Revolutionizing Retail: AI Integration in E-commerce with Deeksha Chugh from Square

0:28Welcome to the AI Chat Podcast. thing I wanted to ask you, really excited to have you on, as I mentioned, is I'm wondering if you could give everyone, you know, a little bit of your, you know, maybe your background and tell us about how you got into tech in the first place, how you got into AI. And like, was this always something you were interested in? Is it something you discovered, you know, in college? Or what was your kind of journey, I guess, to this point? Great question. So to give you a bit of background about my journey to AI is I was born and raised in New Delhi, India, and math was always my passion throughout school.

1:09And I started actually, I wanted to become a doctor when I was in high school. But then, you know, if I ended up studying math as a major in college, and I realized like my passion for math at that time actually grew and I then went to grad school to do mathematics and computer science and I realized like I just want to continue doing something that involved math but then when I graduated in 2008 it was a peak of recession and the jobs were really hard to find. So I started my first job as a software engineer in a very large ID company working on mainframes. And I was coding in Kobo and that was really hard.

2:09So after nine months, I left that job because I wasn't enjoying it. I struggled a bit to find the right career for me. And that's when I knew that I want to work in a space which can utilize my analytical skills that I've learned over the last many, many years. And so I joined this consulting company in India as a data analyst. Okay. And that was 2010, actually. And then I was coding in BBA. If you know what BBA is, Visual Basic for Applications and Excel, it was a lot of fun. I really enjoyed that role and did that for three years. But, you know, I moved to States after that and I realized data science became a real thing at that point of time when DJ Patil actually coined that term.

3:12and because of my background in data analyst which is now called data science I wanted to pursue that journey so I started this master's in data science program at University of San Francisco it was okay it was a one-year program at that point there were not many universities actually that offered that program very couple very few universities. And I wanted to, you know, stay in the Bay Area. So I, you know, joined that. And I think that was kind of the place, junction where I really actually started my journey with machine learning, data science, AI, and learned so much through that program, and then the opportunities that came after.

4:06So I would say like that was the starting point of my ML and AI journey. That's incredible. What a cool journey. And I mean, you changed countries, you changed all sorts of jobs, you got a master's, there's so many exciting things to talk about there. Now you're working at Square. And as I mentioned, you're over a couple different areas. I'm wondering if you can give just like a brief explanation about, you know, what you currently do at Square and maybe the areas that you oversee acquisition, cross-sell retention, and infrastructure? Sure. So for the last seven years, I've been part of Square.

4:43I'm currently leading data teams, which are responsible for developing machine learning and AI solutions for acquisition of prospective and existing customers. And for Square, it's a B2B company. our customers are, the main primary customers are sellers. So if you think of the journey from Square's perspective, it largely consists of three different stages, starting with acquisition, then engagement, and then retention, right? So you start with acquisition. In the acquisition phase, you're focused on sellers who are still not using Square. And those are our prospective customers. So in that area, we use, my team specifically, uses a lot of data-driven techniques, machine learning models to optimize our acquisition strategy so that we use data to identify the right seller at the right time and talk about the right product.

5:58Square has more than 30 plus push party products and, you know, 200 or 1400 plus third party integrations. How do you make sure that you put the right product in front of the right seller? So in that acquisition phase, you want to maximize the ROI of your marketing dollars that you're spending as a company. And data comes in handy in that space. The next area that we focus is on engagement. So now that engagement is about existing sellers, existing customers. How do you increase the engagement with your existing customers? We do that by improving the adoption of our first-party and third-party product through recommendations.

6:53So that recommendation could be about product recommendation, could be about, you know, the feature that we want to recommend to the seller. So in that engagement phase, we are using data that our existing sellers are giving to us to personalize their journey with Square. The third, yeah. Yeah. And then in the third stage, which is now that you have an engaged seller, how do you make sure that they are retained with your company?

7:31So retention is something that, so churn is something that every company faces, right? So in the retention phase, we focus on building some churn models so that we can predict who is about to churn so that we can proactively reach out to them. Oh, interesting. Yeah, we can proactively reach out and then understand what's causing them to churn and, you know, devise some strategies so that we can learn like what our sellers are struggling with so that we can improve as a company. that's so interesting actually so i have a my background is kind of in marketing and something you know that we focus a lot in marketing is like what's called like a win back campaign so like once you lose someone i have to get them back i love that you guys are so proactive and that you have the technology to use machine learning and ai and algorithms and whatnot to look at people that are likely to turn so before it even happens you're proactively trying to get them and i bet your you know your attention is so much better for that so really really cool that you're able to use data in that way.

8:36It sounds like you're working on some very, very exciting problems there. One question I have for you is, you've been in kind of the space for a number of years. How have you seen the, I guess, the market kind of shift and evolve in regards to AI and data and machine learning and everything going on today? Yeah, I think this is something, um it has evolved definitely drastically some of the things are are same um I can share more about that but like for starters data science and machine learning engineer data engineer like those are not the words that you would have heard back in 2008 there was there were no roles like that um there were business analysts there were data analysts and statistical or quantitative analyst.

9:28And then statistical modeling, people used to use SaaS a lot, especially among big banks, you know, for fraud and fraud detection and other strategies. They use SaaS as a software to do statistical models. And maybe, you know, Google definitely had like quantitative analysts, which were doing some algorithm development and so on. When, you know, in 2010, 2012, when these data science world became the sexiest job of the 21st century, you know, Howard University declared that at that time. I remember, you know, every startup, every company wanted wanting to hire data scientists and everyone claimed that their product is powered by machine learning.

10:26even if they're just using rules in the behind the scenes, they're not even using machine learning, but just the fact that they have like one or two data scientists in the company, they said that we're using data science, you know, and that, that was a selling point. And, but then a lot of companies actually became completely data driven. I remember, you know, Squared was one of them that every decision making that used to happen in the company, either to manage or grow the business, they were using machine learning and data science in every facet. And I think the trend has shifted. Now, it was about data science and machine learning in 2013.

11:16Now, it's about general AI. AI. We're in this world with huge democratization of AI and every company is running to incorporate generative AI in their processes, in their flows. And of course, there's a lot of potential with this technology and there's a lot of cool opportunities for improving internal processes, reducing cost, creating personalized customer experiences. And, you know, the landscape has shifted in a way that now we're going to see new wave of roles emerging. Like, for example, prompt engineers are emerging as a result, right? And similarly, I think a lot of existing roles will be redefined.

12:07the machine learning engineer role or the data scientist role or the software engineering role, I think a lot of those roles will be redefined in terms of what companies expect them to do. Okay. Very interesting. Yeah. So a lot of shifts. And that's really interesting to hear your perspective on that. Something I would love to ask you about, right? Because you're kind of in the thick of it and working on all of this kind of new technology. I'd love to know, like, from your perspective, what role does generative AI play in enhancing customer communication and all of the stuff you're working on now?

12:45What is the role of generative AI for you and your team? Yeah, I think if you go back to the basics, right? Customer communication is all about language and content because without the text and the language, you cannot communicate to the end customer. So, and Generative AI is powerful in terms of ideation and iteration and personalization of this language and content that meet the needs of the customer. The most obvious use case that a lot of companies are using or trying to use Generative AI is transforming their customer service. So customer service is where a lot of pain points are, like it's an operational garden, right?

13:44And a lot of customers are unhappy and they're reaching out to customer support or like customer service to get the answers quickly. How can you resolve that? So something else I would love to kind of ask you about and pick your brain on is, you know, what are some challenges you faced with implementing AI machine learning at Square? Great question. So, of course, you know, there are, it's where it's, I think there are different challenges. And you may be surprised with this response or may not be surprised. But the biggest challenge I think so far we face with implementing AI and ML is around people.

14:32it's not just at like at square but i think at other companies that i also work for a lot of people in the people don't really understand how ml or ai works and you know what are some of the obvious opportunities where you can utilize that and so a lot of it comes down to educating people about how machine learning can be utilized to solve a particular challenge um right so it can be sometimes an uphill battle if you don't have the right people in order to get their binds and so on but sometimes you know those people challenges can be around like there's a other spectrum where some people think oh ml and ai is magic and it can do anything and that they think that it can solve all the problems.

15:32And then you try to inject ML and AI in everything. So it doesn't work like that, right? So there has to be, I think, a lot of education around this technology so that people in the company understand where it makes sense to utilize machine learning because it also has its limitations. The limitations is around data and the right business problem. If you don't have any data for the ML to train your models on, you won't get an impactful result, right? So a lot of times, you know, I need to jump in on calls and say like, hey, do you have data in this space? And they're like, no, we just have maybe 100 rows.

16:21And I'm like, no, you can't start with machine learning there. You need to start with something else or a heuristic. So that, I would say, can be sometimes the biggest challenge. There are technology-related challenges. I think those are solvable with so many companies out there who have built infrastructure solutions. solutions. And we have an infrastructure team within our organization that is building tools to get to the solutions faster. You need a lot of machine learning infrastructure, data infrastructure, so that you can maintain the data quality, you can maintain the monitoring of the systems and make sure that what you're feeding into the models is correct and you have ongoing maintainers in the form of retraining your models so that your models always remain fresh.

17:27So there are a lot of technology-related challenges, so you need to invest in infrastructure. So at Square, I think we're lucky that we have huge infrastructure teams that are catered to solving the needs of machine learning engineers so that we can get to a working model pretty quickly. That's really interesting. Yeah, and you guys do have a pretty solid team over there at Square, so I'm always impressed by that. Something else I would love to ask is, you know, like, what advice would you give to companies maybe that are looking to invest in AI for growth, but, you know, they're not really sure where to start with it?

18:06Sure. So if you are, I think you need to understand whether you're, first of all, looking to your data foundations, start with creating solid data foundations. And when I say data foundations, I mean, you have some eventing system, right? And through which you're collecting data about your customers, you want to make sure that you have created some ETLs that creates a curated and high quality data sets before you jump into AI or ML. Because without that curated and high quality data sets out of that high dimensional data that you are collecting from eventing systems, you cannot get to the state of where your algorithm can learn intelligently.

19:06So I would say, you know, that is something that which has stayed constant. Actually, you asked me a question about, you know, how you've seen the field evolve. I think one thing that has stayed constant is very relevant for this question as well is the data quality you know the everything boils down to your high quality data and so investing in those high quality data is is what i would say for companies to to invest in and when they have that ready you know start with heuristics uh where uh don't start with you know just creating a model right like complicated models start with heuristics or rules for example if you are you know trying to maximize a conversion rate you can for a product you can send you can identify the industry where the conversion rate is higher and you know segment your data according to that and then send the communication to your customers based on that rule on heuristics so I would say you need to start with and rules.

20:21And then if that rules proves out to be valuable through your A-B experiments, then you should invest, definitely invest in building machine learning model because building complex rules can be really challenging to manage. So you really want to jump into machine learning at that point of time. Another point that I think is really important before you jump into machine learning or AI is implementing metrics, which basically you want to define the function that you want to optimize, right? If you're not clear on the optimizing function, metric, then you cannot, like, even if you instruct your team to create a model and they don't even know like what they're optimizing for, then you will not get the business result that you're aiming for.

21:17I think that's really, really true. Deekshot, thank you so much for joining us on the AI Chat Podcast today. I've appreciated all of your insights, everything you have shared, and especially your advice to people looking in the field. What I want to do is, for anyone, for all the listeners, I'm going to leave a link in the description of this to Square so you can go check them out and see how their solutions can help play into what you are currently building. But again, thank you so much for Deekshaw for coming on the show and sharing with us today. To the listeners, thank you so much for tuning in to the AI Chat Podcast.

21:53Make sure to rate us wherever you get your podcasts and have an amazing rest of your day. Thank you so much.