AI and online retail - learnings from Etsy CEO Josh Silverman

Stuart Lauchlan Profile picture for user slauchlan August 16, 2023
Etsy has emerged from the pandemic stronger than it went in - and the rise of generative AI offers more opportunity for growth.


Unlike almost everyone else who after the pandemic shrank back to roughly the size they were before the pandemic, Etsy is massively bigger than what it was before the pandemic.

A proud boast from Josh Silverman, CEO of online marketplace Etsy. Speaking at the recent Oppenheimer 26th Annual Technology, Internet & Communications Conference, he said: 

People during the pandemic had very few choices and had to go try Etsy and a number of other places. Unlike almost everywhere else, they have chosen to come back to Etsy again and again and again. Now, even with orders of magnitude more choice, more competition, a highly inflationary environment, we’ve held virtually all of the spend, virtually all of the gains that we've gotten.

He cited some interesting user demographics, noting that Etsy appeals to households with incomes over $100,000. Those below that figure are where there is weakness. This is indicative of a wider trend, he suggested: 

When you look across e-commerce, what you're seeing is people that sell essentials very, very cheaply are the ones who appear to be growing and everyone else is having a tougher time. So for households under $100,000, there is really less opportunity for things they want, and they're really having to focus exclusively on things they need. And even for the things they need, what is the cheapest possible version of that thing they need? 

That's a cycle, and cycles come and go. I'm old enough now to have lived through three and to realize that you've got the down part of the cycle and the up part of the cycle, and our job is to keep building a great business that survives and thrives and grows through cycles.

Growth potential

That said, there’s a lot of runway ahead, he argued, nothing that only one in three women and one in ten men in the US have shopped on Etsy in the past year - and those who have only shopped three times during that 12 month period. So there’s room for expansion, both in the domestic US market and internationally. 

In terms of competitive differentiators, Etsy offers a number, with home furnishings being cited by Silverman as a case in point: 

If you think about all the things you want to buy for the home - wallpaper or throw pillows, couches - in Etsy, you have the opportunity to buy something that's truly unique and special just for you and get it at a really fair price and get it within a reasonable amount of time. Home is something that's about self-expression. So we think that's really a category that plays to our strengths, and it's our largest category. 

But again there’s a caveat coming. While Etsy sold last year “billions of dollars worth of home furnishings”, Silverman admits there’s a recognition issue: 

Yet, if you ask consumers, ‘name places to go shop for home furnishings’, only three percent of people in the United States will name Etsy...We sold billions of dollars of home furnishings last year [but] our unaided consideration is very low right now. So the opportunity [is there] to build that connection - if you think home furnishings, you should think Etsy. We think there's enormous upside in building unprompted consideration for Etsy in the category, and home furnishings is an enormous category.

The other big opportunity for Etsy, he argued, is improved search and recommendation: 

We really have gotten to be very strong in ‘what you're looking for’. I furnished my living room and I'm missing this one piece and I know exactly what I want. Etsy does a great job [finding it]. Our search engine has gotten dramatically better, and we do a really good job of that. But if you come saying, ‘I'm looking to furnish a living room, give me ideas’, Etsy throws listing after listing at you, goes straight to the very, very narrow and has not been great at walking through 'what style do you like?', and 'let me show you different'.

We're making tremendous gains right now on inspiration for people who don't already know [what they want]. For example, we launched Shop the Look just recently, where, when you pick one item that you like for the home, we will find you [others]. If you pick, let's say, an end table, we'll find a rug and lighting that works together with the piece you found. That's a great example of where these new AI technologies are tremendously more powerful than anything we've seen before.

AI, of course

Yes, of course there’s an AI angle! It’s compulsory these days. Silverman makes the case that AI tech is tailor-made for businesses like Etsy: 

For Etsy, more than most, [AI has] the opportunity to unlock really incredible gains, given that there's 115 million things for sale on Etsy and none of them map to a catalog. So the ability for AI to really help to organize the world for us, I think, is a huge opportunity.

There are already clear benefits, he suggested: 

We've made incredible gains in relevance. Now we understand what you meant, not just what you said. So if you come and you type into a search engine, 'cocktail attire for men', we're going to show you mostly sport coats. Even though the words ‘sport’ and ‘coat’ did not appear in your query, we understand what you meant. That's an example of neural network translators.

That same kind of technology in neural network translators is going to sound familiar to many people because that's what [generative] AI is also based on. We've been using that same kind of fundamental technology now for several years. As a result, if you know generally what you're looking for on Etsy, we do a really good job. 

Again, there is room for further improvement, he went on: 

It is the case though that for most queries, we have thousand, if not tens of thousands, of relevant search results. So when you come and look for something, we get you a relevant set. Now the question is, of the 5,000 to 10,000 items that are relevant, which are the 30 we should put on the first page of search results? 

Our big focus right now is understanding quality. We know that items that are visually appealing, for example, convert twice as well as items where the quality of photography, the lighting, the styling of the image isn't as good. We have historically used humans to try to curate what are things that are really visually appealing. Obviously, across 115 million items, you can't use humans, so we're now starting to train our AI models, to use the human labeling to train AI to say, ‘what is a visually appealing listing?’ and really have the ones that are visually appealing come to the surface.

There are other benefits for sellers around pricing and trust, he added: 

On pricing, we're also starting to use much more sophisticated technology to help our sellers, to give our sellers insight in how they should price their items, not just what else that's similar to your item is listed on the site, but which ones are actually selling and at what price. Giving our sellers that data, they can do a better job of setting good prices, and we can do a better job of understanding which items appear to be priced appropriately. 

This is important because there's a bottom line penalty for charging too much and there's a penalty for charging too little, Silverman explained: 

A seller who underprices may not be getting enough value for her work, but also buyers may think it's lower quality than it actually is. So getting it right - not too high, not too low, appropriately priced - is actually the sweet spot. We're doing a lot of work right now on pricing to help give our sellers the data. We don't set prices, our sellers do, but [we]  give them the data so they can do a better job on pricing. 

The other thing our sellers really want help with is promotions. When should they put things on sale? How much should they put them on sale for? For what types of items? That's another area where we can share a lot of insight. So, for example, most of our sellers, when they put something on sale, put it on sale for 10% off. We have some early insights that very often, 10% off is not worth it. It's not enough to cause someone to buy, so you've just sort of given up 10% of your revenue without getting much for it. So if you're going to put something on sale, put it on sale for deeper than 10%. Those kinds of insights, I think we can do a better job of getting and then sharing with our sellers so they can price things appropriately.

Overall, there are potential wins on both buy and sell side to be had from effective use of AI, he concluded:  

I think there's a ton of productivity that can be driven through this. It can make our sellers more productive because it can make their lives easier. It can make our team more productive because it can help us to write code. It can help our customer support team do right customer support. And it can help our buyers have a better experience by helping them to find things and have a better conversation with that, see about what it is that they need. 

My take

Sound theories. The test now is putting them into practice - and seeing how many other online retailers follow suit. 

A grey colored placeholder image