search

Tom Robertshaw on the present and future of machine learning for e-commerce

Cathy Reisenwitz

by Cathy Reisenwitz on October 27, 2020

Tom Robertshaw is Head of Engineering at Space 48, a leading CX and e-commerce platform agency. He created a fun and illustrative video showcasing how he uses Clockwise in August. Both Clockwise and Space 48 are using machine learning to improve user experiences. I reached out to pick Tom’s brain about what product teams should know about the present and future of machine learning in the e-commerce space.

We talked about how product teams are currently using machine learning, how machine learning can power new levels of personalization and customization, how to ensure the artificial intelligence (AI) doesn’t creep the customer out, and whether we’re entering an AI Spring or AI Winter.

How product teams are using machine learning in e-commerce

Clockwise: What’s your advice for product teams who want to use machine learning to provide features that improve the customer experience?

Tom: Interesting question. Day to day, I'm working with e-commerce merchants. Most of my experience with machine learning has been experimental and from my computer science background. The terms “AI” and “machine learning” have been over-hyped for the last four or five years. Now we're coming to the other side where advertising that there’s AI or machine learning in your product doesn't necessarily give you much of an advantage.

Merchants recognize that tools that have some form of machine learning are going to beat those that don't. But it's not the be-all and end-all.

And so as product managers, it's about being useful to those merchants, providing value, and adding features. Machine learning is a strategy to achieve that. It's not an end goal in and of itself.

Moving beyond recommendation engines and transactional websites

Clockwise: Absolutely. I'd love to know how and where machine learning is actually improving the experience for customers or merchants. For example, you have machine learning powering chatbots and recommendation engines. What are some other, really high-leverage places to use machine learning?

Tom: Certainly in the e-commerce world, where merchants are quite used to machine learning in recommendation engines. People are very used to “related products” being powered by machine learning, recommending the products that you would have if you'd manually specified that this product goes with this product. The benefit being that in an ideal world it just works how you expect and it's invisible, in terms of how.

Where it's a little bit more bleeding edge is content management, and possibly even content creation. Content is where I'm spending a lot of my time at the moment in terms of creating great commerce experiences. Personalizing content is the next step in terms of machine learning.

Right now you might get some tailored product recommendations, but it's still very unusual really to get much in the way of personalized content.

So you're on an e-commerce site once, and you start browsing around and initially you just get the default experience. But as you come back, it recognizes you're interested in some trousers in a particular category. You could actually start to tailor those sites that offer a broad set of products with personalized promotions and campaigns.

Having a transactional website is kind of, [laughs] is, is not the goal anymore.

“Having a transactional website is kind of, [laughs] is, is not the goal anymore.”

It's much more about creating the right content for the customer. The right commerce experience. How do we not just create a new homepage? This could be achieved through just customer segmentation, more simple rules and groupings. But even that takes a lot of time.

Next-level personalization

Clockwise: That is, that is so interesting. It seems like the natural evolution of the recommendation engine is just to customize the website. To be clear, you mean that a user would come to a homepage of a storefront and if they've been there before then what's on the homepage would be tailored to them?

Tom: Yes. That could be taken all the way through to emails. You'd be very familiar with email segmentation and tailoring campaigns. So, why not for the website as well, if you recognize who the customer is?

Clockwise: Sure. And could you do this via outside data streams where you could do it for someone who's not been to the website before?

Tom: Hmm. Probably with some platforms’ help. I imagine there’s probably a way to do it if that customer is also logged into Facebook, something like that and you've got a relationship there then I would imagine there's a way to source that information.

Clockwise: And would you change the copy or the videos, or is this just a product thing?

Tom: I would say the copy.

Having that first impression be static, one homepage with one banner at the top with image sliders to try and fulfill internal stakeholders’ requests is less useful than to show the right image to the right person.

That's where content generation becomes very interesting. It's not just about showing like the right product. You've gotta be able to show the right content. You've gotta have lots of different types of content, different types of copy. There are articles that analyze stock market trends written by bots. You could see that technology roll out into more general content management. Most teams are overworked, particularly at the moment. Without that, personalization becomes an awful lot of work.

Making machine learning friendly

Clockwise: Totally. Do you have any advice for making sure that the AI or the machine learning experiences that you create are pleasant for customers to interact with?

Tom: One note there is around the ethics side of it. You asked the question about if we could track them from other websites they later visit. I think if you start knowing too much about a customer, or a customer realizes that you know too much about them, that can be unnerving for them.

It's not necessarily about trying to know the customer ever increasingly better. It's more about knowing what the customer wants you to know.

“It's not necessarily about trying to know the customer ever increasingly better. It's more about knowing what the customer wants you to know.”

So if I've come to the website before, I expect you to remember a bit about me. If I come to an e-commerce website and I've placed an order very recently, then I don't want to have to log in and go to my account page to find my recent order. I should have a bar at the top saying, “This is the status of your order,” cause that's probably the reason that I visited.

So it's about understanding the context as the customer would expect, rather than trying to glean every piece of information, because that can probably be overstepping the mark.

Similarly, some customers will choose to want to opt-out. There isn't a standardized way of doing that. Suppose I were ordering a gift for my wife. I wouldn't want Amazon to continue to recommend similar items, because then she might learn what I'm looking for. Some people might know ways around that. But I don't think there's a standard user interface way, when you start personalizing websites, for people to be able to ask to be forgotten, so to speak.

“I don't think there's a standard user interface way, when you start personalizing websites, for people to be able to ask to be forgotten.”

Clockwise: That's a really good point. I've heard complaints about it, but I never thought about a simple opt-out. That seems really obvious. Any other obvious wins that you can see for making AI less creepy or more ethical or just more pleasant?

Tom: Similar to the opt-out, being able to provide feedback about whether or not this is something you're interested in. I know that's quite common for tools. Let's say I'm not interested in this. The engine could take that as quite a powerful weighting for, “We've got something wrong here,” and that can help correct further recommendations.

Clockwise: That's a great idea. Up or down for recommendations. Is anyone doing a good job at that?

Tom: I haven't seen it in the e-commerce space. I thought about Spotify for music or YouTube for their recommendations. So, you might tailor those. Otherwise you watch something for 10 seconds and suddenly they think I'm really into gardening.

Is AI development growing faster or slowing down? (Are we in for an AI Spring or AI Winter?)

Clockwise: I've definitely noticed the hype around machine learning has calmed down a little bit. Is that just the way people are talking about it or is it the way people are acting? Has hiring slowed? Have clients been less interested in starting AI projects? Like how is it really affecting people on the ground?

Tom: In terms of my day-to-day job, I'm at least a party removed from it because, as I say, working with e-commerce merchants is more about, they want the end goal, whether or not that's like recommendations. They just want great recommendations or great personalization. They ask for that as much as they have before.

But I'm sure machine learning is still incredibly popular because there's still life in this cycle. [The history of AI shows AI development goes through cycles of faster and slower progress]. Even at the end of the cycle, there's still a lot of good uses for it. I hope that society is going through an education piece at the moment. And it takes time to go through that, to recognize that it's not always gonna work out.

Clockwise: You mentioned an AI winter. I've seen predictions that we're heading for an AI winter, and I've heard predictions that we're experiencing an AI spring. For example, GPT-3 from OpenAI. Where do you land on that?

Tom: I think it's incredibly, incredibly hard to predict. I'm not really sure how long those periods have lasted in the past. You know, the Alexa came out, it feels like many years ago now, but it doesn't really feel like, like voice has really taken off. I continue to see predictions about the number of like Google searches through voice beating that of texts. Maybe I'm just misunderstanding it but I fail to see that really coming to fruition. Most people use their Alexas for the time and music, and not much else. Even with Amazon being the largest e-commerce merchant in the world, I still know very few people that have even tried to place an order through Alexa.

I feel like maybe we've come to a point where we've gotten as far as we can with the current sort of wave of technology, and at the moment we're sort of optimizing. Maybe through deep learning or whatever content processing algorithm they've used this time around. Maybe that's the next application is an AI that can generate code so we can build software more quickly, because we don't have enough developers in the world. As the need for software increases, that could be a great way of increasing capacity.

Interested in learning more?

We so appreciate the awesome video and Tom being willing to take some time to offer his thoughts as a leader in the machine learning and e-commerce spaces. To follow Tom and his insights on machine learning, check out his LinkedIn and Twitter.

Cathy Reisenwitz

Cathy Reisenwitz

Cathy Reisenwitz is Head of Content at Clockwise where she oversees the Clockwise Blog and The Minutes Newsletter. She has covered business software for six years and has been published in Newsweek, Forbes, the Daily Beast, VICE Motherboard, Reason magazine, Talking Points Memo and other publications.

Sign up for our newsletter

Ready to try Clockwise?

Get started for free