Customer experience is becoming autonomous - Interview with Andrew Bialecki of Klaviyo

Yeah, thanks for having me, Adrian. And yeah, nailed it. Klaviyo first.

Yeah, sure. So by trade I was a physicist, an astronomer in undergrad, that became an entrepreneur and a programmer. And then, you know, I've always had this obsession with how do we allow people to scale themselves? You know, we think we talk about allowing people to be themselves at internet scale. And so we founded Klaviyo gosh, in 2012, so 14 years ago. And you know, what we help over 200,000 businesses, brands, organizations do is we are their CRM, the CRM for consumer businesses. So we say a B2C CRM. And you know, I think for the first 12 years of Klaviyo's existence we thought a lot about how do we give people the technology so they can aggregate everything they know about their users, their customers. We work with a bunch of educational institutions, their students, right? Understand who people are and then actually action that so you can create, we say, stunning experiences for every single one of those consumers that will result in, you know, could be more sales, more engagements all the way around. And yeah, what's changed in the last couple of years is obviously AI and now we're thinking a lot more about how do you make great customer consumer experiences, how do you make those autonomous. So I think we'll talk a little bit about that. And um, yeah, to give you a sense of the scale, you know, over half a trillion interactions. You can think of, you know, Klaviyo does three different things: we do messaging and marketing, we help with digital experiences so for your website and mobile apps, and then actually customer service. So we delivered over half a trillion experiences last year for brands everywhere from folks starting out nights and weekends like we were way back in the day to big iconic brands like Mattel, Glossier, Liquid Death, Loop Ear Plugs in the UK, Simba, Paul Smith. So yeah, lots going around.

Yeah. So, first let's talk about what that word autonomy means. So look, I'm old enough that I remember the first time my dad took me to a bank to deposit a $5 note, you know, to put in a savings account and start to have it earn some interest. And I had to give it to a person, you know, an actual bank teller. And that was the interaction, right? So if you wanted to deposit money in a bank, that's the way you did it. And I think a lot of us, if you grew up in the 80s and the 90s, you remember this day when a lot of even consumer experiences were very, very much intermediated by you had to go talk to a person for something to happen, called them on the phone, etc. We built Klaviyo because we felt like increasingly, obviously like the last couple of decades, that trend has been changing. More and more of those experiences were not going to be delivered by people but they'd be delivered by software. And I think the quality of those experiences, first it became just about automation, but I think actually now we can use technology, machine learning, artificial intelligence to make those experiences even better. So you can imagine that same bank now might be able to recommend to me, 'Hey, did you realize you actually have a lot of money in your checking account? You should put it in a savings account that's higher yield or something like that,' right? And that's something that software can deliver and it can tailor it to me, right? Or maybe it knows that, hey, I'm saving up to buy a home, right? And so there's some other products that they have that might be useful to me. So we look at Klaviyo is for any consumer business, we think more and more of those experiences you deliver to your customers are going to be autonomous. They're going to be delivered by software. And I'll give you a couple of other examples. You know, obviously when we do work with Klaviyo, a lot of us know us for marketing and messaging. Nobody is sitting there and writing by hand every single message that gets sent out, especially if you're doing personalization. Those are being generated dynamically, right? That's an example of an autonomous experience. We have a customer agent product which is like a customer service always-on digital representative for your business. Can help with questions somebody has that might be understanding your products or maybe getting support, they need to change something about an order they've placed. That's an example of software making things more autonomous. So we think about in the future, if you want to create truly amazing customer experiences, you're going to rely on software to be the actual way you can deliver that experience to every single one of your customers because it can do all the thinking. It won't run out of time and it scales. Like if you all of a sudden have a huge surge in demand to your website or to your business, don't worry, AI can handle that load and can elastically scale up. So that's one is about the delivery of experiences. The other thing that's really interesting is we think even the deciding what experiences to deliver is not going to be entirely done by AI, but more of the work will be done using AI as a thought partner. So even the design and decision-making will become autonomous. And just as an example, a product that we have that we actually just released called Composer. What does it do? Well, it will help you compose marketing campaigns, build them, but it will also look through past marketing campaigns you've run and help you figure out which campaigns worked best for which audiences. And it will surface those results and automatically tell you how to make adjustments to future marketing. So that kind of reasoning, that kind of data research is allowing more businesses to basically get better at the craft of doing marketing. And it's because of autonomy. You have these researchers that are going off and doing all that research rather than relying on you and your spare time to figure it out. So we think about autonomous in those two ways: both the decision-making and then the delivery. And then for humans, yeah, we're not... Well, I do think we have to, as society, get used to the fact that AI will become very good at a lot of the tasks and jobs that we historically do. But I'll tell you, AI is only as good as the data that it can train on. It hasn't really proven itself to be good at novel learning. So one of the things I'll give you two examples: something AI is very good at and something AI is not so great at. When we talk to businesses and brands, I often get the question of, 'Hey, can you tell me what other brands are doing like us that we might want to borrow from or be inspired by?' That's a job that's perfect for AI. AI will do a great job of understanding that. On the flip side, if you are a brand and you hire a creative agency because you want to do a splashy new campaign with new concepts that maybe nobody's seen, right? Maybe it's a little edgy or it's just different. AI is going to have a hard time doing that because it's only pattern matching on things it's seen. And so we very much believe that humans are, if anything, it's going to give us more time to be creative, throw some of our more out-there ideas there, or maybe what direction we want to send our business in, what products we want to offer to customers. That will still be a uniquely human thing.

Yeah. Okay. So two thoughts on this. You know, we normally think of customer experience and customer service as part of that. The first is, we have a value at Klaviyo of being customer first, and what that means for us is you have to start with the person you're solving for and work backwards. I think for any consumer brand, we all recognize, since we're all consumers, we interact with a lot of businesses and products and services. This idea that to a consumer, 'Oh, well, when you're marketing to me, that's one thing. When I'm asking for customer support, that's another thing. When I go to your website or open the mobile app, that's a third thing.' That's not the way any consumer thinks, right? They look at that as all one experience. At Klaviyo, we think about these as the three different surfaces that matter that make up the experiences that a CRM needs to deliver. And as a consumer, if I start a conversation, if I get a message, say, via WhatsApp or an email from a brand, I then click through to the website and I see something, I then pop open the little chat agent and have a conversation. I want it to understand everything about me, stitch through all three of those experiences, right? So one is, I think you're going to see this merger of these three things together. And because of that, consumers, depending on the medium, when I'm in email versus having a web chat versus browsing a site or scrolling through a mobile app, I'm kind of always a little bit in buying mode, a little bit in learn mode, and maybe a little bit in needing some support. These are all artificial constructs. I'm always a little bit doing both. I'll give you a real example. We're based in Boston, the northeastern part of the US. Every year with our kids, we build a little outdoor ice skating rink. We go to the local home improvement store. We have enough of a setup now with plywood boards, but every year I go with the kids and say, 'Hey, we're going to make some improvements.' We've been going to the same home improvement hardware store, so that store knows what we're up to. We're subscribers on their mailing list, and we know some of the folks that work in the store. I think the future of that business's experience is when I walk in there with the kids and say, 'Okay, we have an hour. We're going to pick some improvements we're going to make this year. Maybe we're going to hang some Christmas lights around our rink, or maybe this year we're going to paint a big center line or a logo underneath the ice that the kids can see while they're skating.' So when I walk into that store, I actually have two things. One is sometimes I'm looking for ideas. I've got some ideas, maybe the kids have some ideas, but I don't know. They've seen a lot of stuff. We grow up in a place that's snowy and cold in the winter. I'm sure they've seen some other families come in with some other brilliant ideas that maybe we could be inspired by. So what is that? Is that customer service? Is that marketing? It's a bit of both. And in that situation, I think you're exactly right that these kinds of contact centers, whether it's in the physical place or being delivered over a phone call, because I might be walking around the store having a voice chat with an agent, it's an opportunity for that store to one, help me find the things I want. Often when I go to the hardware store, I bought some light bulbs, they're the wrong size, I need to return them. That looks more like traditional customer service. But it's an opportunity for me to deepen my relationship with that business. This past year, one of the store clerks had some great ideas for how we could have a little picnic bench right next to our rink so it was easier for the kids to put on their skates and take them off. That resulted in hundreds of dollars of incremental sales to that business because I walked in there and had an idea. So I think you're spot on that this is the future, these things are merging. I think the biggest thing that's holding most customer service organizations back from doing this is it's just hard to measure that incrementality. In the era of digital data, that's getting easier to do. You can start to see some of that lift. One of the cool things we do at Klaviyo is when you have a chat, whether it's voice, web chat, or email with our customer agent, our customer service AI agent, we'll actually look to see what happens to that customer in the next hour, the next day, the next couple of months. And then we can show you how different customer experiences actually correlate downstream with customer behavior. That's been really helpful because I think if you talk to a lot of folks that run contact centers, they have an instinct that if they spend more time with customers, there would be lift there. But it's just hard to measure. And once those attribution systems are in place, I think that's probably what that one in five, that 20%, is: they're just getting good at measuring that. But I think it's going to get easier and easier to do.

Yeah, it is. Well, I said there's two things. One is a little bit of a mindset, psychological shift, and the other is a more tactical, like the software we use shift. So on the psychological side, I think you're spot on. I go back to this thinking from the consumer or the customer first and then working from there. I think a lot of organizations have traditionally set up systems, and this predates almost the internet, systems that were there to handle different customer needs. Marketing does one thing, and we market one way, and this is old school back when we were running adverts in a newspaper or a magazine or sending mailers out. And it's like, okay, that's very different. Then if somebody has a question, they need customer service. That's okay. They'll call our phone number or they'll come into our actual store. And so each of these were very divorced, and you thought of them as physically different people doing the work. That is changing in a big way. The fact that a lot of experiences are digital now means that those data sets would actually be hard. In the past, who got the newspaper, who saw that advertisement versus who actually walked in? These are all things we just couldn't track. But now that more things are happening digitally, who received this message and it happened over WhatsApp text message or email? Who visited my website or opened my mobile app? I can see that, that's digital data. Who had, even calling in now, a lot of that is done over voice over IP, it's digital, you understand that, right? Or who's chatting with me? You need this idea that that consumer data, it's one consumer. These things we should sort of rethink the way we organize even as people around these experiences. I think that's the psychological shift. And then the true software infrastructure shift is, okay, do you have software that can talk to each other? Because you're spot on. One of the things we've already seen that's really fun is take the home improvement store. I might be sending a mailer out, a message out to all of my customers, and maybe I'm personalizing that, realizing, 'Hey, Andrew, he's into winter sports, also I know he's got kids, so maybe he's looking for projects to do with them.' And they can see which products I'm clicking on and what I'm browsing on that website. So that's information about me, not super high intent, but it's some information. And then when I actually want to have a conversation with that customer agent, that customer agent can reference back to that data. 'Hey, I see you were kind of interested in some do-it-yourself projects with your kids. Would you like some more ideas? Or here are some of the kits you were looking for.' Or if I reference that and say, 'Hey, I just saw your...' We get this a lot. As consumers, we'd like to use shorthand. 'Hey, I got this message and I saw some cool projects for my kids. Can you tell me where I can go pick those up?' It can intuit which ones you were talking about. And then what's awesome with our customer agent is we've seen some really cool things. For example, in retail, our customer agent is picking up on people's preferences, like size or style if you're doing clothing. A lot of consumers want you to know that, but they don't want to have to repeat themselves. We now have our customer agent that will learn those things: 'Hey, Andrew, he's a size medium, or here's his pant length, or he likes to wear long sleeve sweaters, he doesn't really like t-shirts.' They'll pick up on that and store that in memory against that particular consumer. And then in the future, when you're doing marketing or that next chat experience or that next customer service conversation, they can reference back to that. So we call this a little bit like when you're meeting somebody for the first time, if you pay attention and you learn some things about them, like what cuisine they like, so next time you go out to grab lunch, you know where to go. That feels personal, it feels like somebody's listening, like you're being heard. And now I think every business has the opportunity to do that, and they can do it at the scale of thousands or millions of customers concurrently. So that's one of the really cool things if you get the infrastructure right, these softwares talking to each other. At Klaviyo, our belief is that marketing, your digital experiences, and your customer service, you want all of those softwares talking to a central brain of the business, which for us is the Klaviyo data platform that comes included with all of our software.

Yeah, it's a great question. So first, I'll give you an example from my personal life. We're finally through winter, so that's nice. I was setting up my open clause, these personal assistants, because we like to play with a lot of the newest tech. One of the things we've been experimenting with is just how many of the tasks that we do in our daily lives can these agents take on? It's obviously not great at all of them, but I think if you want to live in the future, you have to constantly play with them and check them out every couple weeks and see what kind of progress we've made. So one of the goals I gave it was, 'Hey, I'm a pretty big runner. I need a new pair of trainers for the spring. The old ones are looking a little worn.' So I told the agent, 'Hey, look, this is the pair that I have now. I want you to go out and check out for this pair of trainers, buy this brand. Where's a great place to buy them? And I want you to balance price and shipping. How fast can they get there? And here you go, do it.' In fact, I even gave the agent a credit card with a limit and said, 'Here, go off and make this purchase.' And you know what? It did the whole thing. So when we talk about the future is agent to agent, everything we design at Klaviyo, whether it's the marketing experiences, these digital experiences, or these customer service experiences, we think about what is that going to feel like to a human? I mean, certainly that's not going anywhere. We're still going to do a lot of browsing around. But also, how will an agent consume these? Are they really consumable by agents? Are they easy to understand? So this is where I think the agent to agent comes in. And I think for some of us, you might find that example I gave of, 'Hey, I'm going to do a bit of price shopping. I know what the pair I want is, but also I want something that will show up in the next couple of days, not in two weeks, or something where it's in stock.' I could go browse 10 websites myself, but I can also just have an agent go do it for me.

I think this is one of the use cases that we're already seeing. When you're designing experiences, it's smart to design them so that agents can understand. What does that mean practically? We're envisioning a world where an agent visits a website and may not want to browse the whole site, but instead connect directly to a customer service agent, like the product Klaviyo offers, and just ask its question.

Right. So we expect that will happen. Regarding whether humans will stop buying, it gets interesting when you consider shopping. Some purchases are more transactional, like buying milk, bread, or groceries. You know what you need, and there's less serendipity. Social advertising works well because people are in a lean-back mode, browsing and looking to be entertained and surprised.

Right. I think we'll see an explosion of creativity in consumer browsing and shopping. I recall a clip from the early 90s where a US news outlet showed early retail websites. They imagined an avatar walking through a virtual mall, but that seems terrible now because the walking is tedious. With AI, the current website experience will feel archaic. We'll repurpose content into experiences for when we have less intent and seek entertainment or serendipity.

Yeah, another example: we're integrating our customer agent with a chatter interface into search result pages for some brands. Instead of searching for tops, shorts, and shoes separately, we're rewriting the search results page so you can say, 'I'm looking for a whole new running outfit,' and it puts together the pieces, shows them on a model, and offers variations. This is coming in the next few months due to AI.

Yes. With generative AI, vibe coding, and now vibe marketing, our minds go to preventing low-quality content. For a brand, representing yourself to customers is everything. In our Composer product, we built 'guardrails.' These are deterministic checks—like a self-driving car staying in its lane. Guardrails include dozens of rules that the agent must pass, and businesses can add more, such as legal disclaimers. Interestingly, guardrails were popular for both AI-generated and human-created content. I spoke with a major athletic retailer in Europe who said these could speed up their marketing teams. With teams in 50 sports and dozens of countries, approvals are slow. Guardrails give local teams more control, letting them be creative while the AI ensures brand consistency. This helps larger businesses with multiple product lines enable their teams to go faster.

Sure. With vibe marketing, we want people to feel safe being aggressive with ideas, knowing software has their back. At Klaviyo, we hate when ideas die in your head or a spreadsheet. Some examples: a health and skincare brand called Odds was an early adopter of our Composer. They used it to generate content about skincare routines, personalized for skin types and conditions. Those campaigns generated 50% better open rates and 40% more sales per recipient. These are significant improvements. Another is Kani Kulani Sikinis, a swimwear brand, using our customer agent for sizing and even helping with vacation planning and packing. Lastly, LifeStraw, a water filtration company, segmented content for home users and hikers. They saw better results and embedded that into their customer agent, which increased sales by answering technical questions without scaling support.

I don't think so. I encourage everyone to think consumer first. At Klaviyo, we believe there's untapped human creativity in delivering stunning experiences. If you can dream it, you can make it possible. AI is now good enough that if you're not asking it for ideas, you're missing out. The best brands constantly ask AI for best practices on launching products, customer service, or incentives.

If you want to improve your customer experience, make sure you find software that has access to your data and either has AI agents built in or can connect to something like ChatGPT or Claude. Ensure the software knows everything about your customers and you have another brain to connect to it.

Loop Earplugs in the UK. They treat earplugs like lifestyle accessories, with different ones for concerts, the beach, or work. They use quizzes, surveys, and AI to help customers build their wardrobe of earplugs. It's creative and engaging, leading to more sales and customer engagement.

I'm excited about a rocket launch from the US and an international team going to orbit the moon, with plans to put humans back on the moon after 50 years. It makes Earth feel like one society taking on hard engineering. Also, compared to 50 years ago, rockets rely more on software and AI, which has lower error rates than humans. It's a testament to our technological progress and the exploration ahead.

Excellent. Thanks for having me, Adrian.