Virtual Meetup co-hosted by Women in AI, Robin.ly & TalentSeer
The COVID-19 pandemic has dramatically changed everyday lives. Our eCommerce and retail experience will never be the same as companies build new forms of connecting with consumers, delivering products, and managing eCommerce supply chains.
This virtual meetup co-hosted by Women in AI, Robin.ly, and TalentSeer on April 30, 2020, gave valuable insights on the impact of AI in the eCommerce & retail industries as well as career advice.
Our featured guest speakers include:
Bindu Reddy: CEO & Co-Founder @ RealityEngines.AI; Former General Manager, AI Vertical Services @ Amazon Web Services
Yuqian Dong: Senior Research Lead @ Wormpex AI Research
View full event recording below. View summary of event highlights by Margaret Laffan here.
Subscribe to Robin.ly newsletter to stay updated for more inspiring AI leader talks:
CEO & Co-Founder @ RealityEngines.AI; Former General Manager, AI Vertical Services @ Amazon Web Services
Bindu Reddy is the CEO and Co-Founder of RealityEngines.AI. Before starting RealityEngines.AI, she was the General Manager for AI Verticals at AWS, AI. Her organization created and launched Amazon Personalize and Amazon Forecast, the first of their kind AI services that enable organizations to create custom deep-learning models easily. Prior to that, she was the CEO and co-founder of Post Intelligence, a deep-learning company that created services for social media influencers that was acquired by Uber. Bindu was previously at Google, where she was the Head of Product for Google Apps, including Docs, Spreadsheets, Slides, Sites and Blogger. Bindu has a Masters Degree from Dartmouth and B. Tech degree from Indian Institute of Technology, Mumbai.
Senior Research Lead @ Wormpex AI Research
Yuqian Dong is currently a Senior Research Lead at Wormpex AI Research, leading its Operations Research team. The focus of her team is to use Operations Research and Machine Learning techniques to drive retail operation decisions. Before joining Wormpex, she was a senior manager on Staples’ Supply Chain Team. There, her team was responsible for helping Staples’ customer delivery division to reduce operation cost, improve service level, enhance network infrastructure, and optimize carrier management strategies.
Full Interview Transcripts
*Has been lightly edited from the original transcript*
Margaret Laffan: In today's discussion, we are going to be talking about AI trends and applications in e-commerce. When we think about this topic and why it's so important, and why AI in retail is such a hot space right now, certainly from an AI research and application perspective, but also from the business, let's look at some numbers.
These are some pretty significant numbers to express what's happening within the market. If we look at a recent survey done by Tractica, AI and e-commerce is booming at such a pace that the revenue growth is expected to reach $36.8 billion worldwide by 2025. The annual global spending on AI by retailers is anticipated to reach $7.3 billion by 2022, two years from now. Capgemini did a survey of 400+ retail executives, they found that AI can help retailers save over $340 billion annually by the year 2022, by enabling efficiencies in several processes and operations.
We know that AI is used in many different ways in retail and e-commerce, and that's why today we are delighted to be joined by Bindu Reddy in this conversation, who is the CEO and Co-founder of RealityEngines.AI, and Yuqian Dong from Wormpex AI Research. We're looking forward to a great discussion with you both, you are heavily engaged at the forefront of AI research and application deployments in your respective companies. And we look forward to learning more about this.
As we get started, to the audience here, be sure to submit your questions in the chat and we can try to get to them throughout the discussion.
Give me a moment to get through the speakers’ bios. I’m going to get started with Bindu. I want to shed more light into your experience and your great expertise. Bindu, you are CEO and Co-Founder of RealityEngines.AI. Before starting your current company, you were a General Manager for AI Vertical Services at AWS. Your organization there created and launched Amazon Personalize and Amazon Forecast, which were the first of their kind AI services that enable organizations to create custom deep learning models easily. Prior to that, you were CEO and Co-Founder of Post Intelligence, a deep learning company that created services for social media influencers that was acquired by Uber. And prior to this, you were previously at Google, where you're Head of Product for Google Apps. So quite a rich level of experience there in various different companies that I know a lot of people would love to have on their resumes as well. And you have a Master's Degree from Dartmouth College and a B Tech Degree from IIT, Mumbai. We're delighted that you joined us, thank you.
That’s a great intro. Hello, everyone.
And I’ll share the intro of Yuqian Dong as well. Yuqian is currently a Senior Research Lead at Wormpex AI research leading its Operations Research team. Wormpex AI Research is the research brand of Wormpex. It's a fast growing cashier-free convenience store chain, with over 1000 stores in China backed by global capital. And the focus on your team is operations research and machine learning techniques to drive retail operation decisions. And before joining Wormpex, you are a Senior Manager of Staples supply chain team, where your team was responsible for helping Staples customer delivery division to reduce operation costs and improve service levels and enhance network infrastructure and optimize carrier management strategies. And you have a PhD Degree from Georgia Tech. Welcome to today's webinar.
Thank you for the introduction. And thank you for having me.
Margaret Laffan: I wanted to spend a few more moments on the introductions because these are the topics that we're going to talk about in AI trends and applications, and I want to make sure that everybody understands the type of background that you bring to the table for this discussion as well as your expertise and experience in this area.
So Bindu, let's ask you the first question. Now it's been a couple of months for you and for the business, what have you observed during COVID-19 that has impacted your business or what you're doing, are you doing anything differently today as a result of COVID-19?
Thanks for the question. Just to give people a little bit of context, RealityEngines.AI was started early March 2019. We are a team of around 22 people right now, so we're a startup in every sense of the word. We launched our beta on January 28th. So we were ready to get going with scaling our beta service. We are an AI cloud service, first of its kind as well. We basically offer AI as a service to various different organizations in e-commerce as well as in retail.
Obviously, in a lot of ways, this whole COVID pandemic was unexpected, at least for startups. And I think it's changed pretty much the course of every startup in some significant way. For us, in the first a few weeks, surprisingly, we saw a huge uptick in beta users. I think staying at home and working from home got a lot of people thinking about how they can learn AI, how they can use that extra time to train deep learning models. So we started seeing a lot of people trying to experiment with the service, seeing a lot of people coming to the surface. However, the interesting thing is in the last two or three weeks, that uptick was coming down. And I almost saw a malaise setting in, people were basically not knowing when the lockdown is going to come out or ease up or what the effects are on their businesses. So we saw our beta signups went up this way, and now they're kind of here. So we have to see exactly what the impact is going to be. But I feel like generally speaking, there is a little bit more interest in AI now, and the revolution which was happening is probably actually going to turbocharge a moment more.
The second thing is in hiring. We were aggressively hiring, to be honest. I’m getting research scientists, as well as ML engineers. And over the last few weeks, we have slowed down just because we have a lot of money in the bank, so it's not like we need to completely stop, but it's an uncertain time. And I feel like everybody's doing that, which is going to basically slow down at least new hires for some time now. So those are the two big effects I'm seeing.
Margaret Laffan: So an increase in people researching and leveraging your product. And then also, on the hiring side, you have a bit of a slowdown right now seeing where the market is going to go as well.
Yuqian, the same question for you. How’s it been at Wormpex AI Research? How have the last few months been for you in terms of the research work that you're doing?
It's a very good question. I'm not the CEO of this company, so I cannot speak on behalf of my employer. What you will hear from me will just be from my own observation. From what I can see, our business remains really well here during this pandemic period. We even opened our first store in the southern city (in China) where we have a lot of big pandemic scenarios going on. And we were also able to execute the first touchless take-out food delivery if I don't remember wrong. So I think this time period actually created a unique opportunity for our company. People start to realize, it's very important to have a convenience store around you and you can go there for essential supplies. So from what I can observe, this whole period was creating a unique opportunity for us to grow and to remain strong.
Margaret Laffan: So you're seeing as a result of the pandemic, for example, that more people are familiar where your store is because you have a convenience store that's opened all the time. It’s cashier-less, even though I know when we were discussing this in the prep, you did say there is a human in the loop. You do have people who are there who can help out and run security etc, but more or less it is a cashier-less convenient store. So when folks are concerned around interacting with others or touching surfaces, they can do this more easily with what you have in your stores.
Yes. Because of our technology, we don't rely very heavily on human labor. And that makes the whole operation much easier during this pandemic period. So people don't feel scared to come in because there's a lot of people in the store. And they can still provide quality service to our customers, to give them what they need for their everyday life with a very short distance to their home, because you're all quarantined at home during this time period.
Margaret Laffan: Thank you both for that. What we want to do now is, we want to put up our first poll to understand a little bit better who is in our audience today in terms of the sectors that they are in, whether it is retail or e-commerce. And this will run just for a few more seconds.
We actually have a lot of people who are not in retail or e-commerce dialed in tonight. We have 63% who aren't. And then next, we have 31% of participants on this call are in retail and then 6% are in e-commerce. So this is certainly a good opportunity for us here to spend some time helping educate around what is happening within the AI space for retail and e-commerce, especially for folks who are not engaged right now.
2. Current Applications of AI
Margaret Laffan: So Yuqian, given your ability and the state-of-the-art technologies to facilitate new retail logistics from storefronts, warehouses, manufacturing, can you outline specifically what AI applications you're working on?
We're trying to use AI technologies to help us digitize the physical world and the physical operations, so that we can make more smarter decisions and more insightful decisions when we make any buying decisions, operation decisions, and all sorts of decisions.
Because we have this kind of technology, we can operate more efficiently and identify a lot of bottlenecks during our operations to reduce cost, and then at the same time, increase our profit. In summary, we use AI technologies to help us to make more insightful decisions and how decisions evolve to products, how to set up our supply chains, how to set our pricing, how to do assortment and store layout, all the elements involving opening and operating a retail store, they try to digitize all of that and they use digitized data to help us do a lot of optimizations and learnings so that we can do more customized services to our customers that bring our store closer to our customers’ needs.
Margaret Laffan: Are you implementing computer vision applications and NLP within the work that you're doing?
Yes, it's not within my team, but within our branch. We have teams working on computer vision, trying to use computer vision technology to digitize our store environment and all the physical operations.
Margaret Laffan: And Bindu, for you, we know RealityEngines.AI helps create large scale real-time customizable deep learning models. And these systems generate accurate predictions that you can infuse into all aspects of your customer experience or business processes. What type of AI projects are you working on today in retail and e-commerce?
Just to take a little bit of a step back, in retail and e-commerce, the 10,000 pound gorilla, obviously, is Amazon. And if you look at Amazon, the reason why Amazon is Amazon, I'd like to say, is deep learning. Because deep learning enables them in pretty much all aspects of e-commerce and retail to some extent, to become who they are, because it enables getting a product from a particular fulfillment center, knowing which product will get sold and when, so the logistics of storing it, optimizing the inventory, figuring out how long it takes to ship, all of that is an AI model somewhere in the background doing the work for Amazon, to the customer experience of recommending the right product at the right time.
One of the reasons why we believe at least retail and e-commerce are very ripe for an AI disruption is that there are right now at least thousands of retailers who are either still mostly brick-and-mortar companies or do have some digital presence, but they're not completely AI transformed or AI first. And I think right now RealityEngines is working with, I would say at least three large retailers - I'm not going to name them, but one of them is a big flowers company, so you can guess what that could be. One of them is a very large electronic components retailer. And what we're basically helping them do is to transform their business in three different ways.
The first one which really a lot of people care about is forecasting. Retail is hard because you have to have the right inventory at the right time, and the right location, if you don't have that you're not meeting people's demands. And if you have too much of a particular product, you are overusing your fulfillment center space. So doing forecasting right and being able to forecast exactly how much of a product you will sell is actually a very interesting and very hard science. And that's where we come in. Our deep learning based forecasting is new, two to three years old, but it's probably the most accurate forecasting there is today. So if you look at some of the papers, which have been published by Uber or Amazon, to some extent, Google, around deep learning and forecasting, you will see that it does much better than other forecasting methods. And RealityEngines right now has a demand forecasting service, which you can pretty much use right off the service without too much customization to be able to optimize your inventory. So that's part A.
Part B is personalization. I think everybody who has used any kind of retail website or an e-commerce website or has used YouTube or Netflix knows the power of recommendation. 70% of all YouTube videos are watched through a recommendation. 30% of all Amazon purchases are made because of a recommendation. And what we're seeing now is the other retailers, the non tech heavy retailers now really jump on that bandwagon to basically customize every aspect of their customer experience, whether it's personalized emails, personalized website, personalized app, and so on. So that you can actually really get to people at the right time and give them the right product. So that's the second big aspect of this.
And the third aspect of this is the last fulfillment aspect, which is basically making sure your fraud and your logistics happen correctly. So this is where you want to make sure that you have zero fraud because again, retail is a very low margin business. And that should be ideally a deep learning model. And then you want to make sure that your product is actually delivered at the right time. And that's again, a model which can predict how long it takes to ship things from a particular fulfillment center, where you should ship this from.
So in all of these three parts of the business, you can actually infuse AI and get your business to be at least 15 to 50% more efficient, both in terms of revenue as well as cost.
Margaret Laffan: And I think with both your answers there, you're very focused on the operational side. Yuqian, your team are working on that as well. So I'm assuming that everything that Bindu has talked about is part of your work here as well, in terms of operational efficiency.
3. Trends and Challenges of AI in eCommerce & Retail
Margaret Laffan: Great. Let's do a reality check on the state of AI applications today, because there is a lot of noise in the system. And then there are some very solid use cases coming through. So from a high level perspective, where do you see AI research, adoption and deployment today?
I think AI research is actually in a really, really good place, generally. I don't know how much of the audience actually is into pure research. But if you are, you're going to know already that all of the top conferences have had anywhere between maybe four to five times of an increase over the last couple of years in terms of submitted papers, and there is a lot of research going on in a lot of the key institutions.
And generally speaking, especially when it comes to neural nets, there's a lot of interest and we see a lot of progress happening in leaps and bounds there. Even today for example, OpenAI came out with music makers of sorts, so you're going to see some press releases or the other from the top institutions. Yesterday, there was a release from Facebook around a state-of-the-art chatbot. So I feel like research is actually in a good place and will continue to be in a good place in terms of AI.
Where things lag is adoption. And part of that reason is if you do a lot of surveys with organizations, and you say, how much if your organization is infused with AI? Or how many processes and machine learning are embedded in them today? You’re going to get answers like 10%, 15%, 20%. Very few organizations are going to come up with an 80% answer unless you're Google or someone. And the big reason for that has been that, AI adoption has been very difficult. Doing AI research is one thing, and it's pretty hard, but actually putting a model in production, maintaining it, retraining it, and keeping it alive is really difficult and really expensive today.
We think we’ll catch up, in the next couple of years, you're going to see AI adoption skyrocket. But part of the big problem there has been talent. It still requires esoteric talent to deploy and adopt AI in a big way. And unless and until that talent gap is somewhat closed, or there are enough tools for the existing talent to use and empower themselves, we're still going to see that issue with adoption. Again, although in the next two to three years, I think that problem will go away.
Now the last one, deployment, I would say it’s somewhat attached to adoption. But having said that, there're lots of organizations that actually are excited about AI. There's so much hype, there’s no doubt about that, and they want to adopt AI. But when it comes to actually deploying it, they have a lot of hurdles. The first one of course, is just plumping hurdles, like how do I get things into production? What are the monitoring tools I need to have? How do I make sure my models are giving me the right predictions?
The second big hurdle is getting ROI. Proving ROI with machine learning is a non-trivial exercise. And I see a lot of people can just tumble from that leap that you have to do from having a model which works in the notebook, but not having a model which works in the wild. And that will also get better, as we get better at things like online learning, as we get better at more robust methods. Supervised deep learning, or supervised learning in general isn't very robust right now, because it sometimes tends to overfit. And when it sees new data, it doesn't really work. So I think all of them in some ways are related. As research becomes better, as we get more and more robust techniques, you're going to see that adoption curve increase. As we get more and more better tools, you're going to see more and more production. So I would say in the next two to five years, especially in retail, you're going to have almost every single retailer and e-commerce company be completely AI first.
Margaret Laffan: Yuqian, before I ask you this same question on your thoughts on the reality check for AI today, Bindu, what about the cost? You talked a bit about the ROI. But what are your thoughts in terms of the buyer opportunity? Are you still seeing companies doing use cases and the trials after that? Where are you seeing this come out in terms of the ROI?
That's a really good question. In a lot of companies, especially retail and e-commerce, they tend to be low margin, as I've mentioned before, so they don't want to spend too much money upfront. And the problem with AI is, especially if you want to have a big AI team, you have to spend a lot of money up front. And so I've seen people, at least retailers either go with the services approach or with the tools approach, so they either get an AI consulting company, or they talk to someone like us, who's an AI tools vendor. And yes, they do want to try before they buy, so almost every one of them will do a proof of concept usually by a paid pilot, and only when they see the ROI, at least in the offline metrics, will they then go put that model into production and see how it works.
The other issue is, even if the model looks good, like I said before, once it goes into production, it might not look that great. So there is, again, a cost in putting that model in production. And we're seeing more and more people now willing to take that cost and get to that place.
Margaret Laffan: Yuqian, do you share some of those same views as well around AI research, the deployment and adoption? What are your thoughts on a reality check on the state of AI today?
For a lot of things that Bindu talked about, I see the same in the industry. So for research, we do see a boom in this industry. We see lots of papers and a lot of people start learning AI and are trying to develop their career in this industry. We do see computer vision, active learning, transfer learning, those kinds of techniques being mentioned a lot in those papers, in publications and in the media. So the research part is definitely booming. We see a lot of good papers and research results coming out.
In terms of adoption, I do see that currently there is a trend for traditional business to adopt more AI solutions. For example, in our business, Wormpex AI Research, traditionally a convenience store is just like a store near your community, your neighborhood, no one will think about convenience stores as an AI-backed company. And now our company has started this trend and our goal is to make sure our decisions are digitized and backed by all those AI techniques.
And we do see the adoption. But just from my experience, I do see it as a slow trend for more well-established companies, because they have a lot of existing operations, existing methods. So the adoption of more state-of-the-art methods requires a lot more preparation than a new company. For the adoption part, I think for big companies, it is a very slow and sometimes painful process.
In terms of real application, we need a more smooth transition from research to real applications. In research, the papers talk about very ideal situations, they talk about very good, clean data. But in applications, we seldom see those situations, our data is tinted and sometimes we don't have enough data. Those produce the first layer of difficulties in the real world application.
And the second one is infrastructure. Sometimes in academic papers, all the problems are really small scale, but when we get this into real applications, we need a much larger and scalable solution. So the hardware infrastructure is definitely a very hard requirement for some AI adoption and real time applications.
Another part is the ROI part, the money part. For AI research from the initial exploration to the final deployment, that requires a lot of investment sometimes, and some companies may just want to solve the problem at hand and they don't want to produce a long-term solution. So those big investments may be a big hurdle for those companies. And they don't want to go ahead to adopt more AI-based solutions.
And finally, I would say the deployment also has real implementation. So let’s say, I have an algorithm, an AI-based solution, I want to implement it in the real world, but when I make it in the real application and operation, I may see a lot of difficulties. I want this AI application to be operated in this way, but in reality, it may be operated in 1000 different other ways, different from what you thought it could be, and then it may give you the false impression that this method doesn't work. This is really not the methodology’s problem, nor the algorithm’s problem, but how this algorithm has been deployed in reality. So I do see that there is a trend going on in academia and in industry, people talk more about AI. But I do see there are some milestones that we need to conquer to make AI more widely applicable in our everyday world and especially in traditional industries.
Margaret Laffan: Let's talk a bit around the trends, when we're thinking about new trends in retail or e-commerce, again, can you both share some of your thoughts around what you see trending right now and what we might expect to see in the next couple of years evolving within either the store or the e-commerce process and so forth?
I see two main trends. About a decade ago, you saw a lot of retailers - some of them, unfortunately, are going bankrupt now - but companies like Target, Gap, Macy's, Walmart, all of them went digital, clearly. And they spent quite a bit of time and money on getting their digital presence up and running and getting their e-commerce sites up and running. And then just recently, I would say even a year or two ago, there hasn't been that much excitement about their digital properties, because a lot of their sales still come through the brick-and-mortar stores. And there wasn't that much super excitement about their digital site.
That has very recently changed dramatically. Part of that is probably because of the pandemic, where people are realizing that, eventually people are going to be very comfortable shopping online, whether it's fashion, whether it’s groceries or other retail goods. So we are definitely seeing a look back to how do I make my digital website much more AI friendly or AI first? They all have some level of machine learning or some level of rudimentary personalization on their sites, but they all want to take it to the next level. So at least we are seeing a lot of interest with retailers or with e-commerce websites saying, we need to do a little bit more.
But I also see a move towards things like more of those what I call “fancy features”, things like subscriptions. So you're seeing, at least the fashion retailers are thinking and wanting to do things which are a little bit more like Stitch Fix, where you have user profiles and you could potentially send out recommendations and even better, boxes, to users who can try things on and then take it back. So we're definitely seeing interest in that.
I am going to go back to saying that there's a lot of interest amongst CPG (Consumer Packaged Goods) companies as well as retailers to get better at forecasting. The COVID situation actually makes it more important that you have to be able to forecast how much of something you need.
Unluckily, AI models aren't great at predicting COVID. Obviously, when we had this pandemic in the beginning, we had this huge need for toilet paper. There's no way an AI model will get that very well. But what AI models can do is, they can say, based on the weather, based on Twitter trends, based on seasonality, this is how much you will need for a particular good. And the more companies can get better at it, the better off they are. So I would say the two big things are personalization and forecasting from an e-commerce site. And the trend is very high on those things. And then there is a significant number of people who are thinking of, what I'm going to call the “fancier AI features”, whether that is the Stitch Fix style stuff, or even fancier stuff, which I'm sure that deals with computer vision, where you're basically trying to model a particular dress or have lipstick on your face, which you can see through your app, I think there's more excitement around that stuff as well.
Margaret Laffan: So you certainly see from the operational perspective, forecasting will continue to mature and get better, because it's not there yet. But from the frontline in retail, the commercial aspect of it, things will get fancier and more attractive, do more things for you when you have a new app - do your makeup, do your styling. I know one of the big issues coming out of Thanksgiving last year and Christmas was the returns. It seems like this whole return angle has not been fixed yet. And there's still a lot of churn and a lot of money being burnt in that space. So it'll be interesting to see how retail addresses that.
Definitely. I think the whole issue generally with returns has also been like, how do we make that more and more streamlined? And to every aspect of this, there are predictions about how much of the inventory gets returned, what would you do with it? And retail hasn't been great at that. But the second aspect of this to what you just said, the more you can have the consumers like and try and decide that they want the good, the better. So in a lot of ways, the Zappos model, or the Stitch Fix model is what I'm going to call a “band aid” model. Because one of the reasons it exists is that people can really try out their clothes online. And the more you can get better at that, the less the problems with logistics and shipping and returns and all of that stuff, because now you can actually virtually see how you look like when you try things out.
Margaret Laffan: Absolutely. And Yuqian, for you at Wormpex, you have a whole lot of problems to solve in some different ways because it's around what the future on the operational side is going to be. But then also, grocery is different from, let's say, you're a retailer like Macy's or Neiman Marcus, it's a very different product, you're offering very different things. Where are you seeing the grocery aspect mature? What are you seeing here in the trends for grocery?
I think one trend that I see here is making the whole shopping experience more customized. So we talk about customized shopping sometimes for some expensive stuff. Now, we bring this concept to convenience stores and we want the customers to come to our stores, feel comfortable and enjoy their shopping trip. We don't want them to feel that it's very deliberate and we set it out for them. We don't want them to feel anything different, so they don't even know what is going on. But all the decisions we make are very customized for the customers, and they just feel like it is the right place for me to go shopping for everyday needs. We want to give them this type of feeling.
I see a trend that all the AI goes to customized solutions, it is not just one solution for all occasions. We want to provide very customized solutions for the locations, for the customers and for the products. We also want to make the AI solutions very fun, not only for our customers, but also for our industry, so that people can be freed up from certain repetitive work, and they can transfer the knowledge to the AI algorithm or AI technologies. They can collect the expertise from the human beings who used to make the decisions and then to help them make better decisions.
That also brings up another trend I see. So traditionally, from my perspective, I used to think AI as something that frees people entirely from the whole work and from the whole decision making process. But as I'm more involved in this, I feel like it is not true. It is more like a human in the loop. We need humans in this loop to provide guidance and we also want to get humans in this loop to provide feedback to this whole AI technology or AI algorithm lifecycle of products. So those are the two trends I see - use AI to make things more customized, and have more humans in the loop of AI products or AI solutions.
And finally, I think AI is also a good way for helping us document our knowledge of everyday operation. I do see that for certain companies, when we talk about operations, it is really just words by words, so you have different generations of workers transferring those knowledge from one generation to another generation. I think AI technology is a good way to help us document and digitize those knowledge and transfer them and have them streamlined from one generation to another generation, from one version to another version.
Margaret Laffan: How do you see that maturing within the operational solutions that you're working on? If you're taking all the knowledge in and digitizing that, how is that changing? How do you do your different operational activity?
Bindu has mentioned about forecasting. I think forecasting is a very big challenge for all those retail companies. All the retail companies want to know what is the right amount to buy, what to buy and when to buy them. This is a very important decision that drives the profit and the gross income for all those retailers. AI technology definitely plays a big role in more accurately producing those forecasts. Because AI companies can forecast those numbers and occasions more accurately, the supply chains can be less disruptive. So you can plan your supply chain much longer before the operation launches. And you can identify those hot products much earlier, so you can buy them before they are running out of stock from the manufacturers. And also, you can plan the downstream supply chain better to arrange the delivery to your stores and arrange the delivery to the customers. So all those AI technologies can help you shape the operation and make it more streamlined and more cost effective.
Margaret Laffan: So we can see the whole end-to-end supply chain. And as you talked about customizing, I’m assuming that if you're going to the grocery store, there’ll be a lot more indications or alerts around what it is that you're purchasing every week, and where that is, if there's going to be an issue around it. Maybe the weather is going to get hotter, so you're going to have some items in more demand. So you're ahead of all of these different types of forecasting and predictions as well. But it ends up in the basket in a different way for the customers too.
In the interest of time as well, I want to get into the last section that we want to cover with both of you today. And that's career advice for professionals who want to enter the industry.
Before we jump into some of the questions, we do have a second poll that we want to run. We want to get a sense again from our audience participation, where people are in their professional career, and some of our questions, then we want to tailor it to the audience that we have online with us today.
This is interesting. With both what you said, Bindu and Yuqian, both of you are saying that you're seeing the same type of trends moving forward in retail, so we're going to continue to see personalization and forecasting. You mentioned it was customization there, Yuqian, and I think that's pretty similar to the personalization aspect, but very different from a grocery to a retailer, but we're going to see these areas continue to mature.
So when we look at our audience here, 40% are in the early stage of their career, 1 to 5 years. And within 5 to 10 years, we have 27% of our audience. And then 10+ years, we have 33% of our audience. And we don't have any students online. The two questions we have that I believe will suit our audience pretty well.
4. Career Advice
Bindu, you have a Master’s Degree from Dartmouth and a B tech degree from IIT. And you've spent time at some stellar organizations, especially who are leading in the whole retail and e-commerce space, and you sold one company to Uber and now you're CEO and Co-Founder of your current company.
We are looking for two to three specific areas that you would recommend to those who want to enter the industry or are very early on in their career, what type of advice would you have for them in terms of how you can progress and how to approach or take a strategic approach to your career?
Let's first focus on the people who have 1 to 5 years experience. And assuming that, there are two classes there, there's one class where you already have an AI degree, whether it's in CS with a lot of AI, kind of a specialization. So you're almost there. And then there's a second class where you actually may have a CS degree or may have some other degree, but you really want to move into AI or data science or machine learning. We’ll talk about the second class first because it's somewhat more interesting, because the first class is kind of obvious.
So my advice would be, do the following three things if you want to move into a data science or an AI position. The first and foremost is, especially if you want to do something hands on because it's always a good thing if you're earlier in your career to do at least a couple of years of hands on work. As an example - it doesn't get mentioned in my bio - when I first started, I was a computational biologist for a couple of years, so I actually did modeling. And then for a couple of years after that, I was a software engineer. So my big advice to people who are out of school is, especially if you're interested in it, definitely go and do something very hands on.
And the first thing I would do for that is, to do a couple of online courses. I particularly like the course that Google offers, which is free of charge. It’s a 15-hour bootcamp style machine learning course, very easy to take. It’s something which everybody should get started with, if they're interested in machine learning. You want to go and start there. And then you can actually add a couple more courses from Coursera, or any of these other things that you might like. Andrew Ng’s course is very famous, and that's something which you could try. That comes second.
Beyond that, I would also like to recommend going and using YouTube a lot. YouTube has a lot of resources, a lot of AI influencers, a lot of the companies, especially DeepMind and Stanford, put out their courses on YouTube. So if you're interested in doing that, I would do that as well. That's the education process part of it.
The second part of it is - this is probably the most important thing, and nothing really beats this - is somehow trying to get yourself involved in some data science or AI or machine learning projects. You can do that within your existing company, or you can do that when you're switching jobs, or you can even do that with a hackathon, maybe. But you want to do something which is hands on and practical. If you can't do any of that, you can at least start going through some AI workshops. Doing courses is one thing, implementing things, doing a hackathon or producing a model is a completely different thing. I would definitely recommend doing that second.
The third thing is, once you know of all the concepts, once you understand things like what's transfer learning or what’s supervised learning, or whatever, all of the stuff that you really need to know, I would keep track of what's going on in this field. That's very important because the field changes dramatically year upon year. So if you did all of these three things, it's actually quite easy to get into the field and also to do well, even if you are early on in your career, and not having that much AI expertise.
If you already have the AI expertise, things tend to be obviously much easier here, my recommendation would always be to pick a company, which is much more of what I'm going to call a “pure" AI company over an “applied" AI company. And by that, what I mean is if you have that choice. But then what I mean is, there are lots of companies today, which talk about applying AI to a particular use case. That's interesting. And there's even more examples where people are like, let's take Uber. Uber is using AI for figuring out the ETA on rides. That's a really good ML project. It's one example. It's taking machine learning and applying it to that particular example in that particular company. While that's somewhat interesting, I think it's far more interesting if you work in a company, which is maybe taking AI and applying it to the whole domain, or to a particular class of projects, that's even better. I don't mean this to be a pitch to our company, but even better for a company like ours, where we are basically building new AI techniques for a whole array of use cases. The reason for this is, this way, you are going to solve a multitude of problems as opposed to be fixating on one particular problem. So that's what I would say: pick a job or a career which is more towards the techniques.
If you're further on in your career, and you're looking for more management jobs, again, I think nothing beats understanding basic concepts and learning about ML. So I still would recommend the 15-hour ML course, which is really good for even beginners. And then from there on, try to use Twitter a lot. I find Twitter to be a very, very useful tool, especially if you're going to follow a bunch of AI influencers, very easy to look up. And a lot of people tweet out a lot of useful insights about AI, I would do that.
And then again, I would start thinking about how you could incorporate AI in your current job. Usually, for people who have more years of experience, it's a little more difficult to go pivot in or go to a new job. But it's much, much easier where you are in your position to find something which has machine learning associated with it, which is good for the company, which is good for you. And in a lot of ways, you're going to be the change maker. I would strongly recommend that more and more managers and individual contributors and organizations do that.
Margaret Laffan: That’s great. There's a lot of different sections there that folks can follow. And we can help to summarize that as well. Certainly the online courses, hackathons, keeping track of what's happening in the field of research, and following the AI influencers on Twitter or thought leaders and so forth as well, in terms of what they're doing or how they're addressing the topics. I'm definitely hearing from you, it's a bit harder, perhaps in some ways to pivot in your career right into deep AI. But find ways within your company which you can start to channel yourself, be on the business side where it is that you can add value from an AI project perspective, and do different things that your companies are doing. These are all great career advice sections.
And Yuqian, for those in the industry and who want to progress or be promoted, what are the two to three different areas that you would recommend that they focus on?
The first thing that I think is very important, is you have to be really passionate about this. Don't just enter this field because it's hot. You want to enter this field because you're really passionate about it. So it's very important for your future career and also for your future success.
The second thing that I think is very important is, to be really good at data processing, data manipulation. A lot of time, I do see candidates that are really good at understanding the model, they can do a lot of cool equation derivation, but they're not really good at data. That can be a really big blocker for them in order to build very meaningful machine learning or AI algorithms for products. Because data is the core for all of those technologies or algorithms, you need data to train those models, and you need to build meaningful solutions or meaningful insights based on data. So you have to be really good at manipulating your data, make sure you get the right input into your model, and have a very streamlined pipeline, so that your model is not handicapped by those data inputs.
And the second thing is the foundation. I do see people sometimes talk about different types of machine learning models. I feel like sometimes, it is important to go back to the basic understanding about what are those models built upon. And this is really important for you to go further in this career. Because knowing the model’s name is important, and knowing how the model works is important, but it's also very important to understand what is the foundation of this model, what cost function it uses, what is the mechanism of this model, so that you can apply this model in the right occasions, you don't just simply try the parameters. You may draw a very wrong insight just by doing so without a very good, solid foundation of understanding the model.
The third thing, Bindu has already mentioned, is hands on. When you read a book, or listen to those online courses, you learn a lot about machine learning models, but it’s very different from building the model yourself from scratch. And a lot of times, you will find out it's not as simple as people described in the YouTube videos or online courses. You really have to do it yourself, understand what it is about, what are the difficulties, and how to make it really work for your problems or applications.
And the fourth thing I think is being creative. In the classes or books, a model sometimes is taught with an example, telling you this model is used to solve this type of problem, and it's well applicable in this scenario. But when you go to your own application or real life work, you will see a lot of problems. It's not exactly the same as a textbook or not exactly the same as what is described in those YouTube videos. You really have to be creative. So sometimes you may use one type of solution and build a recommendation tool to solve a totally different problem. So being creative and being able to fluently use different solutions for different scenarios is very important for you to have a very successful AI career.
And then finally, if you're not in academia, you're in industry, this is important for you to understand the business - what are the implementations and what are the needs for AI applications. And when you apply those AI solutions to business, you need to understand business to drive and to draw meaningful solutions and to make good decisions upon those application algorithms.
Margaret Laffan: That’s a lot of great career advice from Yuqian around making sure that you're very passionate about the data, about the field. And around understanding the data, knowing the foundations, understanding how models have been built, their names, and so forth. And then of course, spending hands-on time and being creative and so forth. These are really great advice, especially the business side as well, knowing how this is going to add value into the business.
Thank you both for a great discussion!