Latest / Talk Commerce / ShopTalk Fall Live Exclusive: Session AI and the Future of Ecommerce with John Reagh
Transcript
- Brent Peterson (00:02.222) Alright, welcome to this episode of Talk Commerce Live from Shop Talk. Today I have John Ray, the CRO at Session AI. good to see you. We're in a giant equity room, the podcasting room, which is to be 4,000 square feet or something. Just us. Yeah, just us. No equipment this time, but hey, we're doing the best here. John, give us your day-to-day role, one of your passions. Yeah, okay, day-to-day, I get the enjoyment of heading up the revenue side of Session AI, which is outgoing after new clients and expanding existing customers. So I have a sales team as part of that, a sales engineering team, and of course work closely with the rest of the go-to-market group on the marketing side, and as well as our customer services teams. Passion. Well, we got a fun little introduction on exercise and things. I would say definitely I like that part of my life to keep my body in tune with my brain. But also, I love flying. So I have a pilot's license. And whenever I have time, I like to get up and kind of move around in the sky. that's a big passion of mine. Yeah, that's awesome. My son was a pilot. trained in Bozeman. He works for Delta now. nice. Very cool. He would always talk about how Bozeman, the mountains areas, Yeah. unique in the ecosystem of how the weather works. Yeah, because you're training in the mountains. It's a whole other level than training in the flatlands, right? You got to know and look around weather and obstacles. That's great. Yeah. So give us a little background on Session AI. tell us kind of what the core business is? Yeah. So the company started with three founders, very technically astute. So we're talking, know, masters and PhD degrees. Actually all from India originally, and looking to create a model using machine learning that would create or help companies understand intent. Brent Peterson (02:20.538) And that started as just being funded as a project, but no real business solution for the initial years, and then evolved into a product that was brought to market about five years ago. And kind of catching a little bit of the winds of change in retail is what has put the company in its position today. You'll see a lot of AI. I mean, if you're walking any trade show floor, AI seems to be kind of all over the place. Most of that is what I would describe and you would understand as generative AI. that's kind of outcroppings off ChatGPT with some layers built on top. This is behavioral AI. And so the difference for us is... We snap in pretty easily into what I would describe as more of an enterprise-sized retail site. And the engine is watching and learning once you tag the site and we get the traffic flow. And we're basically giving you a signal on traffic to say, hey, is this group or individual, are they highly likely to purchase? Are they not likely to purchase? Or are they potentially the most interesting group, which is, are they on the fence and maybe could be influenced? to purchase. So we generate a score. Brent Peterson (03:43.18) And then you could serve that up to a person who runs promotions and or marketing responsible for conversions on the site and generate some pretty impressive outcomes. session and session AI, will you do that in real time in a session? Yeah, exactly. It's a good point. We're generating this score in kind of the nanoseconds, if you will. scoring and giving this information to the retailer while the site is live. And that's what is also just quite a bit different is you can act and market or promote live during the session. The other kind of, I think, pretty significant difference here is with the deprecation of cookies, where the identity of your site traffic is becoming less and less known. we're doing this completely independent of other sources of identity. So it doesn't matter. We don't need to have any other information, no PII. We're able to generate the score regardless. So you're using the existing traffic to kind of learn the behavior of the customers and then maybe you group them based on the behaviors of them on the site itself? Yeah. So we're using about 70 different measurements and we're scoring them off of each other. And that is the outcome is a score from one to zero. Brent Peterson (05:20.334) And with the segmentation, think of if you knew someone was going to purchase, you don't necessarily need to get in the way or want to get in the way of that. You also don't need to incent them any further. You're already going to get that transaction. So you can save some margin there. And then the group that's not likely, I mean, we've all been to sites where you're solicited to enter your email address, usually within a second or two of landing for the 15 or 10 % discount. You can kind of, you can still use that as a tactic, but this maybe gives you a more creative and a less intrusive way to do that. But where you make the real money is if they're on the fence and they're looking, I mean, think of if you or I go into a brick and mortar store. and you're looking around, what would you do to incent someone to maybe become more interested or to have a purchase? Those are the things that we're talking about with Session AI. So when you're, so I'm familiar with the scoring model. think the scoring model was used a lot in post or during purchase with fraud detection. How do you work with, and the big problem in fraud detection is false flagging, right? Blocking somebody from checking out that's a real user. And you did address a little bit the question about if they're intending to already check out. How do you work with false flagging on what somebody may want and then you're giving them something, the wrong thing? Is it the learning part of it that plays into that? I think what you're talking about, if I understand it, is maybe further down the purchase funnel. So all of that fraud detection is typically when they're in the cart. Brent Peterson (07:07.766) So the other piece that's really unique to this is we're doing all this identification and scoring usually by the fourth or the fifth click on the site, which is in most cases in advance of the cart. So exit intent tools that are looking for people when they've added to cart and they've been through the PDP pages, et cetera, those are different, potentially complementary. But we're doing this all up front. So most likely well in advance of a fraud detection run that you're doing at a checkout page. I think you mentioned earlier just generative AI. That's what people think about AI nowadays. The predictive analytics and the analytics as finding patterns, that's what you're doing. You're well ahead of that. You're well ahead of the time that ChatGPT came out. How was the customer feedback or even peer feedback before ChatGPT and now after ChatGPT? Yeah, I think, again, going back to the tailwinds that we're enjoying, because AI has become such an incredibly compelling thing for executives of especially retail companies to talk about, I think we're benefiting from people being tasked with, I don't know what it is, but we need more AI in the company type directives. So that's only helped us. I would say some of the things that we see, happening though when what we're doing is we are in a way transforming how a retailer deploys discount and incentives. And we've tried to find maybe a balance on, hey, you don't have to transform completely right out of the gate, but there are areas or maybe gaps into your promotional calendar where you can put this AI tool to work, drive increased conversion rates, increase revenues, and kind of step your path. Brent Peterson (09:10.657) into what we would hope to be the end case and that's if it's a math equation and you're kind of crunching the numbers, the machine is gonna have a better outcome than a human is in terms of choosing what discounts need to go or what incentives or what nudges need to happen with a consumer based on data. So that's the long game. But all the way back to your question, early on, maybe there was some skepticism. about could we do this or is that the right thing to do? But we're benefiting from the marketplace today around the buzz and the excitement to get onto the AI bandwagon. From a merchant standpoint, from their tech team, does this also give them a tool? Brent Peterson (10:04.504) thinking of somebody going through a funnel and they get stopped because of a technical problem on the site or there's a 404 inside of that. Does it kind of help them understand we should have gone all the way down this path and we're getting blocked at this point? Good question. I don't think that's been a use case that's come out of the customers that we have deployed. It does give a pretty insightful view of the early flow of the customer journey, which I have to tell you, sitting with some pretty big companies that you would also recognize very clearly, kind of surprised that when we show our flow and where the consumer is going, that they're like, wow, looking around the room, why don't we have this other places to analyze? in kind of this granular level of detail. On the 404s and where people get stuck, not something that we're hearing much about. So yeah, not today at least. Where does a merchant see the biggest uplift in... Is there a PDP page, category landing page? Is there a place on the site that everybody seems to make some mistakes that aren't covered? I mean the product pages for sure tend to be where the people are. we're making a decision and serving either ourselves a nudge or a prompt or an incentive at the fifth click. Typically, that's where the consumer is. In terms of the biggest outcome, This is kind of bread and butter on ECOM. It's conversion rate increases and top line revenue growth. So those are the measurements of success. When you look at it in a broader sense, kind of that step function of how you could evolve it, we've got case studies with, I'll give you an example, a large retailer above $10 billion. Brent Peterson (12:06.586) spending on an annual basis about 100 million in incentive discounts to drive traffic. That's a site-wide discounts to come to the site. When we put the model behind that, what we found is that almost 50 % of the traffic that was highly likely to purchase based on our outputs was getting the discount. And so this is a group that you didn't necessarily need to incent. And so you start thinking about where can I redeploy or redirect those dollars to convert the customer that maybe came and was humming and highing and could have used that incentive to stay and maybe become a new customer and then have an added LTV factor on top of that. But there's lots of different things, but to your point or to your question specifically, we're looking for conversion rate increases, new customer acquisition, and then we're driving top line revenue growth. in A-B testing? Is this going to replace it all in A-B testing? Yeah. That's a accurate question to ask because we are running A-B the entire time. So we're always keeping either a 5 % or a 10 % as a control to compare. But we've talked to some really, really large retailers. And because we can deploy so quickly, We've been testing other things in an AB type scenario just because we can get that in center of that nudge out so quickly. So yeah, there's a possible use case or possible use cases there as well. I was involved in one of the PayPal mobile studies where we got clients on board to try different PayPal buttons and things like that. And one of their big drivers was putting a lock Brent Peterson (14:04.15) check out button. Little lock that says go ahead and check out now. It was like 50 or 60 % uptick and conversions Have you found any magic little magic things that you could share that you see that? Hey, this is a no-brainer But I guess in that case was that that one seems like would be a no-brainer but there's other ones that like different colors would spark different things and you wouldn't like you intuitively you wouldn't look at it and say that but the data shows that this works Yeah, yeah, that's interesting. think I are the sixth the majority of the success that we've seen is if you took a Retailer that was maybe becoming more and more dependent upon discounts or incentives and kind of became drunk if you will on the discount hamster wheel that being more targeted with the incentives drove the outcomes that they were looking for both on a conversion rate increase and then protecting the margin that again didn't have to go to those that were already on the path to purchase. The nudges though that have worked for the most part, thinking of that type of retailer, are just simply small financial incentives. Other nudges are around loyalty programs. So if we work together with the CDP and we can be told that this is a kind of a known or a loyalty customer, we can change that incentive to be points. Brent Peterson (15:41.062) And that is another just small nudge that moves this group again. That's interested, but maybe not quite all the way there to come across the edge and add to cart and to check out. When you talk about a learning engine, I think people know now JetGBT learns, but it learns from everybody. Yeah. And I'm assuming this is private. Like this is a private model for each client. how much data do you have to get to learn and how fast does that happen? Yeah, so the overall engine, I mentioned about 70 variables. That is a compilation of billions of sessions. But you're exactly right. When we deploy for each individual client, the engine is tuning for the way that the traffic flows and the way that that site operates. But for us really to go today based on the models that we use, we have a pretty tight ICP. So I would say, we would say maybe in public a million sessions a month would be enough. And depending on the use case, that is usually fine. But I would say it operates maybe even more efficiently at higher volumes. So at the of the five million and up is it's kind of really humming. I think if you look at the way we're going to market, our focus is starting with the large, very high-volume retailers, and then our plan is to expand into the middle market later. So a lot of clients that we're engaged with are in tens of millions in some, well over 100 million sessions in a monthly basis. That allows us to do really fast test and learn, just because you have so much information to crunch. and you use the billions of transactions as an anonymized data to feed the engine to get things rolling and then the customer learns or the customer traffic teaches the private engine. Brent Peterson (17:45.792) everything else as we're going along. Yeah, and it is deployed independent. But if there was some significant shift maybe in market, then we could tweak the core model way that it interacts appropriately as needed. And for most platform standpoint, you're agnostic, you plug in with APIs, or how does that look for a... Yeah, so kind of the fast path to get the technology on is to tag the site. And usually that's deployed through a tag manager of some type. But we're really agnostic to what the e-commerce platform is itself. Once it's, if it's running in a large site, there's a possibility to also run it in an SDK type format. But it's cloud based, it's also cloud agnostic. So we have a lot of partnerships. We have partnerships with the three main cloud providers. And between AWS and Google, which is where the majority of our clients run. That's a very, very easy thing to spin up and to scale up as native for traffic. John, we have a few minutes left. I said at the end of the podcast, I give everybody a chance to do a shameless plug. say this. Again, as we touched on earlier, there's a lot of excitement and a lot of interest in bringing artificial intelligence to business. And there's no question that machines can crunch and do things faster than humans in certain cases. Brent Peterson (19:21.826) But if you're gonna invest in AI, I think one of the things that you need to see a clear path to is does this generate a return for the business? And I think what we've been pleased to get in terms of feedback from our customers is with relatively little disruption to the way that they're operating today, we're able to get in and provide an ROI positive outcome. inside the confines of the way that they're going to market today. And sometimes that's the tune of 20 times plus what their initial investment is in our technology. So there's a lot maybe of noise in AI, but if you're a retailer that's above a million monthly sessions a month and you're concerned about, which this is a pretty broad brush, increasing conversion rate and driving top line, we have a really, really compelling offer. And so we're excited about what we're gonna be able to do and what continuing to bring to the retail space. Yeah, and I'll just add on that personalization is going to only become more personalized as we progress down the path of being able to personalize, hyper-personalize every single user. so important. Absolutely. John Ray, CRO of Session AI. Thank you so much for being here today. Yeah, my pleasure. Nice to be with you. All right.