Navigation
X Close

Resource Center

Staples’ Principal Data Scientist Drills Down On The Nuances Of Intelligence

Type: Blog

Innovator Interview by Ernan Roman
Featured on CMO.com

Elisha Heaps understands the value of data.

But Heaps is pragmatic about data, too.

1200x800_data_scientist-min.jpg

She is, after all, principal data scientist at Staples, the office supplies retail company, where her responsibilities include developing artificial intelligence (AI) and machine-learning techniques to drive its e-commerce business. Her work also has implications for Staples’ personalization, targeting, and pricing strategies, as well as how it approaches customer service.

“It’s not that every single piece of data is this magical unicorn that tells you everything you need to know,” she says. “But maybe there’s that one unique something that’s a bit weird, something that not everyone is going to buy, that is going to tell you something about that person. Once you have that, the conditional probability they’re going to buy something else is going to increase or decrease correspondingly.”

In this exclusive interview with CMO by Adobe, Heaps talks about how data has been a game-changer for Staples, fueling the retailer’s ability to drill into demographic clues, target each individual customer, and create a more personalized experience.

Can you tell me a little bit about your background and what led to your present position at Staples?

Sure. I started my career in ad tech, where I had the pleasure of working with billions of data points, basically deciding, algorithmically of course, which ads should be served to which user or which device. I was exposed to a lot of different types of technologies. Right before I joined Staples, I was a chief data scientist at a company called Digilent, where I was leading the ad technology. Then I heard of this great opportunity at Staples. My first project was working on the chatbot. The team worked like a startup within the company.

I was excited by the idea of working with retail data for the first time. It’s a different side of the coin going from advertising to retail, going with the customer on the entire journey—from when they land on the site to their final purchase. You can tell more of a story in terms of their behavior, so it was a very intriguing field to be in.

What are the key challenges you’re trying to address in terms of the customer journey?

There are many. We want to be able to predict how a customer is going to behave, ultimately to drive purchases, right? When you have somebody who first lands on the site, they may or may not be signed in, but the trick is to gather all the data you possibly can, to figure out the best way to help guide them. We are going to be looking at their views, their clicks, and their purchases. We’re going to be looking at the product information on the purchases they’ve already made to construct this story about them, all algorithmically, of course.

For example, if they’re looking at a given product, what else should you serve alongside that as they’re considering their purchase? When they’re on that checkout page, what are those last-minute impulse buys that will help round out their purchase, expanding what they would have otherwise purchased from you? You want to make that experience as good for the customer as possible because that helps you sell to them, but also it helps the customer.

Companies are surrounded by data. How do you sort through and prioritize the data that matters to drive engagement?

Not every single data point is going to be constructive. If everybody is buying the same ream of paper from Staples, it tells you that’s a really great product, but that’s not really going to give you a sense of how that person may be different from anybody else who walks in the door. But when you see that that person has made another purchase, like buying even something as simple as glue or a particular type of pen, now you have more nuance about them.

And what’s really great is that mathematically we can pick this up in a very strategic way and know that somebody who’s purchasing this particular style, this particular product, even a particular color, that’s what separates them from everybody else who buys that same ream of paper.

What signals and intelligence can you pick up from a transaction to intelligently develop a series of logical assumptions?

There are little demographic clues that every single product or SKU may carry with it. If you wanted to construct this whole persona for a customer, you could, but the value is predicting how that person is going to behave. You want to create behavioral personas about what somebody is going to purchase and what they’re going to be motivated to look at on your site.

We employ clustering algorithms, where it will tell you, “Hey, somebody who looked at this product is likely to look at this product as well.” There are a lot of different ways of grouping items together that are not necessarily associated with each other on paper. You have to observe what’s borne out in terms of what people are purchasing together, what they’re viewing together, and once they’ve purchased something that maybe is a little bit rare or maybe a little bit more unique—perhaps an unusual color choice—that’s going to be the ticket.

That’s the meaningful variation you can then latch on to, to find a way to make that whole experience feel incredibly more personalized for that given user.

Given the amount of products Staples sells, how are you ensuring the products you’re presenting have the highest likelihood of purchase?

Yes, that’s the trick. Studies have shown that if you present people with too many options, it becomes a choice paralysis, and they may walk away and not get anything.

It’s helpful to show them something that’s slightly less expensive, slightly more expensive, or maybe something that’s totally out there that you don’t expect them to buy because perhaps it’s three price categories higher, but just so they know what other features are available. They also see what value they are getting for their dollar if they were to purchase where they have landed.

Then there are the add-ons. You don’t want to inundate the person with too many things, and you also want to make it feel like it fits, like it’s right. It could be people who are buying a particular color of pen are also going to want to have Twizzlers in the breakroom.

The more we know about you, the more we can make this feel like a custom, personal shopping experience.

This clearly requires a huge amount of artificial intelligence. How are you succeeding at this sophisticated level of automation and yet also ensuring that human beings are available at key points in that journey?

You want to respond automatically whenever it doesn’t diminish an experience. Something as simple as processing a return, if you can automatically serve the form that needs to be filled out or have the bot redirect them to that information page where the return information is located, that’s great. You’ve just saved some time, especially with something that’s really straightforward for the business.

You want to switch to a human when more nuance is required, such as if somebody wants to ask about the differences between two products. That can be difficult for an AI to do with something that’s beyond specs. Now, the AI can look at the product information for these two different SKUs and say, “This has 100 more units of paper than this one, or this is blue and this is red,” but those are things the person probably was able to do on the site themselves.

It takes a lot of work to get an AI trained in a way that really feels like it’s a person behind it. That brings us to the other place where I think humans are really useful: The only way these automated systems are going to get smarter is if you have humans investing the time and training them the right way. Ideally, you have a devoted team of people go through all of the interactions that the AI would be predicting and saying how well it’s doing, and the AI would automatically learn from that. It gets better and faster with each iteration that you go through.

Then people can focus more on deep expertise and knowledge of a given space. That’s really where humans are going to thrive and let the AI do the low-hanging-fruit approach.

What are the three most useful tips you would provide readers about data?

The first tip would be to not focus on the quantity of data, but to make sure to focus on data quality as much as possible. Often you’ll see people focusing on just storing things and not taking the time to make sure that what they have is really good and recorded properly.

The second tip would be to really think about how the customer is using your site and to focus on allowing them to complete what they came to do as efficiently as possible and to also make it as pleasant as possible. The third thing is it’s really important if we are going to turn more to AI solutions that they work in lockstep with the human experts so that ultimately they can improve.

How will you define success in your position in 2020?

The easier we can make it for people to buy the things they need, then that’s success. It’s really all about helping the customer find what they need in an efficient way, in a way that is enjoyable. It’s not so important that people know the algorithms that are going on behind the scenes so much as they just walk away feeling like, “Wow, that felt personal. That was a nice experience. I’m going to come back and buy there again.”

What Our Customers Say

We could go on forever about the benefits of preference management, but sometimes it’s better to let our customers do the talking.

Listen to their challenges and solutions — and get inspired!

Connect

  • Twitter
  • RSS
  • YouTube
  • LinkedIn