critical insights, to inform procurement decisions like, hey, what are the items are currently, you know, running low on stock, and how much should we be purchasing?

Ibrahim Ashqar:
Right. And so Lumi would, would do an analysis to figure out, okay, here are the items that are on, you know, low weeks of supply. You're going to run out fairly soon. And here is the recommended quantity we should purchase. And embedding that into their workflow was, was really helpful. And you know, again drive drove a lot of kind of value for, for, for their team in kind of avoiding lost sales.

Ibrahim Ashqar:
And on the flip side of that, I know inventory optimization is a very interesting topic. It's one that's very near and dear to my heart. We worked with another chocolate manufacturer who was dealing with excess inventory problem. Right? Where there's too much inventory, kind of, you know, their procurements actually resulting in too much inventory on hand.

Ibrahim Ashqar:
And that's tying up cash flow. And so using Lumi as a very easy way to pick out these unproductive areas within your organization of inventory was was a big one, right? Like, hey, what are the items with the most weeks of supply or you know, are there any, you know, this was a chocolate factory. They had this concept of expiring inventory.

Ibrahim Ashqar:
So, you know, are there items where we have expiring inventory soon and how much of it and all of that stuff. So we were able to kind of really hone in on certain use cases that tie to, like, measurable outcomes, in this case, freeing up cash flow through optimizing inventory levels. And it's all through kind of having the, the, the AI agents powering Lumi just to be familiar with their underlying data structures, SAP, Oracle, Microsoft Dynamics, their business terminology and being able to kind of, you know, query the data as needed whenever they want.

Ibrahim Ashqar:
So those are some of the smaller ones on the larger ones that we've seen that have done well, like a fortune ten retailer. It's around kind of giving them boots on the ground. What's happening on the item store day level? Usually when when you're at that scale, like a fortune ten retailer, you're dealing with hundreds of millions, if not billions of rows, and you won't be able to kind of put all of that into a dashboard, into power BI or Tableau.

Ibrahim Ashqar:
And so you're, you're forced to kind of aggregate it to a higher level in order to, to kind of get any sense of what's happening. But when you aggregate things to a higher level, you lose visibility on what's happening on the lower levels. Right? So you're unable to see what's happening at the item store day level, because just too granular and your your dashboards aren't capable of of surfacing insights at that kind of level yet.

Ibrahim Ashqar:
And so to me was fantastic in surfacing these anomalies that are hiding beneath the aggregations, to help them weed out, areas of their operations that are underperforming or, you know, bottlenecks in their operations and kind of action on them. So that's kind of some of the things that we've seen. But I'd love to get, you know, Kunal, your take Colin on on what you're seeing as well.

Ibrahim Ashqar:
I know you work with a variety of clients.

Louisa Burwood-Taylor:
No, maybe I can I just jump in and maybe you could share, you know, example of something of an AI implementation that was above expectations, but also one that was below expectations, you know, as a way to kind of give some of the people here, you know, guidelines on maybe what not to do.

Kunal Thakker:
Of course, you know, I think, well said, Abraham and I think, you know, I'm going to piggyback on that with the real life, solution where it did exceed expectations. It was more around, you know, we've sold in my previous companies on Amazon and, many marketplaces, and having inventory in FBA obviously is always, you know, is a proven, con because fraud, because, yes, you are being able to offer, prime service and you being able to increase sales and being in stock.

Kunal Thakker:
But on the other side, there's also holding cost associated with that, especially in Q4, where it quadruples, on the, on the holding cost. So you've got to have the right balance of inventory sitting in an FBA or a, other marketplace, offered solution. And your three PL to be able to do proper replenishment. And having leverage in the past, obviously, I wish we had, tools like Lumi, but, you know, we've leveraged, data science teams to create models for us and, generative AI to then be able to provide insights on what is my safety stock, what is my run rate, what is my lead time in terms of being

Kunal Thakker:
able to go and replenish from the triple center into an FBX warehouse, or create a purchase order based on what inventory is in the warehouse in the pipeline, so we can send a purchase order to the manufacturer or supplier. So by the time you know, we are running out of inventory in the, you know, point of sales front, we have enough inventory in the pipeline to come and be able to replenish, into these nodes or models.

Kunal Thakker:
And, user generated AI has definitely, in my opinion, exceeded our, in expectations in giving us results. That's allowed us to be in stock, avoid any out of stock or, customer reviews getting impacted by that. And also allowing us to be able to plan it correctly. One other area I would say, which was very beneficial, but it needed a little bit more hands on, which is, you know, addressing to users, where it has not met our expectations was into something I called inventory aware pricing.

Kunal Thakker:
When you are selling online, and you have a certain price based on competitive analysis and business assessments, what happens is some to some inventory, you have it, you know, in lower in stock and you know that your pipeline of bringing inventory from the, you know, sourcing facilities to shipping to getting it to the warehouses may take time sometimes because your sales have exceeded what you had predicted.

Kunal Thakker:
And inventory, their pricing using generative AI would then help say, hey, if you're lower in quantity, let's raise the price so we have better profit margins, and we'll stay in stock till the inventories back in. On the contrary, when we have excess inventory, it say, let me discount the pricing by a little bit, which is then going to allow me to, release my inventory a little bit faster because then that ultimately, make sure that we do not have too much of holding cost, or storage costs associated with that excess inventory.

Kunal Thakker:
So with that being said, I think there is definitely, certain areas where in the supply demand planning in terms of replenishment at or below placement, it's exceeded expectations. Sometimes in inventory of our pricing. It, it is still learning. You know, the, the, the, the behaviors of the customer, to be able to give us very accurate.

Kunal Thakker:
So this is where information, which is where a bit of a manual intervention still comes into play in making those decisions. Or it may not have it exceeded expectations in those areas, but those are my two real life examples. I think that, you know, it's it's allowed us to either exceed expectations or meet Collin's, to you.

​Colin Kessinger:
Yeah. So unfortunately, most of my, real life experiences mirrored yours. So I won't dare do all of you the disservice of repeating, I think the themes it can all laid out, we've now seen play out over and over again, that there is definitely a fairly clear path to games. I do want to go back to something that was said earlier.

​Colin Kessinger:
You know, it's easy to confuse this. This oh, I can just get dashboards. Right. And the things that I would leave you with is I haven't met many organizations that didn't have a backlog of dashboards that needed to be built. Everyone's always waiting on some bi team to go spin something up. I haven't been to many organizations that don't have more dashboards that aren't used than used, so we spent a lot of effort waiting around for things.

​Colin Kessinger:
They all get created and then we forget why they were even created. Along the same lines, I can't remember the last time. I mean, it happens all the time is, Hey, by the time the dashboard was done, I no longer cared about the problem. Right. And so, you know what? We're, you know, the use of AI here is that, yes, AI has a standard set of reports.

​Colin Kessinger:
We all know what they are, and that's great. But there's so much dynamic activity in our supply chains. It's a great example that very often in any given week or any given month, I need to go chase a problem for four weeks, and then it's resolved and I move on. That supplier went down, shipment got lost. Whatever it is.

​Colin Kessinger:
So there's something that I need to go watch closely. And the beautiful thing about AI is that it enables adaptability, right. So there's a productivity aspect to it. Hey, it'll write some code on my behalf so I can stand up a new dashboard much faster. I can get through the iterations much faster. Therefore I'll have the relevant data much faster, hopefully before my, you know, crisis is blown up.

​Colin Kessinger:
And so I do want again, in that lens that I laid out, you know, I've laid out a number of cases that it enabled better decisions for sure. Right. That's the end game. But the speed and adaptability to get there is as important as the quality of the answer that you got. And so I want to make sure that as you look through your own application opportunities, that you put equal value, on that productivity gain and that adaptability and that flexibility, versus just, hey, now have a report that I didn't have before because almost any bi organization would just tell you, hey, tell me what report you want.

​Colin Kessinger:
I'll build it like that. Doesn't feel any different, except that hey bi are going to say you take too long. You have too many requests. The requests aren't always clear. Right? So there's a there's a lot of interference in that process that I in this application. But across the board has a tremendous opportunity to reduce, mitigate or eliminate, by.

Ibrahim Ashqar:
Calling on and on that note, and just in the interest of time, I think the next slide would be quite awesome. You mentioned how do you think about laying out, the framework of use cases in your organization? So this is the, you know, this is also a an excerpt from the, McKinsey, publication that was just this month, I believe, on how they're trying to, map out use cases around horizontal versus vertical, so to speak, across the different business functions, supply chain being one of them.

Ibrahim Ashqar:
But one thing that's interesting to note in that in our article, they're saying that at the heart of that paradox, and it's something we were touching upon throughout this whole conversation, is this imbalance between horizontal, widespread copilots and chat bots and, you know, which have scaled quickly versus kind of these more vertical, type applications that, that, you know, harder to measure still, you know, not not widely spread as some of these horizontal counterparts.

Ibrahim Ashqar:
So, you know, I'm curious to to get your thoughts on one, the, the explanation McKinsey put out here into this framework as a whole. To kind of thinking through, you know, how to think about the use cases within an organization.

Kunal Thakker:
Sure. I think I can, I can, provide, a little bit of my perspective here that, you know, one of the big things for, these initiatives of leveraging generative AI, whether it's within supply chain or across functionally, it is that it has to be cross-functional, right? We can talk about supply, demand planning and other examples of where inventory management and optimization as a supply chain function.

Kunal Thakker:
But think about it. When you look at the functions on the top, procurement plays a big role because we got to source the inventory, the right right quantity at the right time at the right place. Supply chain is the one who's taking the action on it. Marketing is the one, that is, deciding on when the promotions are going to be that's going to dictate the, the, the demand, that if I have a promotion, I better have the right inventory in place.

Kunal Thakker:
Technology plays a big role in terms of helping us implemented. And ultimately, the customer service is going to provide us the information on what was the post-purchase customer experience, having done all that we did upstream. So in a way, there is, one of the areas where I've seen companies fail or not seeing the impact, is when they're not working.

Kunal Thakker:
Cross-functionally we can call that supply chain as a leader, but supply chain in using AI and supply chain. But it it is totally dependent on other cross functions. So that alignment is necessary in my opinion. Second thing is the the data, you know, we need to make sure that the data is clean. The master data file is accurate for us to even come to a proper conclusion.

Kunal Thakker:
And that, in my opinion, again, goes across the board with, working with the tech team, working with the other functions to make sure that we have collected the right data and populated it into our engines to then, you know, say input equals output. So that's, just by a high level, thought process. And, at the end of the day, you know, in looking at these vertical use cases or horizontal use cases, I don't, feel that you're going to see a very big impact unless you work on initiatives that are more cross-functional and have an alignment, versus when you start working in, vertical areas.

Kunal Thakker:
Of course, there are exceptions, no doubt about it, in every business. But, vertically, I think getting insights AI will help a lot. Getting insights faster with large data sets is going to definitely help. But again, when you go to act on it or you do the action, it is all interdependent and cross-functional.

Louisa Burwood-Taylor:
Colin, did you want to add anything there?

​Colin Kessinger:
Yep. Just one little lens on this. What's fascinating about the list at the bottom is, well, first of all, the fact that there's a horizontal and vertical is just a reinforcement of.

​Colin Kessinger:
There are things where that domain knowledge is just less important. And so you're going to get, some of that, but they'll only get you so far, and it won't be as far as what some of the vertical specific solutions are. So take cursor, which is a great post, development tool. It's different than ChatGPT the same underlying technology, but again, very specific.

​Colin Kessinger:
For a certain set of activities, cursor applied to other activities actually doesn't perform well at all. So you know, again that vertical specificity matters. The second thing I would tell you it's interesting is I look through the, ten boxes. Seven of them are unique to gen AI versus old tools. So the creation of content, the brainstorming of ideas, you know, the support assistant, the ticket categories are those are all things that Jenny AI is going to do a lot of things, I think, to introduce new capabilities, prioritizing and qualifying accounts.

​Colin Kessinger:
You know, historically, there's great tools for it. You know, the the non gen AI version of it, the insights provider, you know, good data science. With that scientists could could do many of the great things. Jen I can provide some augmentation. And then of course the demand forecaster is also in that category. So for me it's interesting where they're sort of net new big opportunities.

​Colin Kessinger:
And you'll look in those other the other seven and then again it amounts to automation and productivity at the end of most of them aren't really about insights, you know, summarizing research and trends that that's an automation activity, etc.. So I think it's interesting to reflect on where are you going to see traction first and sort of, watch the progression for me in the forecasting space and the inventory space and even in the sales assistant space, I think the great opportunity is the conversion of unstructured data and into structured data.

​Colin Kessinger:
So our traditional machine learning models love structured data. And there's a lot of things we like about traditional machine learning models, because ultimately we can dissect how they actually work and explain why they did what they did. And for some of these functions, that's really important. But still, there's that opportunity to say, all right, hey, there's a bunch of industry reports, or projections about, you know, microclimates and you know, what the yields and production quantities will be and, and all that kind of stuff.

​Colin Kessinger:
And the question is, how do I take all that unstructured data, reform it into something that I can then build a model against? And I think we tend to think only about the decision itself. But in my opinion, one of the data models with more data always beat models with less data like that's just vertical. You know, if you can see more, it may not use all the data you feed it, but the more opportunities you give it to be successful, the more likely it is, and that quickly becomes more important than the, how exotic your models.

​Colin Kessinger:
And so to me, one of the great things, having been a lifelong, work in this field, is the untapped potential of taking all this data from all these different places, putting it into a structured format and then feeding it into models.

Ibrahim Ashqar:
Yeah. That's interesting. I think we talked about this before thinking of Genii as a way to enrich the data that you need to, in, you know, feed into like a demand forecasting model powered by traditional machine learning algorithms. That's a super interesting, idea. And, you know, this is just one framework, that that McKinsey put out.

Ibrahim Ashqar:
I thought it was interesting. Thought, sort of framework. And, you know, they're using this framework as the reason to explain why we haven't seen material impacts on the bottom line yet. It's because horizontal use cases are more widespread and it's more about productivity boost. And so that's kind of why you're unable to see the impacts of the bottom line, versus the fact that these vertical ones maybe are not in wide scale adoption just yet, but it's just one framework.

Ibrahim Ashqar:
I think OpenAI has come up with their own framework for enterprise and how to think of AI. Use cases are genuine use cases within the organization. Louise, if you want to go to the next slide. Yeah. So so they thought about it in the form of like six primitives is what, what they said for for they call it AI.

Ibrahim Ashqar:
But this is for Genii. What they meant. And this is just for enterprises on how to think of use cases. So when there's content creation, there's research, there's coding, data analysis, ideation strategy and automation. And you know, they're like, okay, well, there's a whole paper we can kind of dive deeper into, but they're thinking of these are the main use cases of of AI across any function that you think of.

Ibrahim Ashqar:
Right. So within within within any within supply chain where you're spending lots of time coding, for example, and obviously extracting insights from, ERP or a master of masses and obviously area that's ripe for disruption for AI and that's, you know, an area that Lumi kind of thrives in. But that's kind of the realm of what what they're trying to do or, you know, content creation, creating SOPs, for employees that follow some sort of guidebook or handbooks and things like that.

Ibrahim Ashqar:
The research part, I think, is really cool. We're seeing a bunch of tools, you know, especially in like, the procurement contract management space. Kunal, I think you've seen a bunch of these tools, where they actually pull information from third party, summarize it, and give you, like, a vendor risk score and all of that stuff. So this is some of the use cases.

Ibrahim Ashqar:
Sorry. Can I'll let you jump in I you're mute.

Kunal Thakker:
No, I think that's right on. Ebrahim and I think, you know, first things first is, always the dicha model, de for data AI for information you gain from it, K for the knowledge and then for the action. So, yeah, as much as as much as, you know, we are, pulling the data and doing the research at the end of the day, we got to act on it.

Kunal Thakker:
And I think, some of the, applications you rightly mentioned the contract. It's allowed us to not only monitor vendor performance is not only allowed us to identify cost saving opportunities, is also allowed us to understand how do we build capacity, right. Because there are certain contracts that we are seeing that, hey, we are maxing out on those.

Kunal Thakker:
There are certain that vendors are not performing. And it's sometimes becoming a hindrance in terms of growth and scaling the business. So you act on it, you get to, you've leveraged it, and you've done the data analysis, the ideation and the strategy happens. And then, you go act on it by taking the action and then automatically, getting new vendors, or doing better vendor performance management, right, to mitigate any risks that might be coming with it.

Kunal Thakker:
So I think that's one area. The other area is also cost optimization through. You know, I mentioned about transportation management system earlier is, hey, I've got a ton of these, shipping carrier partners that I have rates from and they're plugged into my DMs. But what are the variables? Right. What am I defining as a variable?

Kunal Thakker:
As the variable that select a partner for me who can get me from point A to point B in a certain time in transit? Who has the service level that I'm looking for? Who has the right volume discounts for me? Is that carrier already maxed out on the rebate or the discounts I can get? So maybe move the volume to the another carrier so I can start gaining over there.

Kunal Thakker:
Oh, and one more variable. How about customer experience, right. Is the rate best? But is the customer experience with that carrier the best? So you got to balance that profitability with the post-purchase customer experience. A lot of that decision making that we've started to use generative AI in to play versus having to have done those decisions manually before.

Kunal Thakker:
Right? So I think these are just, some amazing, areas where we could, you know, tap into each one of these at some point with this data analysis or strategy or automation, to help us make sound decisions for the customer.

Louisa Burwood-Taylor:
Great. We are running out of time and we've already got some amazing questions. So, I think we can jump to those. And so first one is with the growing interest in implementing generative AI and supply chain operations. How are companies addressing concerns around data privacy and sensitive business information? And have you observed any resistance to adoption due to these concerns?

Louisa Burwood-Taylor:
And how are organizations navigating the balance between innovation and data protection? I don't know who would like to.

Ibrahim Ashqar:
I think I'd like to I'd like to answer that. We've we've done our fair share of security clearances and privacy policies and things like that. At this point. It's a it's a it's definitely top of mind. You know, I think companies are very sensitive, with their data and, obviously at the start, maybe in early 2023, there was a lot of kind of we don't even want to use gen AI to begin with because they may be using their our data to train their models and things like that.

Ibrahim Ashqar:
I do think I do think we're we're a bit better now, but I think you still got to adhere to all the enterprise standards and protocols they expect about data not leaving their environment, data being handled in a proper way. Most of these corporations have, security clearances. Intensive like 300 question questionnaire is a times that you got to kind of honestly answer.

Ibrahim Ashqar:
There's no way around it. You got to you got to have your SoC two compliances. You got to make sure you're handling data in the right way. Things are encrypted in the right way. So all of those are things that you have to do if you want to cater to enterprise. And more recently, I've seen that a lot of these companies that have AI committees, so that this concept of, there's the IT security clearance, which is just going over the basics, you know, making sure their data is, is, is secured and not subject to potential leaks or hacks, but also then a separate AI council that is just considering the, the, more

Ibrahim Ashqar:
tactical uses of AI's ability to influence or spread misinformation or, or things like that. And if you want to sell into an enterprise, you got to be able to kind of walk through both of those conversations and, and have solid responses to kind of, each, I suppose. And, you know, in that process, you'll discover features that you need to build and things that you need to account for in order to cater to.

Ibrahim Ashqar:
Like the most stringent of organizations. We've done a fair share of them at Lumi. Happy to say that from the start, we built it in a way that ensures that data is remains within the client's environment. So addressing the concerns of, hey, I don't want to share my data or anything like that with, with, with AI or with you guys.

Ibrahim Ashqar:
So that was really good from the onset. But, you know, I don't think there's that you have to be very mindful about it. I don't think that's going to go any, any anytime soon. So you have to innovate while, you know, the question is like balance between innovation and data protection, I think the innovation has to happen in light of the data protection framework as well.

Ibrahim Ashqar:
So you have to kind of have that it's constrained to begin with.

Louisa Burwood-Taylor:
Great question. What is the best way to get started using AI. I think you know this might be a good one for you. Is this something that we should focus on. Okay.

Kunal Thakker:
So I think you know, it's a good question and, something really basic that we all need to, ask when we get started. So I, I've mentioned this earlier, and I'll, I'll repeat, is that, you know, AI projects don't scale unless we have a good cross-functional ownership. So getting an alignment from cross-functional partners is very critical.

Kunal Thakker:
Having a good master, a data quality, it will lead to more reliable predictions and solutions. And ultimately, you know, tech for tech's sake, with unclear ROI is not a good idea. So having a good mindset in terms of I want an ROI, whether it is improving my time in transit or improving cost or improving capacity or quality, or getting more accurate reporting, those are the, the ultimate outcomes that we need to have in mind before we start creating a solution around it using AI.

Kunal Thakker:
So those are some basic fundamental, you know, initiatives that I'd recommend, for starting with the AI.

Louisa Burwood-Taylor:
That's great. I really running out of time. I think we have a poll, which maybe Anthony from New Me can put up. We'd love to, see what you guys think on that question there. Where do you see the biggest opportunity to apply AI in the supply chain? And as you guys, answering that, another question we have is what is the biggest obstacle for adoption?

Louisa Burwood-Taylor:
I don't know, Colin. You have or.

​Colin Kessinger:
Yeah, I will sort of tie back. I'm going to wrap up the two prior answers into this one. But I do think that security, for whatever it's worth, will be a solvable problem. Like, we think it's too important to solve. And every vendor in this space has to come up with a credible solution. IT organizations sort of as a group, will come up with acceptable standards.

​Colin Kessinger:
So I think it'll be great progress there. I think. But on the flip side, it's almost nice that it's painful in that it'll keep you from jumping in on the wrong things, like, it's so easy, like every function your organization is getting hit by some, genie vendor who wants to sell you something, and if there isn't enough pain associated with it, you're going to do exactly the opposite of what Kunal said.

​Colin Kessinger:
It says, I don't know about you. All right? Doesn't matter. Looks kind of fun and cool. And, you know, your proof of concept is great. And, you know, you've wasted way too much time, on this stuff that the only other thing that I layer onto it is I honestly would start with a problem that you understand very well.

​Colin Kessinger:
It may be a little counterintuitive. This is the problem you don't understand is where you might bring in a consultant. Then you're going to get all the insight and say, wow, there's a lot of value. Assessing the performance of these technologies is a non-trivial task, and it goes back to asking the right questions, knowing what you're getting out of it, being absolutely sure that it's creating value, and not having to fight a change management battle, I think, makes great criteria for getting started.

​Colin Kessinger:
The threat is lower. Your ability to assess and learning how to assess these technologies is much better, because it's an area that you understand, you know, there's there's immediate, you know, workflow and automation and therefore productivity gains associated with a bunch of these things. So I, I would get your feet wet like it's like I said, like on that on the lead generation and the inventory optimization forecasting.

​Colin Kessinger:
If you haven't been an active participant in evaluating this technology for the last five years, this is a hard place to start and it's easy to get sort of, bamboozled by all the fancy demos and all kind of stuff. And that's why I keep coming back to that ROI problem. I really understand it won't be debatable.

​Colin Kessinger:
I don't have to fight the organization on it. I think that's a great way to get started in these, both financially but also just organizationally great.

Louisa Burwood-Taylor:
And then with like 30s, what is, long term possibilities? What is the North Star opportunity this technology could unlock for the future? I think given all celebra.

Ibrahim Ashqar:
Something something yeah, I have this is a great way to end, I think I think that the promise of this technology generally isn't what isn't what one singular large language model can do. It's what a combination of multiple other or large language models can do when they put together, and they able to communicate and talk to one another and solve a more complicated task.

Ibrahim Ashqar:
And so that's the difference between like single agent and multi-agent applications. And really the promise of these multi-agent or a genetic workflows. There's multiple terms terminology going out. There is their ability to not only auto extract insights for you, but the ability to take actions as well. And that's where I think things become really interesting. And that's like the promise of of this technology as a whole is that you now have this incredible technology that's I can dig up insights for you.

Ibrahim Ashqar:
And then with a human in the loop kind of a button, take an action for you. For example, issue a DC to start to store stock transfer or, you know, mark a vendor as a cat do not, do not purchase from this vendor or whatever, but be able to take these actions and streamline a lot of manual processes.

Ibrahim Ashqar:
That's like what I think the true promise of, of this technology is. I was headed in and, you know, in our world with Lumi, we're trying to build that a genetic workflows for kind of supply chain. So an agent that can extract insights from supply chain data sets surfaced, to the appropriate stakeholders, provide recommendations. And then with a click of a button, someone be able to action on those recommendations as well.

Ibrahim Ashqar:
That in my opinion is the future.

Louisa Burwood-Taylor:
Amazing.

​Colin Kessinger:
And at the end of the time, I'm gonna add one one last week because it's so short, like the aspiration for this technology is that we should all enjoy our jobs more obviously, be that, yes, we just enjoy our jobs more if we do this right. Yeah, I love that we.

Kunal Thakker:
Can make this more successful period.

Louisa Burwood-Taylor:
Solutely yeah, a great place to stop. Well, thank you so much everyone for attending. We will be sending out the recording to, to all of you so you can follow up. We are also going to be co-hosting another webinar in about a month's time. Do follow Lumi on LinkedIn or org funded to get more information about that.

Louisa Burwood-Taylor:
And yeah, thank you so much for joining us. Thank you to our speakers. And Colin.

Kunal Thakker:
Thank you.

Ibrahim Ashqar:
Thanks.

Louisa Burwood-Taylor:
Thanks everyone.