Retail and eCommerce companies are embracing AI and machine learning (ML) technologies in order to gain valuable insights, automate marketing, forecast demand and much more.
Implementing a successful AI system includes model building but it goes far beyond that. It needs data, a clear strategy, infrastructure and the involvement of various departments at the company.
We caught up with Resham Sarkar, Principal Data Scientist at Slice, at DSSBreak. Slice is an online food ordering platform for over 16,000 independent pizzerias, where Resham leads Machine Learning and Personalization and develops ML models that contribute to the growth of the business. She will be talking at the upcoming DSSVirtual for Retail and eCommerce on August 25th and gave us a sneak peek of the ML success story at Slice.
DSS: Slice won this year’s Fast Company’s Most Innovative Business Award. How did data and machine learning contribute to this success?
Resham Sarkar: Slice is a company where data and strategy go hand in hand. They can't exist without each other. If you have data and no strategy, you’re just going to keep building and dissecting models all day without a strategy to pursue. And having a strategy without the right data makes the strategy unactionable. At Slice, there is a perfect storm of both these things. They drive one another. This is the secret behind what makes us innovative.
"Slice is a company where data and strategy go hand in hand. They can't exist without each other."
Additionally, we have a really good pipeline that collects data from all of our channels (apps, websites, etc) and a platform that curates all this data. This maintains the fidelity of the data. For example, we get triggers when there is something unexpected with the data. Additionally, the decisions made by everyone at Slice (marketing, sales and products teams included) are all driven by what data is telling them.
DSS: Where do you see the most potential for AI in retail and ecommerce in the future?
Resham Sarkar: We have seen a lot of growth in this realm during the last year as people are no longer shopping in person. It became more clear that we need a lot of AI and ML to drive and improve experiences online. One of the big things that will grow is chat boxes - more interactive and sentient chat boxes. Although there is importance in speaking with real people, data scientists will become starters at building these more sentient chat boxes which will have a better impact and efficiency.
DSS: What will the audience learn from your presentation at the upcoming DSSVirtual with the talk “Physics, Personalization and Pizza”?”
Resham Sarkar: I’m very passionate about telling people that you need a general understanding of how numbers work before you get into becoming a data scientist. In this talk, I would like to give an overview of what we do at Slice that drives personalization in relation to physics and pizza! How do we use the fundamentals we learn in school to build personalization infrastructure.
"I’m very passionate about telling people that you need a general understanding of how numbers work before you get into becoming a data scientist."
DSS: There's a long way from having an innovative idea to actually implementing a successful model. What does that look like in practice?
Resham Sarkar: A lot of talking. Model building is the least challenging part because you know these are hard skills (you know it or you don't). However, when you start building something you need to have a lot of buy-in, meaning there must be infrastructures in place to make that happen.
If you want to build a personalization tool, for example, you need a lot of data for that. You need to understand your customer. Otherwise, you'll have a ton of data that isn't useful for this goal.
First figure out, what do you need? Do I have that? What do I do if I don’t have that? Where do we go from there? Then, what do I build with this? There's so many ways you can go with data. Next, you must have a very good partnership with your data engineering team to know if this is supported by your infrastructure. How do we pull the data and how will we digest it? Additionally, you need machine learning engineers to actually build the model and take a deep dive into the data and find issues that may need more data. They will let the data engineering team know, so there is a lot of back and forth between these teams.
"When you start building something you need to have a lot of buy-in, meaning there must be infrastructures in place to make that happen."
There should be good alignment with the product people too. This is what we’re going to build, why we’re going to build it and what it will bring us. Then, we want to deploy it. We must have a good partnership with software engineers if you’re digesting it on a website or app because your model won't do any good if it's not deployed. Then once it's deployed, there must be experimentation. The feedback cycle is crucial. A lot of talking with a lot of different people is crucial to make sure everyone is aligned or on the same page.
DSS: What tools do you like to use at Slice?
Resham Sarkar: From a machine learning perspective, we found that it is very helpful to have some sort of a notebook infrastructure like Databricks as it is very easy to go on, build things and see quick results. We tend to use Python for most programming, for ETL we use SQL and Optimizely for the experimentation process. Overall, I’m very tool agnostic. There's a big push in the industry to use the most complicated deep learning models, but I tend to challenge that and find a cheaper way of building. I like to use whatever gets the job done.
"There's a big push in the industry to use the most complicated deep learning models, but I tend to challenge that and find a cheaper way of building."
Join Resham for her full presentation about "Physics, Personalization, and Pizza" at the DSSVirtual for Retail and eCommerce on August 25th. Use the discount code DSSRoundtable to redeem one of 10 complimentary conference passes.