#185 Revolutionizing Supply Chains through AI with Ankur Verma

Transcript generated by Podium.page
Help us spread the word by tweeting about us at @podiumdotpage and including us in your shownotes! https://podium.page

NOTE: There were 2 speakers identified in this transcript. Podium recommends using "Find and Replace" to change the speaker label to the appropriate name. Speaker separation errors can arise when multiple speakers speak simultaneously.

0:00:02 - Mehmet
Hello again to a new episode of the CTO show with Mehmet. Today I'm very pleased to have with me joining me from India, Ankur Vermav Ankur, thank you very much for being on the show with me today. Can you tell us a little bit more about yourself and about your journey, and then we will discuss more about TrueGradient as well. 

0:00:22 - Ankur
Absolutely. Hi, mehmet. Pleasure is mine. Thank you very much for having me in the show. Yes, definitely. I would like to talk about who am I and what am I building over here. 

So this is Ankur, as you mentioned, and I come from a hardcore machine learning background. I have 10 years of experience after my graduation into machine learning I work with. By education, I'm an engineer. I did my engineering back in 2013, post that I started doing analytics and my data science journey and machine learning started really at Walmart Labs, where I was working in merchant technology team, taking care of initially pricing, like churning out the price elasticity solutions at a scale of 300,000 active items. Moving on to the replenishment area, wherein I was responsible for designing the neural architecture for determining the solutions of if an item store combination is an out of stock, excess inventory and determining the result inventory, henceforth to be able to get the purchase order right. So that's what? That was my Walmart experience. 

And then my latest experience was at Amazon India, wherein I was part of the Amazon pay here, wherein we used to design the neural architectures. Again, here I was the neural architect, design identifying what is the right offers that we should be giving to the customers. So, let's say, uber in India, or like different types of online merchants in India, they collaborate with Amazon to get out with offers. So, at a scale of 65 million Indian customers, we are trying to turn out what's the right offer given the budget. So that's what my Amazon experience was Over the period of time. I also worked with some of the software as a service company, as a founding member, like at Samyarai. I was a founding member over there, wherein I was the architect like the neuroscience architect of the demand for casting engine right, wherein I worked with 12 enterprise companies over there to be able to establish what is the right demand, that assessment, to assess the demand at the grand 11 level, like the most grand 11, usually it probably happens to be the item store. So, yeah, that's my sense. 

0:02:42 - Mehmet
Yeah, so really that really very impressive, I would say. Now, with this experience, across these old, various giant, take giant that you mentioned anchor, like from Amazon, Walmart labs, as well as you know the field of machine learning and deep learning. That's been so what inspired you to start through gradient and how the previous experience helped you in shaping this startup. 

0:03:12 - Ankur
Right? So that's a great question, Mehmet, and I come from an optimization background. Like in my past experience and always, I've always been a big fan of automation. Frankly speaking, I always like to automate things and to be able to get the manual work lesser as much as I can in my past 10 years of experience. 

So, while I started my analytics journey back in 2012, right and data science journey, ml was not a very common norm at that point of time, like machine learning was just coming up, very like just the academicians had started publishing papers and lesser videos was existing, and there was another trend called competitions. A lot of data science competitions was also was also up during 2012-13. Kegel is a very good platform, which a lot of today more than 2 million data scientists sits over there and does competitions. So it started doing competitions on these platforms back in 2013 and 14. And that helped me to understand some of the articles that are some of the solutions that are being designed by the great folks sitting at MIT, stanford, building solutions at larger scale, right. So then I was very clear that and hence doing these like I would have done 60 plus ML challenges, frankly speaking, in my eight years of journey, right prior to Amazon. That motivated me to come up with a very strong automated machine learning architecture, which is called auto machine learning kind of thing. 

Yes, during the COVID time started working over that and I saw that wherever I was going in my experience, the main thing which I was doing actually repeatedly, or any project that I was working I was trying to design I was trying to design that machine learning architecture in an automated fashion which can be used as an example, something which I designed in 2016 at Walmart is still being used, which is a very thing to think of, right. It has just been enhanced, has been enhanced, but the base still remains. So that was the motivation behind coming and that was the learning that was there to be able to go deeper into the solutions like, for example, my experience and my company folks experience, like in supply chain area, right. So we thought that why don't we go ahead and build something in the supply chain area, automate the machine learning part in the supply chain area as much as we can, and hence I really garnered this particular skill set. 

0:05:38 - Mehmet
Yeah, that's great to hear. So if you can give the audience a simple explanation of what True Gradient does and how does it stand out, I would say, in the competitive AI landscape, correct? 

0:05:53 - Ankur
yeah, absolutely so. Truegradientai is, I would say it's a deep tech technology platform. From the backend perspective, yes, and what type of problems it is solving. It's AI technology that we are building over here. So, if we look at today in the area of retail business, cpg business, fmcg, d2c brands, like all of these types of organizations, one of the most important thing in this competitive world has become are you able to assess your demand and change in consumer behavior right, well, and be able to fulfill the demand at the right point of time or not? So that has become a very important thing and in order to do so right, all of these organizations needs to be on the top of having their inventory levels right, be available for what they're trying to sell, right, and, at the same point, of being cognizant that they're not over stocking items right, like, as an example, if you consider food and beverages grocery department right as an example, kind of thing, then here, yes, availability is an important factor, but there is a self life constraint for these kinds of business, so you cannot be, you know, over stock every time, because that leads to a lot of loss in terms of items, because the items are going to be expired. So in order to strike a balance right at the inventory level, what is the right inventory optimization is a complex problem to solve right in the supply chain area, especially because to be able to determine the inventory for the consumer right at that level, you need to solve very small problems at a different node level, right like from the manufacturing to the distribution, to the, to the, the tertiary to the consumer, and find retail into the consumer. So that's what we are. 

We identified as a larger problem right from inventory perspective and we started. We started building solutions from machine learning standpoint and optimization standpoint right, taking all the constraints from business and solving the, the volume, for the right inventory levels. That was the first first part of our protrusion phenomena. Secondly, we also realized that when we talk to a lot of retailers, what becomes important is out of multiple product mix which is available, like a lot of assortment can be possible or products can be available. Are you having the right assortment mix while the customers are visiting you or not? 

So that's called assortment planning or planogram. So our company also solves for the entire end. To end what products at what time at what shen is is something which we solve for. So this is the second part of the problem that we are solving in terms of assortment, and the third part is the McDonald optimization. So a promotion is very crucial in this competitive world. Right is are you as a, as an online seller, as a, as a ecommerce business, are you having the right type of incentivization for the customer or not? 

right, so in order to perform that promotional planning, you need to strike a balance between your earned money versus the consumer right, so what consumer is paying for so the optimal pricing that should be set up is the third pain point, which we are going to introduce that into our model. 

0:09:08 - Mehmet
Yeah, yeah, that's. That's great to hear. Actually. Now, one thing you know when, when I was preparing and we have checked also, like last week I get to know that your company uses no code AI for supply chain solutions. Correct. Now how do you ensure that these complex systems are accessible and usable for businesses without special technical knowledge? 

0:09:38 - Ankur
Absolutely. And that's where, mehmet, the deep technology which I mentioned on the first place before going into business, comes into perspective. Over the period of three to four years and in the experience of last 10 years, having worked with like multiple firms like Walmart, amazon, is really the. 

SAS, work with multiple enterprise companies. What, what? What we have done over here is to be able to build systems at the back end AI systems at the back end right, which are far more complicated right, introducing certain machine learning techniques, introducing certain deep learning techniques and building it in automated fashion. So that is the layer that we have solved over a larger period of time, like bringing in different types of data into it and trying to solve it. Now, the key question was for us can we go ahead and build a platform on, over and above this complicated system, which can be a very, very simple platform, right, just taking few inputs and be able to solve the problem? So what we have done, we have built a platform over this auto machine learning architecture right, that we have built, and that platform only seeks inputs which are which are inputs which the demand planners, from our perspective, or supply chain managers from our perspective, or growth managers have been doing day to day, have had a day to day basis to we see. We see just like five to six basic inputs right from the, from the customer, with the data connection that we should have the data. 

There are some prerequisites of the data that we need to. As an example, for demand forecasting, we need the POS or the sales data right as a prerequisite. That should be a part of it. For replenishment, we need sales plus orders data as an example. So just having the data information, the basic data tagging information, along with that, the objective of the problem in a humanly simplified manner or in a layman fashion, right, I want to do a forecasting at an item store. Granularity of my sales for a particular period of time is good enough for my machine to, from the back end, to go ahead, run and produce the results, which is whatever is required. So that's what we have been solving in the past. Have we have been busy solving to simplify this person the past one and a half year? 

0:11:46 - Mehmet
Yeah, that's. That's great anchor to know. Now you mentioned couple of you know problems you tried to solve actually with with your company through gradient and you know personally I read a lot of you know tech articles and I love when I see you know technology going to solve certain problems I would say vertical problems, and here we talk about supply chain. So now you talked about forecasting, you talked about merchandising and replenishment, right, what's your approach to tackling these diverse problems and do you see any communities in the solutions like, are they, you know, tackle the same way? Because now you have also your own experience and you are doing now with your start up. So do you see any communities there in the solution? 

0:12:40 - Ankur
I think that's a great question and that's what I always talk to when I talk to a customer, when I always try to mention that that there is a huge common entity in all of these problems that we're trying to solve, because supply chain at different nodes are being driven by the end end consumer behavior. That is being driven by that on the priority basis. 

So we always see that at the center sits your demand planning. That how good, the forecasting or demand assessment that that has happened over there. So all of these problems that I mentioned about, right, be it demand planning, be it merchandising, be it replenishment or the pricing and promotion solutions, right, a demand planning solution or demand forecasting sits at the center, if you have. So what we typically do is like we run the demand forecasting right for solving any of the problem and then, while solving demand forecasting, we determine different types of outputs. Right, we determine different types of outputs and we run certain simulations and optimizations over the same demand forecasting model, which is a trained model. 

0:13:43 - Mehmet
Right, to be able to estimate. 

0:13:46 - Ankur
Let's say you know like. Let's talk about replenishment as an example. So for solving replenishment, you need to assess the demand for a particular period of time and secondly, you need to know what is the availability for you at this point of time, like what is the stock on hand or the purchase order that you have already done right. So the main thing that happens is like a demand forecasting again over there and then writing certain inequalities and different optimizations to get your inventory. So we have created our product in an interconnected fashion is what I would say right. And in order to solve one problem, right, in order to solve multiple problems, we have abstracted it to the level that you go ahead, feed inputs right and be able to solve all the problems together. 

0:14:32 - Mehmet
Yeah, great, great. Now again, the supply chain was on the spot, especially you know the past few years right, so we saw a lot of challenges that was facing supply chain. It caused, like some global disruptions also as well. Now do you see your company through gradient leveraging AI, being able to help business navigate in the future? Similar challenges? 

0:15:04 - Ankur
Absolutely, because in today's world, the world is changing, if you think about it, people are getting lazy right. In today's world, if you think about it, channels have become a very, very important factor. So in the past, like I was just talking about what my experience right If you go back in 2000 area, right, 2000 time, then brick and motor was like you know, go to go here, I want to go to shopping, be in the store and everything right, that was the norm at that point of time. Today also, I wouldn't say it's not a norm like it's a norm today, also because people enjoy it and it was going outside, but people have become busy. So today, the channels through which the transaction is happening is just not the you know retailers, but it's e-commerce. It's different, it's even modern traders, general traders, like e-commerce, and there are different types of channels which are coming up right Now. 

The approach to solve these problems right are not necessarily the same. 

If you think about it, like you know, in terms of what is the right stock in that you should be giving, or what is what is the right strategy in which we should be, fulfillment should happen for all of these customers that have become slightly different. 

So what we are trying to do through TrueGradient is we are trying to bring all of these aspects, like the futuristic aspects of the way the trends are moving in the supply chain industry. We're trying to bring them together and solve customer problems in that particular fashion as well. So, as an example, with one of our customers, we are seeing a huge change in trend from the customers doing purchases from general general traders to modern traders, especially in one of the area, right, so a lot, of, a lot of change. Due to those changes, a lot of strategy and planning changes also needs to happen. Right, for them, you know, be available more in the modern traders, for example. So what are the ways in which you can run optimization? What are the ways in which you can cut the cut the threads from, from threads from the general traders to model traders, like, as an example, are the key things that we're trying to do? 

0:17:00 - Mehmet
Yeah, that's, that's great. And one thing I want to ask you also. I'm kind of like like with all you know, significant research and you know, you, you, you you told me that you do these, you know, participate in these competitions and and so on. So I'm sure you have a kind of, I would say, your own vision on how AI will be, on machine learning, of course, will be adapted in businesses. So how you are seeing this, you know, coming in the future with, with machine learning, deep learning and all of this, where do you see things heading and how you think through gradient will adapt also to these changes. 

0:17:54 - Ankur
Absolutely no. That's a great question again, because in the recent past you think about it. So the chat GPT framework kind of thing was. It's not that it has come now, it was available long time back as well but it's just that it's accessible. It has become accessible to people in the past six months. I still remember you working over transformer architectures back in 2016-15, this is like seven, eight years back in time. Right Now. 

One of the things that I observed last year like it's in sparse last year, once this chat GPT became a norm for people to go to yes, marketing is one of the things that people have started using significantly, but even a lot, of a lot of companies who wants to, you know, leverage AI, ml techniques and and not sure that is this going to work or not are very much ready to at least adopt and see that. What are these AI techniques? How can I leverage, let's say, chat GPT to forecast? As people are asking me questions that, hey, is there a chat GPT available for forecasting? Is there a gap for inventory optimization? So that's the level wherein people have started thinking about from the business perspective, right, and the reason for that is what what I have assessed over the time and seen certain companies grow in the past 10 years in the, especially in the supply engine area and the FMCG and CPG area, is that the competition, as I mentioned, is a big, big thing. 

Because, let's say, in these kinds of areas, brand making is a very, very significant aspect because when I I am a person who will always go and do which is a mango yogurt, yogurt of a specific brand, it will be very hard for anybody to come and cut my like to, to, to, to take a share of share and cut my mango wallet, mango yogurt trend to some some other brand trend, right. So it's a it's a hard industry to do that Like. Now people are trying to identify what are the ways in which we can do that and especially with the data availability there, I think there is a significant change in the mindset of all these demand strategy builders, right, the planning and the strategy guys, vice president of operations, guys in different industries, to adopt the AI techniques and see and also go to the level of making strategy that what should be the right business strategy for next quarter or the next six months. So that's the trend I have observed from my perspective, right. 

Over here yeah right. 

0:20:10 - Mehmet
Yeah, remember, like I was, you know, when we met last week, I was telling you that now people, when you go to them and you say you have an AI powered solution and another friend also I met last week told me the same the first thing people ask hey, what is the chat box here, right? So people think that it's only chat, while you know there are a lot of things that happens in the background, that measures of the people. Maybe they are aware. But I mean because I think us humans, we like user interfaces and I believe the reason why charge PT from, I would say, commercial perspective, is a success, it's because we like, we are like a, you know, we humans will like to chat, right, so we would like to interact, whether by typing or by speaking or you know. This is why Siri at the first place was a success, for example, because you can interact using voice. 

Now, charge PT. People went back, like I would say, maybe they are not aware. They went back when something like ICQ and MRC was very famous, where you type and you know everyone is saying the person there is real, is a bot, but now they are very happy that actually machine is answering them, which is very interesting. Now I'm good you know you from your company perspective. We talked about supply chain, but who is your ideal customer? I would say who is your ICP? 

0:21:45 - Ankur
Correct? Yeah, currently, yes, absolutely. That's a great question. So, as I mentioned, we are solving problems and like for type of I would say for the harder type of categories right over here. So primarily the FMCG retailers, cpg companies who are every day dealing with, let's say, some in the food and beverage category, as an example, is one of the most important categories that we have considered right over here, where, in the shelf life are low, you need to have a very robust demand for casting architectures right to be inventory planning, optimization solutions to be able to get the solutions right right over here. 

So our ICP happens to be these organizations right and the category and who are in the categories where, in the low shelf, life exists right for you that is. That is the first kind of ICP that we have created and the second one is in the fashion brands, because fashion brands are tough to predict because the fashion changes every time. Right, if you think about the footwear, the jeans, lifestyle wear and all of these things that we change is significantly. And the reason for having these as an ICP is because of the deep technology which is required to solve these problems right. In the traditional way, it's very hard to solve a fashion related problem or a food and beverage related problem because they are not just average solutions but they are far more complicated solutions wherein you need to build neural network architecture robust enough and, like some of the great optimization solutions to solve such problems to be to realize revenue gain. So that's what I would say. 

0:23:22 - Mehmet
That's great. Now I'm asking you these questions. Of course, you and I we know, but one of the things that I like in the show to highlight and your startup. You already have some traction. Who is your ultimate persona to go and talk to when you approach any company? 

0:23:42 - Ankur
Absolutely so. Typically, the strategy and the planning folks right who are building strategies for their companies, the demand, the supply chain planner guys right who are there primarily the chief operating officers as well, ceos who are responsible for, I mean, who are the decision makers with respect to how the operations going to run as well will be the ideal person for us to go and talk to. 

0:24:08 - Mehmet
Yeah, that's great and maybe you understood Ankur by asking these questions. So because I want people to understand and the thing that I liked about what Ankur and his team they have done is that they have done it all right when from a startup perspective. So they spotted their problem. A lot of people they have it and this problem it's worth to solve it because it saves them huge amount of time, money, reduce risk and even it can increase revenue with the demos that you showed me last time. So it's amazing story, I would say from from from your anchor. Now, as we are coming almost then, this is like a little bit not related to machine learning and AI is a little bit more you know to your own experience. So you've worked with the large corporates Amazon, walmart and you know the big names all over there and now you have, you are running a startup. What advice would you give to fellow entrepreneurs or aspiring data scientists maybe, who may be considering a similar path? 

0:25:14 - Ankur
Yeah, absolutely yeah. I mean I would say firstly that today, wherever I am it has been, I will attribute that significantly to my experience at companies like Walmart, amazon, even other companies where I have worked at, but primarily these companies where I have worked. So one of the advice which I would give is is, basically, what becomes critical is we are doing work in day to day basis. From data science or machine learning perspective, we are saying we have seen a huge scope in terms of solving, utilizing our skills that we are learning over here to be able to go ahead and implement it right in different factors and different organizations. So one of the things that becomes crucial is somebody picking up such kinds of technology learnings that they are getting and the business learnings, obviously, that they're getting from these companies, and it's a very, very fair chance in today's world to go ahead and try it out and go ahead and look out for, as you rightly mentioned, if there is a problem space which is existing or an opportunity which is existing and build a product around that. So that is one thing that I would say that the world is looking for such kinds of solutions to come, especially in the AI world, to go on. And, yeah, I mean people are very open to talk about I mean firms are really open to talk about the different types of problems that people are solving as a startup or as an entrepreneur. 

That's the first thing, and the second thing to a lot of data science practitioners is always be deep in what we are trying to do is a very important thing which I have tried to at least abide by in the past 10 years to be able to, rather than the tendency is just to understand, have a superficial level understanding and try to implement it. But that doesn't really work well unless you have gotten into the depth of mathematics and understood the algorithms and understood the problem solving or the problem that you're trying to solve well enough. So go deeper in whichever algorithms you are going or whichever problem spaces you are going, and try to understand it end to end, because any problem and any solution I mean as an entrepreneur I've seen as like a area wherein you can really, if you go deeper into it, you can make a billion dollar business as well if you can find the right customers. So depth is one of the things which I always focus on and I try to follow in my life as well. 

0:27:38 - Mehmet
Yeah, and of course you had to do the validation as well. Ankur right. 

0:27:43 - Ankur
Absolutely yes. You have to do the validation. That is super important. 

0:27:47 - Mehmet
Yeah, so Ankur, where we can find more about your company. What's your website? I will put that in the episode description. 

0:27:56 - Ankur
Absolutely so. Our website is truegradientai, so that's our website and over there we have a demo instance as well. Right, if somebody would like to try out a demand forecasting, inventory forecasting, price optimization module, so we allow that like a subscription, like pre-subscription, like somebody can go ahead validate the business, do some kind of diagnosis over the business, that how is the business running, feeding in the data over there. So, truegradientai, there is a sign up button, sign in button over there. So, and that is our presence on the website, and somebody can reach us, book a demo as well. If you press on book and demo, we get a notification and we automatically send out the demo instance as well. 

0:28:40 - Mehmet
The demo into demo instance. Yeah, that's great, Ankur. Is there any question that you wish I asked you? This is my famous last question. 

0:28:50 - Ankur
I think you've covered a great detail of the questions that I would have wished to talk to over here, right? One of the questions that I think can be a part of this conversation would be around, I guess how does the transformation of a data scientist happen to an entrepreneur? 

0:29:14 - Mehmet
right, Like being a data scientist and moving to yeah, yeah, please, if you want to give like a quick hint. 

0:29:22 - Ankur
Yeah, yeah, yeah right, because when you are a machine learning scientist and especially being published papers in the past, like you, drop the e-tech now. How do you convert yourself from a deep tech person to be able to do a business? Right Is a thing, like that. 

So, yeah, I mean, basically it boils down to how we can. In my experience, what I've seen is it requires a lot of personality change, right, and really takes a year one and a half year to be able to do that. Initially, when we do the, then I used to do the conversations with a lot of customers or, generally talking, I used to be very technical, right. So then how do you understand the consigns of the person and try picking the abstract out that, hey, you need to hit what the person is looking to hear and understand the pain point of the person, empathize with the person rather than talking about all the technology and what you are excited about. So changing personality altogether is a key skill that I am learning. 

0:30:17 - Mehmet
Yeah, there is a saying that goes like once an engineer, always an engineer. Right, I first it also back in the days, although, like I shifted to a completely sales role, but when I was still talking, the customers used to end by asking can you let your account manager send us the quotation? I said, okay, I'm the account. And because I want so much deep technical details to refer with you, anchor, like some people they love it also as well, because it shows that you have both the strong technical background and you have the business acumen which is for every customer, is like the ultimate, I would say, thing, this company that we are now about to sign up with them, like your company, like, yeah, the guys, they are not just selling for the sake of selling. Actually, because they have the technical background, we are confident if they have any support case in the future. So these guys knows what they are doing and, at the same time, from business perspective, they understand how business usually works and they are solving this problem. So, yeah, it was great discussion, anchor. I really enjoyed it. Thank you very much for joining me today and, as anchor mentioned, the website will be available in the episode description. 

Thank you very much anchor for your time today and, as usual, this is the way I end each episode. If you have any feedback or question about this episode in specific order show in general, please reach out to me directly. You can send me an email. I'm very active on you can see now, with our new cover all the social media handles. So feel free to reach out to me and if you are interested to be a guest same how anchor was a guest today also don't be shy, just hit me a message or a DM and I will be ready to discuss. 

We can arrange for the time and the day and all the logistics, I would say, but please reach out because, as you can see, I try to have a diversity in the show, bringing CEOs of startups which is my favorite one, honestly to talk about how they are solving problems or CTOs sometimes, and also we discuss some other aspects of entrepreneurship, which is more related to marketing, sales and all the other stuff. So I hope you enjoyed this episode today. Thank you very much and we'll meet again very soon. Thank you, bye-bye. 

0:32:48 - Ankur
Thank you members, thank you guys, bye-bye. 

Transcribed by https://hello.podium.page/?via=mehmet