Champions
of Data + AI

データドリブンな革新を推進するリーダー

EPISODE 18

AI in Retail

A never-ending fixation on enhancing customer experiences is true for most industries but even more prevalent in retail — where consumer preferences for style and fit are critically important to get right.

In this episode, Vimal talks about how AI is used across four different retail brands and the impact of a hub-and-spoke model when tackling common use cases for the various lines of business. Plus, we touch on the need to innovate quickly in order to stay competitive and how that’s balanced against the need for operational excellence.

headshot
Vimal Kohli
VP, Data Science and Analytics, Gap Inc.
Vimal Kohli is the head of Data and Analytics for Gap Inc. In this role, he leads a team of data scientists, analysts and consumer researchers to provide insights into the Gap Inc. brands and help inform strategic decision-making, long-range planning, and transformation. Prior to joining Gap Inc. in 2020, he led data science organizations at Dick’s Sporting Goods, L Brands, and Wyndham Worldwide. He has his MBA from XLRI Jamshedpur and his Bachelor’s of Science in Mechanical Engineering from AMU in India.

Read Interview

Chris D’Agostino:
Welcome to the Champions of Data and AI. I’m your host, Chris D’Agostino. A never ending fixation on enhancing customer experiences is true for most industries, but even more prevalent in retail, where consumer preferences for style and fit are critically important to get right. In this episode, I’m joined by Vimal Kolhi, VP of data science and analytics at Gap. Vimal and I talk about how AI is used across four different retail brands and the impact of a hub and spoke model when tackling common use cases for the various lines of business. Plus we touch on the need to innovate quickly in order to stay competitive and how that innovation is balanced against the needs for operational excellence. Let’s get started. All right, Vimal, thanks for joining me today on Champions of Data and AI.

Vimal Kolhi:
Good morning. Thank you for having me, Chris.

Chris D’Agostino:
Let’s talk about Gap and more importantly, let’s talk about the challenge that you face as the VP of data science and analytics, because it’s not just Gap, it is multiple brands within kind of the Gap umbrella.

Vimal Kolhi:
Right.

Chris D’Agostino:
So, Banana Republic, Old Navy, Athleta. Can you tell us and the audience kind of what have been the big challenges with coalescing a data strategy across four very distinct brands with likely different demographics for your customers.

Vimal Kolhi:
Yeah, yeah, no great question. Great question, Chris, you it’s funny. So I’ve been with the business a little under two years, and as you know, my team supports all four of our iconic brands. What’s interesting is everywhere I’ve been in this industry in the last 20 years, one of the first questions people will ask me and we’ll get into a debate will be, should you have a centralized or a decentralized model, right? And I’ve found both of them have their challenges. And that challenge gets even more magnified when we are in a structure here, as you correctly said, where we are supporting not one workflow brand, right. What we’ve tried to do is to find the best of both worlds by creating a hub and spoke model.

Vimal Kolhi:
So, there’s two kinds of members of my team. One portion of the team is embedded in each of the brands. Those are as much a members of our central team as they are members of the brands themselves. Pre-COVID when we were still in the office, they were co-located where their brands were. They weren’t in a central location. And that part and parcel of whatever functions within the brand that they’re working most closely with marketing, for example.

Vimal Kolhi:
On the other hand, in the hub portion of the team we have people that are focused on specific domains and solving specific problems, right? So if you have someone who’s focused on inventory optimization, for example, or yield management or pricing or forecasting, well, then that’s what they’re doing, right? They’re honing their craft to build a forecasting model that can operate at high scale, high velocity with a large degree of automation that can be part of a best in class inventory management solution that we as a company can build and deploy it all four of our brands with customization.

Chris D’Agostino:
So, Vimal, that’s interesting. So I love the idea and we talk to a lot of customers about embedding the data science teams within the lines of business because they understand the business objectives more clearly, they understand the data sets that are collected, but you bring up a good point. You’ve got a bunch of different retailers that from my vantage point, and correct me if I’m wrong, seem to have a very similar business model.

Vimal Kolhi:
Yep.

Chris D’Agostino:
And so, how do you optimize for common use cases that span those four businesses?

Vimal Kolhi:
Yeah, that’s exactly right.

Chris D’Agostino:
Do you centralize a team around inventory optimization and is it applicable to each brand in an equal way? Or do you have to tweak it?

Vimal Kolhi:
Yeah, no, that’s right. So, well we have tweaked it. So what we’ve typically done is, it is the consumer insights professionals and the data analytics professionals that we have embedded inside of all the brands and inside of all the functions. The data scientists who are building really the machine learning applications need to work very closely with the technology organization. Those are the ones that are not building solutions on a brand by brand by brand basis, because that would be extremely duplicative and wouldn’t give us scale, right?

Vimal Kolhi:
So when it comes to forecasting or pricing, or each of the four brands will have their distinct pricing strategies, but the capabilities they need in any apparel retail business that we own or run will be reasonably similar. So it makes sense to build that scale by building that capability once and then deploying it to all the four brands. Now, the rollout would still be brand by brand. We may build the capability once, but we will deploy it one by one. They will be customization level look different, the outputs and the reports and the monitoring of the machine learning algorithms could and sometimes will look different.

Chris D’Agostino:
So talk to me a little bit about how the consumer demographic and their use of technology changes brand by brand. Is it very similar in terms of the shopping experience and the digital experience, or do you see a differentiation between say a Gap purchaser or consumer from say a Banana Republic?

Vimal Kolhi:
Yeah. So it is and it isn’t. So I’ll give you a specific example. So one of the things we’ve done is we’ve created a very sophisticated and proprietary segmentation of the entire US apparel customer. Right? And we’ve done it in a couple of very unique ways. One is unlike what some people will do, we’ve not done it just for our current customers or buyers. Right? So to that extent, again, we get the best of both words because when you do it for the entire US apparel market, obviously we have four brands that have a very distinct positioning, which kind of play a very different role in the consumer’s life, right? These are very complimentary brands. But the coverage is not only of all of those four brands, but also of customers that haven’t yet shopped either of four brands. That’s the first thing.

Vimal Kolhi:
The second thing we’ve done uniquely is a lot of times this segmentation is either based on surveys that can be very rich in the kind of information, but very limited in the extent of coverage, in terms of the number of customers you can score in that segmentation. Or people will do the reverse where they will build it only off of internal data, in which case the richness is lost. What we’ve done is we’ve done both, right? So the segments are created using very rich survey data without any predisposition to what the segments themselves should be. It’s a combination of some of the things you were asking, demographic psychographics, behavioral attributes, just how consumer shops apparel.

Vimal Kolhi:
Then we’ve led on a predictive model that then has the ability to score everyone into those segments. And then what’s happened is, working very collaboratively with each of the brands, they’ve gone really deep into each of the segments and then looked at what their brand stands for and what their strategic plan is. And based on that, and that’s not static or set in stone, but based on that now we’ve created a mechanism or a framework for each of the four brands to pick the segments that they think are their target segments and come up with strategies.

Chris D’Agostino:
Yeah. So, online retail has obviously taken off over the last 10 years, and it’s challenged the business model of traditional brick and mortar businesses. And you of course have a hybrid model where you can purchase online, but you can also purchase in person at a store. Clothing is really personal for most people and the fit, the look, all the things that they care about from a visual and aesthetic standpoint. Let’s talk a little bit about how that B2C kind of business model and AI and the need for people to understand what the clothing’s going to look like. How have you used AI or what has Gap been doing sort of more broadly as a strategy for not only embracing people’s desire to see what the clothing looks like in person, but also trying to minimize cost and scale the business.

Vimal Kolhi:
Yeah. Oh, great question. You said it right, right? When it comes to apparel, fit is one of the most difficult things to get right because how each individual customer see themselves fitting in a piece of clothing is very unique, right? So I think it is a combination of both of those things that you mentioned. On the one hand, I think we are fortunate. We have the advantage that we have a model where across all of our four brands, we have a very large network of retail stores where customers can walk in and experience our products. Right? On the other hand, we make it easy and convenient for our customers to shop, right? Whether it’s shipped from store or buy online pickup in store or just have the product shipped.

Vimal Kolhi:
Over and above that, what we’re trying to do is again, solve this problem in a very unique way from an AI standpoint. This is a problem that has not been solved well by anyone all the folks who have attempted it so far, whether they use size charts or size guides or this, that, or the other, none of them come close to replicating the real fit experience, right? Gap, Inc. really put its money where its mouth was. And a few months back, we announced the acquisition of a startup named Draper that has a very, very unique technology where they actually use AI and fundamental body scanning techniques to provide a fit experience to a customer in a very, very unique way.

Chris D’Agostino:
Sounds great. Yeah. The last time you and I chatted, we talked a little bit about Draper, and then you also talked about the need to operationalize AI and some of the challenges where a company that’s been around as long as the Gap has been, and there’s a lot of pressure to streamline, be very efficient, be very cost effective with, especially in the retail space given all the competition that’s out there. Help me understand your point of view because we chatted a little bit about it last time about maybe too much emphasis going into operational excellence.

Vimal Kolhi:
Yeah, yeah, yeah.

Chris D’Agostino:
And that for or slow down the progress. And then maybe layer in, do you think you could have created a Draper like capability within the current construct of Gap or is it something where you had to be acquisitive and go and acquire that sort of speed of innovation?

Vimal Kolhi:
Yeah. I’ll take the second first, that’s really straightforward and easy. Again, we need to stay focused on what the larger goal is. Right? The larger goal is to give our customers a great shopping experience and to drive shareholder value for the company. Right? So I personally, as the leader of analytics, I don’t subscribe to this kind of it has to be built in my backyard kind of mindset, right? So we will do whatever is right for the business.

Vimal Kolhi:
I think there’s two challenges from my standpoint that I see in AI. One is that companies struggle to scale. They struggle to scale their data science in AI and ML investments. Most companies today have done a lot of interesting POCs, but then they don’t know where to go. I think the second thing that happens is to your point, when you focus too much on operational excellence, you run the risk of getting disrupted because then the focus only becomes on incrementally improving the processes you have today. Whereas in many instances, the business would probably be better off just replacing the process completely and re-imagining it. Right?

Vimal Kolhi:
So, the simplest example I give is automation. I usually don’t want to go in assuming that an existing business process is perfect and I just want to automate that. I want to take a step back and question what outcomes that process is driving. And I like to start with what the north star ought to be and work backwards from there. I think that’s a much better way to go. And that then, to answer your question about innovation, that’s where I think if you, operational excellence is tremendously important, don’t get me wrong. But I was being provocative by saying, I feel like a lot of companies are scared to break with a past. You have too much of focus only on operational excellence it can lead to incrementalism. It can take you away from 10 X thinking and you don’t take any big swings, you don’t go for the home runs.

Chris D’Agostino:
So it’s interesting because you have kind of just built in scale, whether from day one now, meaning anything that you do given the number of customers that you have that are loyal to your various brands, given the number of retail locations, given the number of garments that you produce, right? The range of products that you produce. It’s like everywhere you look there’s scale in your business in every dimension. So anything that you’re doing automatically gets tested at scale if you roll it out, right?

Vimal Kolhi:
Yes, very much.

Chris D’Agostino:
It sounds to me like it’s a bit of that balance of, “Hey, we can spend all this time trying to perfect it, but we’re going to get eaten by another competitor if we do that.”

Vimal Kolhi:
Right. Yeah, exactly.

Chris D’Agostino:
That scale to actually test things.

Vimal Kolhi:
Right.

Chris D’Agostino:
And forgive me for being so blunt, but it’s not like this is life or death, right? So, if something goes wrong, certainly there’s business impact, but maybe you shipped the wrong supplies to the wrong manufacturing plant and the wrong product potentially gets made or something. I don’t know. These aren’t drugs that are saving people’s lives and-

Vimal Kolhi:
Yeah, no, exactly.

Chris D’Agostino:
It seems like you’ve got to balance how innovative you’re trying to be is how disciplined you need to be.

Vimal Kolhi:
Yeah. Very much, very much. And I think that you said it right. The balance is exactly what it is. And I’ll give you another simple example. I think just close a home to the work we do for your audience. I think here’s the example I would use. Right? So what the scale does is, the scale does two things. One, it gives us, as an analytics professional, as a data science professional, it gives us really, really interesting problems to work on, right? The fact that we have a long history and are still in the process of transforming ourselves also gives the opportunity for members of our team to work on really interesting problems that we may not yet have solved in this company. Right? That makes a big difference. So you’re not coming in for example, to tweak an algorithm and nip at the edges of it. You’re building fundamental stuff. Right?

Vimal Kolhi:
And then thirdly, as you said, very rightly what happens is we have processes in place. We have a lot of data science products that are already in production. Those are running at scale and already productionalized. Now, that also gives us the opportunity to experiment with a lot of new stuff without disrupting the core operations. And then it allows us to follow a CICD mindset, where again, I go back to my simple example of forecasting. We have a production forecast in place, right? And we are free to experiment with new forecasting algorithms every single day. And the day we come up with a forecast that beats the current one in true champion challenge and more, we can switch them and we can decide how much risk we want to take and what the upside and downside is. And we can roll out faster versus slower. That’s the advantage of having the scale. Because it gives us to some extent, the luxury of being able to work on tomorrow’s problems while still supporting today’s business.

Chris D’Agostino:
Let’s talk about a wishlist since it’s the season for wishes and all of that. If you’re in your data and AI role, tell me, the importance of a data architecture and just all the things that make your role successful in an organization as important as Gap. Tell me what three things you have on your wish list this season.

Vimal Kolhi:
Yeah. Thank you. Yeah. Great. It is the season of wishes. You have to promise me that my three wishes will be granted. No, I’m kidding.

Chris D’Agostino:
They will.

Vimal Kolhi:
I think the first one to me is I want to move from data science as a service to data science as a product, right? So, microservices architecture has been around for a long time, but oftentimes the data science components or the ML components exist inside the microservice. And what I want to do is productize that, so the data science becomes a microservice onto itself that then interacts with the other technology microservices.

Vimal Kolhi:
The second wishlist I have is I still feel like far too much time gets spent in prepping and provisioning the data versus actually developing the model. So, again, that’s a journey we are on. I want us to keep going on that journey. And my vision the future is my dream state is the first version of an ML model we should be able to, and the performance of that first version may be completely lousy from an acceptance standpoint. I don’t care. But the speed of it, the data should already be pre-provisioned to an extent that the first version of the model can be built in days ideally, at most in a week or 10 days, that’s it. And then we can do CICD from there and keep improving.

Vimal Kolhi:
And I think my third wishlist is around talent and I feel like sometimes being in San Francisco, being in Silicon Valley, obviously, we have access to a lot of great talent, but we are also competing with a lot of big names for talent. And I think there’s a lot of things we do very uniquely. So what’s on my wishlist is, all your listeners, all the talented people that we are trying to hire in my team, that’s my wishlist that I could reach out to all of them, because if I could, I would say the following things.

Vimal Kolhi:
One is we have a great culture in this company. I’ve been in this space 20 years and candidly in my role, in roughly similar roles, as head of data science and analytics, a very large chunk of my role historically has been convincing the executive leadership of the importance of analytics, work of data science, work to make the transition to AI. And I have not had to do that at all here. We are blessed at Gap, Inc. to have an executive leadership team that truly believes in the power of AI. And I always say, put your money where your mouth is. Well, this company did it. We acquired Draper, we acquired CB4. They continue to invest in this space so the belief is there.

Vimal Kolhi:
The third thing I would say is, and I touched on this a little bit before, sometimes you can join a large organization and become just a small cog in the wheel and you can be doing some fairly operational work, even though it’s all AI and ML work. And we don’t have that situation. We have really, really interesting fundamental problems that still haven’t been solved here. So when a new member joins our team, they get to work on really cool stuff and it’s exciting work.

Chris D’Agostino:
All right, well, those are three tall orders and I promised you in advance that we were going to grant them. So let me tell you how we will satisfy that. So the first is you want data science as a product instead of a service. So we are really, as a company, we’re moving towards a low code, no code approach. Obviously the platform is designed for expert people with ML backgrounds and data science backgrounds, but it increasingly is also very easily addressable by people with skills in the non-programming space. And so we’re working on being able to create more citizen data scientists, if you will. So that’s one thing.

Chris D’Agostino:
The second thing you wished for was less data wrangling and the ability to be able to work with data more quickly and spend more time on the model development and less time on all the data prep. And so that’s been a big focus of ours as well. So I think, we can satisfy your two wishes with one platform and that’s Databricks in the Lake House Architecture. That’s the only pitch I’ll make.

Chris D’Agostino:
And then the third thing in terms of talent, this podcast, I just won an Emmy. I don’t know if you heard about that. We have 37 million listeners.

Vimal Kolhi:
Awesome.

Chris D’Agostino:
On a weekly basis. So you now have probably the broadest audience you’re going to get.

Vimal Kolhi:
Awesome.

Chris D’Agostino:
I’m pretty sure those numbers aren’t true. I’m going to run with them and just say that they’re true. In closing, what kind of advice would you give to people? You’ve had a great career, you’re doing some amazing things at Gap. What kind of advice would you give to people that are pursuing a career in this space?

Vimal Kolhi:
Yeah, I would give one piece of advice. I would say, learn how to learn. I mean, again, I was fortunate to get into this space almost 20 years ago when the field of analytics was very, very new. So I grew up with the industry. I consider myself very fortunate. And I have always felt that if I couldn’t learn on the job, I would not have survived in this space. Because whatever you learn in school, by the time you come out of school, the technology has changed, the environment’s changed, something has changed. So, I think that is the advice I would give people, is learn how to learn, retain that curiosity. Be a self-starter. Don’t be shy to ask others questions. Don’t be shy to ask for help. Read a lot, learn from your peers, learn from the people around you and just keep learning every day.