Recorded live at K:LDN 2026 in London, this conversation is about the thing every brand now has in common. Everyone has access to the same AI tools. So what actually separates the brands winning with them from the ones just using them?
.jpg)
Recommending the wrong whiskey to a loyal customer does not just miss a sale. It breaks trust, and once trust breaks, no amount of personalization copy fixes it.
Recorded live at K:LDN 2026 in London, this conversation is about the thing every brand now has in common. Everyone has access to the same AI tools. So what actually separates the brands winning with them from the ones just using them?
Phillip Jackson sits down with Jake Cohen, VP of Insights at Klaviyo, and Tim Martin-Harvey, Head of Ecommerce at The Bottle Club, a UK multi-brand alcohol retailer carrying roughly 9,500 products. Their answer: memory. Not the AI kind, the brand kind… meaning the stored, structured context a business builds about its own customers and products over time.
Tim explains how one mandatory checkout question, asking whether an order is a gift, for self-consumption, for hosting, or for trade, reshaped his customer insight and exposed why standard RFM and lifetime value metrics break down across different buyer types. Jake widens the lens, arguing that loyalty is better measured through engagement across touchpoints than through money spent, and that the brands seeing real gains from AI are the ones writing customer and product knowledge down as reusable context, what Klaviyo calls "artifacts."
The conversation gets specific fast, down to the exact wrong recommendation that can cost a brand its credibility, and closes with Jake's straightforward plan for putting this into practice over the next 90 days.
"The word of the moment to me is actually context, and that context, if you can store it effectively and leverage it effectively, is the way that you can create a moat, because you can serve more people more personally, more memorably, which will create deeper relationships and, of course, more durable business over time." — Jake Cohen, VP of Insights, Klaviyo [2:08 to 3:04]
"What starts to become very important in the world of AI post LLMs is that the most important thing a brand can do is show up for someone the way that they need when they need it." — Jake Cohen [9:09]
"I genuinely think Klaviyo agent makes the most sense to be the agentic storefront, and that's not just me Klaviyo championing it. It's genuinely got the most context from multiple sources." — Tim Martin-Harvey, Head of Ecommerce, The Bottle Club [18:49 to 19:48]
"The answer should not be, 'Great, here's 10% off, go buy one.' The answer should be, 'How long are you running? Do you have a color you're interested in? Do you have a race coming up?' As you start to collect that information, that helps build the context for that individual, and they become the type of customer that will stay with you for a lifetime." — Jake Cohen [10:26]
Have any questions or comments about the show? Let us know on futurecommerce.com, or reach out to us on Twitter, Facebook, Instagram, or LinkedIn. We love hearing from our listeners!
[00:00:01] Phillip: We've been hearing for years about the ways that brands can personalize every experience with their customers. For maybe fifteen years now, I've been hearing that this is the year that personalization is going to really take hold. But I think now AI gives us the ability to actually deliver on that promise. But it can't do that without putting the data at the center, giving context to the brand, and actually giving memory to every experience. When you do that, memory becomes a moat for your brand experience. Why do we need a moat? Well, the truth is everybody's using the same tools today. For years, we've had the homogenization of experiences where everybody's using the same tools to build the same websites with the same best practices. You have the same top rail and the same left side filters. Every website looks the same. And we're on an accelerated path today where every single piece of text that we're generating has the same exact effect. It's your data at the center that makes every touchpoint unique. And so your memory becomes the moat. Today, we're going to sit down with Jake Cohen from Klaviyo and Tim Martin-Harvey, and they're going to tell us how brands like The Bottle Club are taking this data that they've been sitting on for years and they're putting it to great use.
[00:01:21] Phillip: And then how they're going to build new software, new services, new experiences to get new data to pilot the experiences of the future. Because from now on, memory context is how you differentiate yourself from the competition. Alright. Monday morning, I want you to walk into your all hands meeting, and I want you to ask the question, do we have a three sixty unified view of the customer? And when they say, yeah, absolutely, sure we do, I want you to ask, where is it? And can our AI tools make use of it to drive some concrete outcome? Something that actually ticks a box on our road map for the year. And undoubtedly, they're gonna point to a spreadsheet. They're gonna say, yeah. Absolutely. It's right here. It's in a spreadsheet named customer underscore final underscore final, all caps, real one dot x l s x. That is how most business is done in 2026. And that's why Klaviyo's latest release feels super relevant to most teams right now, because Composer takes this busy work off your plate and helps you actually focus on the customer relationship work that requires a human. So Composer can audit your campaigns. It can identify what you have missed already, what you haven't built yet, and then it can build those missing pieces, and it hands it back to you with the data that triggers your flows that you just got set up with Composer. And Klaviyo is bringing all of that back into the customer profile, which makes it part of the flow, which is part of what happens next in personalization so they always feel like they matter to you as part of the brand. And that is where CRM means it's more autonomous, more connected, and a little less dependent on everyone in your team remembering which tab the customer is hiding in. So you can see everything new from Klaviyo at klaviyo.com/new. That's klaviyo.com/new. And remember, you heard it on Future Commerce, klaviyo.com/new. Let's get back to the show. Hello, and welcome to Future Commerce, the podcast at the intersection of culture and commerce. I'm Phillip, and we are live at K:LDN 2026. And today, I'm sitting with Jake Cohen, VP of Insights here at Klaviyo. Welcome back.
[00:03:59] Jake Cohen: Thank you for having me.
[00:04:00] Phillip: It's been a long time.
[00:04:01] Jake Cohen: Thank you for having me.
[00:04:09] Phillip: I'm glad to have you. We're gonna talk a little bit about the deep customer understanding and data intelligence layer that is happening right now in AI and how customer intelligence is more important than ever and the context layer that's required for you to deliver on customer expectations when executing AI in your brand. And no better two people, I think, to talk about what is required today than the brands that are doing it and the platforms that are helping to enable it. I've heard a lot about the age of AI. But I heard you say something that I think is a phrase I hadn't heard yet, which is memory is the new moat. Could you explain that a little bit? And let's talk about how memorability influences customer behavior with a brand.
[00:04:56] Jake Cohen: So memory... there's two things that we're talking about. Memory that a brand can use to create an experience that is memorable. That memorability word that you mentioned has a lot to do with the... where variant from the norm of what we go through. That's how the brain works. What's difficult today is to use machines to take advantage of the promise that AI is giving that out of the box, can use something, and it magically will deliver a surprisingly good experience that allows us to stay even more connected with a brand. The way that you can do that though is by storing lots of context about the person that you are serving and about the things that work well given how people feel about your brand. So the word of the moment to me is actually context. And that context, if you can store it effectively and leverage it effectively, is the way that you can create a moat because you can serve more people more personally, more memorably, which will create deeper relationships and, of course, more durable business over time.
[00:06:16] Tim Martin-Harvey: I do.
[00:06:17] Phillip: We're all using the same platforms. We have been using the same platforms for, like, ten, fifteen years almost now. The same language. And so the context is highly important now. Yes. Because that's the thing that gives you the differentiation between your brand's experience and every other brand's experience that's trying to use the same tools.
[00:06:45] Jake Cohen: Exactly correct. And specifically, we can get into as much detail as we want here, but specifically, you have to create artifacts, written information that represent who your brand is, what it stands for, who the customer is, what they're interested in, and make that available to the tools that create the experiences that you deliver to your customers. Mhmm. At Klaviyo, we are investing a tremendous amount of energy to reduce the friction to create those artifacts and leverage them so that you can have these special experiences. But that foundational approach is true whether you're using the Anthropic ecosystem, the OpenAI ecosystem, any other tools. You need to write down what matters to you and to your customers so that those tools can use it for you.
[00:07:30] Phillip: Tim Martin, who's the customer? What artifacts are they creating? What are they interested in?
[00:07:35] Tim Martin-Harvey: So for us, we realized that we were making assumptions of what we were purchasing. So we're a multi-brand alcohol retailer. We got nine and a half thousand products. But the most important thing for us is why they were buying and what they were buying for. So we had an assumption that 50% of all of our customers were buying for gifts. That was assumed by the fact that they were buying products that were in a gift box or a gift pack, which logically makes sense, but we had nowhere to actually genuinely quantify. So a couple of years ago, Shopify bought Checkout Blocks, and they basically opened it to all... yeah... gas merchants. So we now have one mandatory question at checkout that basically asks the simplest question that's most important to us as a business, and I think every ecommerce merchant has a question they need to ask, and they might not have worked on what it is yet that defines the context. But for us, is it... was that order for a gift? Was it self consumption or were you hosting or was it a trade order? And that has unlocked so much more insight for us to be able to do so much more with it. And we've learned a lot more. And then we've even tested, like, push Klaviyo to its current limits with RFM analysis because of... and also we realized that the products that are the most gifted are ones that weren't in the calculation before.
[00:08:44] Tim Martin-Harvey: The ones that were in the calculation weren't correct. So I'll give some examples. One of our most gifted products on the website is a naked bottle of Grey Goose vodka, which is a standard bottle. I wouldn't have said it was very gifty. And then we've got the McLaren edition of Jack Daniel's, a nice orange bottle. It's in a gift pack that's classed as a gift in our logic, but actually it's a self consumption product because it's a collectible. And collectibles are something we ignored. There were things that we would never have kind of thought of. And until you actually went and looked at that data, it unlocks so much more about that product in particular. And I was just on that RFM point. The repeat purchase rate of a gifting customer, maybe annually. So they're not at risk. They're not churned because they haven't purchased in the last three sixty five days. They're not going to... or they're going to buy a specific key buying period during the year, which RFM doesn't allow for if you look at it on top level. So look and obviously, Klaviyo is built around segmentation and its individual audiences. But if you don't apply that same kind of logic to the calculative metrics that we're using, they kind of don't really mean anything. So, like, LTV is false for a gifting customer versus a self consumer. So it's applying that context that is paramount for us.
[00:09:53] Phillip: The old way of looking at it, especially in looking at RFM analysis, is we have these cohorts of customers that behave in certain ways. And in order to try to influence behavior, we would try to get them to move from one customer segment to another, and we would create flows in order to push them into those behaviors. Because if one customer buys like another customer, their behavior would theoretically be the same. But I think we all intuit that the world doesn't quite work that way.
[00:10:21] Jake Cohen: Not so clean.
[00:10:22] Phillip: Right. So how does this world actually work, and how does loyalty actually... ?
[00:10:31] Jake Cohen: Yeah. So I don't fully agree that the buckets that we might use to inform how loyal or deeply connected... I like better... someone is with our brand. I don't think those are wrong necessarily, but what matters is how you define them. And so if they are purely based on a certain amount of money spent, then yes, I think that's pretty antiquated, and I also think it's exceptionally shortsighted. However, if it is based on how engaged someone is across multiple channels in terms of the touch points that they have, in terms of how frequently they interact, more than just the purchase, but like the full experience, then you actually get a richer sense of who cares. And who cares is really what matters because then you can solve their problems, and then revenue will come. So for example, I'm from Klaviyo, so I'm biased. I know that world well. Our definition stretches far beyond simply how much they buy to be used. What starts to become very important in the world of AI... positions... you... we've invested so much in the service side of our business, and the reason we put so much into the data layer, is because when someone comes to your business, you need to be able to give them what they need.
[00:12:03] Jake Cohen: Gone are the days of slamming a promotion in someone's face, hoping that it's compelling enough, cheap enough, rare enough that someone will buy it. Now, because people can get such personalized experiences, that's becoming the expectation. So what has to happen is when Tim comes to a site and says... let's say it's a running brand... I think I want new sneakers. The answer should not be, great, here's 10% off, go buy one. The answer should be, how long are you running? Right. Yeah. Do you have a color that you're interested in? Do you have a race coming up? This is a whole different thing. And as you start to collect that information, that helps build the context for that individual, and you can train a system to teach it to ask those questions instead of push a promotion... you for a lifetime.
[00:12:57] Phillip: Brands are sitting on years, if not decades, of this type of first party data already. How do we make better use of it now, and how do we go forward in using and in collecting actionable data in, you know, like, if today is day zero going forward? Because I think those are two different stories and two different actions.
[00:13:20] Jake Cohen: There's two options. Right. You can be very, very smart like Tim, and you can have built a whole bunch of homegrown technologies and approaches to mine a bunch of data to uncover the insights that he was just sharing. I don't... it's hard to beat Tim. In the other case, you can use tools that will do this mining for you. So again...
[00:13:41] Tim Martin-Harvey: I think you also need to work out what questions are in that case.
[00:13:43] Jake Cohen: That's fair.
[00:13:44] Tim Martin-Harvey: I think you gotta work out as a brand or as a business, what does that customer need from you? And then you said about... in a skincare, you know, what ailment, what skin condition are you trying to remedy? Rather than... and I think you can be, especially with case service, really proactive. So say for example, you've got a cream that's good for psoriasis, but they bought it for acne, you've actually gathered that at some point within that customer journey, you can go, great choice, I'm glad you bought from us, but I don't think that product's right for what you achieve. And I think that's when you build a relationship, and you said about what loyalty is. Loyalty is not so much about point scoring as you mentioned. It's about being the first to mind, and if you've had that positive experience of going, not only did they, you know, listen to what my issue is, they recommended something better. Proactively, to not waste my money and just take more from me. I think that's how you build a proper relationship.
[00:14:41] Phillip: The role of the head of ecommerce is quite expansive in 2026, and I think it's changed quite a bit. What are you doing right now with the first party data that gives you that edge? And I heard a lot... I mean, Jake said it... what is this competitive edge in building custom tools and tooling? Are you vibe coding your way to...
[00:15:04] Tim Martin-Harvey: Actually, not vibe coding. It's a lot of Google Sheet trickery and backing with Matrixify.
[00:15:10] Phillip: Yeah. For sure.
[00:15:11] Tim Martin-Harvey: We've got our own, like, margin based merchandising tool that we built as well, but just...
[00:15:15] Phillip: 10,000 people that are listening are like, oh, thank god. I'm not...
[00:15:17] Jake Cohen: One of...
[00:15:18] Tim Martin-Harvey: Yeah. Google Sheets is a good tool. Let's be honest, especially with Matrixify listening to every... But for us, I think on the segment side of things and the Klaviyo side of things, it's... we now understand that the customer may be a gifting customer.
[00:15:32] Phillip: Right? Mhmm.
[00:15:33] Tim Martin-Harvey: But our goal is now to work out how do we turn those gifters into self consumers? How do we work out what they like and how do we nurture them into self consumption and vice versa? The people that buy them for themselves, how do we incentivize them into gifting? We also see this huge opportunity in hosting. So then, you know, because those basket spends are much larger, but they're less frequent. And you wanna make sure that you've encompassed everything that they need. So our email communication for the audience is recipes and party ideas and more content based stuff where you build engagement rather than sell, sell, sell all the time. And then where it gets really nerdy and exciting is when you can start applying that into things like objects. So I was chatting to one of the other champions the other day and they... a pet food brand. And they're trying to work out how can we gather the data in such a way that we can store each pet as an individual object and, you know, because they might put two items in a basket, but actually, is that two for the same dog or is there two dogs? And you don't know that. Sure. You can't work out repeat frequency and then... you're not building that relationship, and obviously, pet food's always one where you wanna make sure that you're being sympathetic to the... you can be doing.
[00:16:44] Phillip: That's an important... memory is the moment. Yeah. For that's high context. You need to understand household information, etcetera. So it looked like you were gonna jump in there.
[00:16:55] Jake Cohen: Yeah. Can I mention the margin thing?
[00:16:56] Tim Martin-Harvey: Yeah. Of course.
[00:16:57] Jake Cohen: Okay. So one of the very cool things that Tim has done is done an analysis of what products are interesting at different points of the year, even in different geographies in some cases. Mhmm. And from that information, he can look, let's say, of the top five products that are interesting in this time of year in this area, which one actually produces the best margin profile for him. Because selling two different products for $80 is really different if your margin is 20% on 140% on...
[00:17:22] Tim Martin-Harvey: The other.
[00:17:22] Jake Cohen: Right. Of course. And so, as I was saying, not everyone can be Tim. One of the things that people can do is as they start to... prep messages or text messages that will be sent or communication through the service agent, you can say to this as a skill, when choosing products, I want you to prioritize margin profile. That's not something that will happen out of the box for any old system. That's something that you need to tell it matters more to you. That is part of this bucket of context. And if you store your COGS in the catalog such that the system can see it...
[00:18:06] Phillip: Yeah.
[00:18:07] Jake Cohen: It can start to look through and say, oh, I was going to recommend this pair of running shoes, but because three of these would work, I might slide up the one with higher margin. Mhmm. Still solves the problem for our customer, which is utmost importance, but it also does better on my business. And these... if you... these little things are now very, very, very possible. And the more you invest in teaching the system to do what's right for the customer and for your business, that's the context I'm talking about, that packages brilliant Tim into tools that everyone can use.
[00:18:38] Tim Martin-Harvey: Yeah. And I think there's a lot more that you can do with it as well. It's like making sure that those product recommendations are smarter, and that we've got a really detailed level of metafields that we need to reference. So actually bringing that in and, like, with the margin thing, we don't just do the margin thing. What we also do is we look at, like, weeks of sale. So with the weeks of sale, the amount of inventory that we have take to deplete plus mixing in the margin. But then you layer on that we also know on every single product if it's more of a gifting product or is it more of a self consuming product. Some people, when they're asking an actual agentic question, a proper query, they'll be asking things like price point. So taking the price point, taking the margin, you know, what's the occasion? It's for a gift. Like, hey. It can't look cheap and not showy. It needs to be quite smart looking. So and then also, they'll give things like context. So someone will say, oh, my dad really likes this product. What else do you have? And making sure that you've got enough of a structured data... recommending something that's got no margin in it because you're losing at the end of the day. But it's also for us with alcohol being so subjective and so nuanced through all of its categories. If somebody said my dad loves Jameson's and it went, yeah, perfect...
[00:19:54] Tim Martin-Harvey: Try Jack Daniel's. Like, they're completely different. That's, like, the worst recommendation. It also ruins our credibility. Like, we're the drinks experts, and our AI is giving just a crap... oh, it's whiskey, and they're about the same price point. Grab that. We've got enough structured data. And I think that's the challenge to Klaviyo, like, as you're already building it. But the Klaviyo AI stuff at the moment is so customer data and touch points. But then when we start layering on all the product data that's sitting out, like, it's not just gonna respond back to it's whiskey and it's ÂŁ20. It's gonna be, okay, it's Irish whiskey. It's blended. It's more used for gifts. And I think that's what our skill says, and that's why I was always really reluctant to put on an agent. And I think I was actually saying today, I genuinely think Klaviyo Agent makes the most sense to be the agentic storefront, and that's not me just like Klaviyo champing it up. It's genuinely like it's got the most context from multiple sources. Shopify is gonna be great for a product recommendation, but it isn't getting customer data. It's not asking enough questions. It's understanding the service side of things. So actually, if you're thinking about bringing in an agentic storefront, I kept thinking about who's gonna be the best play... which form is gonna be the best to get that data from. And realistically, Klaviyo is gonna be the best one because of all of the bits of information it's already got. So it's storing a lot of people data. It's storing a lot of order data. It's storing product data.
[00:21:18] Phillip: If a brand operator is listening right now and they're, you know, sitting on this mountain of data, they wanna use it as their context. They likely... maybe they're already using Klaviyo in some fashion. They just wanna go deeper. What's the thirty, sixty, ninety day plan to operationalize and start deepening that tactical side of getting going?
[00:21:41] Jake Cohen: In the first sixty days, you should set up your service agent and start making skills. We saw a demo earlier today that... the way that you do that, it's a basic text interface where you describe what you want, and there's an embedded agent that will help you create the artifact that is the context that should be stored and used as a skill by the agent that talks to your customer. So in the first thirty days, 100% start doing that. In the next thirty days after that, I would start to use the Composer tool to ask questions about your underlying data. Start to explore. Start to wonder. Start to have crazy ideas. Like, if I literally... for people who... in the case that you have Tim's data, for the people who bought a gift, you know, can I find a way to see if they liked the gift? And the answer might be, well, we don't have that information, but let's create an experience where we can reach out to those people and ask how the gift was. And then you could start, like, all... just start to dream. Right. Basically. And then as you get the feedback either from learnings or from new data that you... get ahead of AI because it turns out that of the brands that are seeing 75% of the gains that are coming from these tools, it's only the top 20% of brands that are getting it. The reason is because they are creating these artifacts. Yeah. They are embedding in the tools, and each subsequent interaction that gets better reinforces and makes the system smarter, and so you actually get this compounding curve of performance. If you start with the context, ask questions, and then improve it, you will start to see that too. So that's my guidance.
[00:23:27] Phillip: I love that... that is something very concrete to take away. We tend to be very dreamy and very visionary. But in this world right now, every conversation that's being had around AI sounds like it's something that's coming. This is something that's here. We can do it. We can do it right now. And we had a conversation earlier about scaling from, you know, 1,000,000 emails sent to 10,000,000 within... beyond... a month. That's the scale of acceleration that we're at right now, and I don't see why we can't hit the ground running today. I think this made it very real. Jake, Tim, thank you so much.
[00:24:00] Tim Martin-Harvey: Thank you.
[00:24:01] Jake Cohen: Thank you.
