Latest / Cheeky Pint / Des Traynor on reinventing Intercom twice and the “four horsemen” of good AI companies

Transcript

Des Traynor on reinventing Intercom twice and the “four horsemen” of good AI companies · Cheeky Pint

• John (00:00:00): Did you ever work in a bar? Des (00:00:01): I've never worked in a bar. I have a Guinness keg in my house. John (00:00:03): Actually. Do you want another pint? Des (00:00:04): Yeah, sure, sure. I was talking to a guy who runs procurement and he was saying like, you have to understand, I'm just getting bullshitted to all day. It's just all day relentless, absolute nonsense in my face. John (00:00:13): People do not have enough empathy for the procurement person who just has to endure nonstop nonsense. Des (00:00:21): Absolute garbage. ChatGPT launched, I think, on a Thursday. I had a call with our head of AI on a Friday. Owen made the decision on the Sunday, I think it was, and we started working on the AI version of Intercom on the Monday. It was a lot easier to invest when being a founder was uncool. I blamed, genuinely, “The Social Network.” I blame it’s just kind of the entrepreneurial lifestyle. I blamed TikTok, I blame all these things. Soho, how? John (00:00:43): Soho House? Des (00:00:44): Yeah, exactly. John (00:00:48): You just show me your technique and I will learn from you. Des (00:00:4448): This will be the first time I've ever poured a pint of Guinness in America. John (00:00:51): My fellow Irishman Des Traynor is the co-founder of Intercom, the customer service giant turned AI company. He's also a prolific blogger and one of the most respected voices on product strategy. Cheers. Des (00:01:16): Cheers. John (00:01:16): OK. You are my first Irish guest and so I actually have a critically important question that has been burning this entire time. Whenever we release these episodes, all the YouTube commenters are obsessed with people splitting the “G.” This was never a thing for me growing up, a thing I never saw in Ireland. It's like an invasive species maybe from TikTok. Is this a thing? Des (00:01:26): No, it's not amongst anyone who you would respect. It's very much an actual TikTok thing. It's a slight bit of a tourist thing because of that, but it basically means you're drinking, I think it's a quarter of a point in your first mouthful. And it's like, I don’t know. John (00:01:39): Waste of good Guinness. Des (00:01:40):. Yes. John (00:01:41): Yeah, OK. It's a little, when I first came to America, drinking games. It's like back home drinking is a very serious activity. Des (00:01:46): Exactly. John (00:01:46): It wouldn't make a game out of this. Exactly. Yeah. Yeah, I was very confused by that in America. OK, and then my other stout-related question is I saw you opining on Beamish, Murphy's, everything like this. I dunno, what is your view on the stout landscape in particular? Des (00:02:01): I probably default to Guinness. There's been moments when I go Kilkenney, Beamish, Murphy’s. Usually it's like when you're in their respective hometowns, but for the most part I default to Guinness. For one brief day I tried Island's Edge. I don't know if you remember when Heineken launched the Guinness killer and it lasted all of six months or something like that. They were literally, I think literally they were giving it away at the end and still couldn't get rid of it. Yeah. John (00:02:26): Have you heard all the Guinness stats about that Guinness used to be a majority of Ireland's stock market? Obviously the canal system was built for Guinness distribution, but sometimes we can tell people that Guinness used to be a very significant part of Ireland's economy. They don't believe you, but the stats are really there. Des (00:02:41): And it's weird. It trickles into modern day where I live by Castleknock, a lot chunks of the Phoenix Park are still owned… exactly where the house used to be. Maybe they donated it to OPW (Office of Public Works) or something like that. But yeah, it's very much still kind of carries forward. John (00:02:57): Yes. OK. We're not actually going to talk about Guinness the whole time. We should also talk about Intercom and I was thinking in preparing for this. I'm very impressed by businesses that can reinvent themselves or maybe even reinvent themselves multiple times. You think about Netflix, they started with the original DVD-by-mail business and then, oh my God, the internet's coming and they had a few abortive attempts at streaming, remember the whole Quickster debacle, but then really cracked streaming movies and so you can watch “The Godfather” or you can watch whatever movie you want on Netflix. And then, of course, as they got more squeezed by the rightsholders, you actually can't go watch “The Godfather” on it anymore. Or you type any movie into Netflix and it's not there because now Netflix is all first-party content that they have developed themselves. And so they've reinvented the platform once again to be, rather than watching other people's content, watching Netflix content. They've twice reinvented the company from DVD-by-mail to streaming third-party content to streaming first-party content. Intercom strikes me as a business that has similarly been reinvented twice, where you guys got started with the intercom feature of you can talk to your customer through the website, and then you guys became a customer-service company, which is actually different for reasons we can talk about. And now you're becoming an AI customer-service company. That's my theory of Intercom. Is that actually an accurate theory? Des (00:04:16): Yeah, that's true. There's a bit of extra context I'd give it. So when we started, it wasn't really, it was about—this is literally when we started—and our initial plan was like—hey, you remember the internet because Stripe was in its early days back then as well, but there was no tooling to run a SaaS business. There was literally nothing. You were kind of abusing PayPal for payments and you're using MailChimp for talking to your customers and stuff. We had this idea that talking to your customers is really important so someone should work on that. And there was so much stuff in the early releases of Intercom. It was the first live CDP. You can actually see who's live in your product right now and what you're doing and you could store data against them. You could say, “how me all my premium customers,” and weird use cases like that still exist today. (00:04:59): I was in Denmark last week and I was like, “Oh, I should message all my Copenhagen customers and see you,” and all that stuff is you're still like, who else is doing this? So we started out with a very, very general-purpose idea, which was let internet businesses talk to their customers. And then we kind of fell in love with this jobs-to-be-done methodology, and one of the things you do in that is you look at how your product is actually used and then you iterate on it from that point of view. And that kind of led us down this path of sales, marketing, and support. And then I guess to skip a load of years here, let's just say 2020 things were great, 2022 things weren't as great and then it was like, we need to focus … John (00:05:34): Concise— Des (00:05:34): Yes, exactly. Let's just quote a lot of the reasons why or whatever, some sort of disease as well along the way. But yeah, so 2022 it was like, hey, Eoghan returned—Eoghan had left in 2020, I think—Eoghan returned, business had been in declining net-new revenue. He said, “We need to focus. We're going to focus on customer service.” And then a short time later, I think maybe like 10 weeks later, AI happened. ChatGPT launched, I think, on a Thursday. I think I had a call with Fergal, our Head of AI, on a Friday. I spoke with Eoghan all through the weekend and we made the decision, Eoghan made the decision on the Sunday I think it was, and we started working on the AI version of Intercom on the Monday and that was 2022. And I think, were it not for that, it's hard to say exactly where things would've gone, but certainly that's the reason I'm sitting here. John (00:06:18): And obviously people are naturally wired to be skeptical when they hear the AI version of Intercom. You have everyone and their mother out there saying, “We're an AI company now,” but you actually are an AI company now. And so describe what the AI version of Intercom actually means. Des (00:06:32): The biggest thing it means for us is our product Fin. So we launched Fin in March, I think 2023. We launched a few little AI features. We were the first people to actually build anything on the GPT-3.5 and then we launched Fin in March, GPT-4 launch day, and Fin was basically the first chat bot that worked, is the best way you could think about it. What that really meant was it could actually have conversations and answer questions, and when we launched it, it was doing I think 25% resolution rate and that was crazy numbers. (00:06:59): Today it's like 65%, and today Fin's resolving about, I think it's over a million conversations a week. It's handled about 40 million actual end-to-end customer service scenarios to date. It's growing over 200% year over year. We charge a dollar per answer so you can work the revenue out, or 99 cents even. It's becoming just this AI sort of growth story inside Intercom, which is already a mature SaaS business into hundreds of millions of revenue. But I think when we think about, “What does it mean to be AI?”, it's like, first of all, what is the future growth of your business? And the answer is AI. And then over the last, say, six months, we've been going hard on being a kind of properly deep AI company. We're now at a point where we're using our own models inside Fin, we're using custom reranker or custom retrieval summarization, etc. And we're doing a lot of this work—we have an AI lab of 50 people and we really just have gone all in on the idea that, obviously, Intercom still has a help desk product, but the entire future of CS is clearly going to be AI, and that's what we're all in on. John (00:07:54): Yes, yes. Maybe there's the repeated pattern in tech where the enthusiasm for technologies comes before the technology being ready. And so people are excited about computer games before the wave of good computer games. Des (00:08:05): Yeah, yeah. John (00:08:06): Or people are pitching mobile internet and you'll buy cinema tickets via the web. It's like … Des (00:08:10): J2ME and all that. John: Yeah, exactly. Des: And we actually had our version of that. We had a product called Resolution Bot. It was originally called Answer Bot, but I think Zendesk threatened to sue us because they had a competitive product. But Resolution bot was actually a good AI product at the time, but it was just AI wasn't there. And so that was the actual reason why we had such a headset start. We actually already had a little AI group ready to go. We'd already built a rag engine ready to go, so we were able to jump a lot quicker than a lot of folks. But yeah, I think a lot of these products, you're right, have kind of two or three stabs before they go mainstream. John (00:08:36): And there was this whole enthusiasm cycle for bots in 2017 I want to say, and the tech just wasn't there at all. Des (00:08:42): It was a horrible experience for customers. It was also quite clunky to set up for businesses, and at some point I think everyone looked, there was a genuine question at times where it was just like, is a web form not just better? And I think in a lot of cases it was. John (00:08:54): Yeah, yeah, yeah. Whereas I think now people are starting to have the experience of—the classic thing is you're talking to a bot, it's like please, will you please just connect me to a human? Whereas now it's like, can you just connect me to a bot? Des (00:09:05): We see that a lot. We see like, “Oh hey Jenny, sorry to bother you. Can you put me back onto Fin? It was actually doing a great job, it just thought it wasn’t. I asked a few too many questions.” But yeah, I think there's a general pattern we're noticing, which is that a lot of experiences are just better digitized because, partially because of human considerations. One of the reasons people go to the kiosk in McDonald's I suppose because as opposed to their actual counter is because they don't want to have to think out loud in front of a human. They're like, oh, “Give me a second. Do I want fries?” A lot of the reason why people prefer Waymo, for some people it's like I just don't want to have the conversation, I don’t want the awkwardness around tipping or whatever it might be. And I just think what we see a lot is once Fin answers one question well, people are like, “Oh, this thing's paying out. Now that you're doing that, I'm going to ask you all the things I was wondering about.” Whereas I think they'd feel probably nearly weird unloading all that on one CS rep. John (00:09:52): Okay, so you're seeing a lot of induced demand for people using interactive customer service. Des (00:09:56): Which is interesting. It turns Fin into not just being a cost takeout, but also it's like how much better would your business be if everyone knew how to do everything they wanted to do? And the answer is a lot of times it's a lot better. And it's not just the whole, I don't want any human, it's also people often—one of the biggest fallacies in AI is people compare it with this perfect human that does not exist, the driver that never crashes or in our case it's like, well, a human artisanal handcrafted answer and I'm like, yeah, let's pretend that would be there in seven seconds. It won't. It'll probably be 18 minutes. It's also not going to be perfect. It might not speak their language… John (00:10:29): Let's presume you're doing those handcrafted answers—which you're not—someone is busily trying to close the case before they move on to the next one. Des (00:10:35): So I think comparing it, I dunno, we have this thing where we expect our AI products to be flawless and we're totally tolerant of humans showing up only speaking one language and only working six hours or whatever. It's just, it's a funny contrast. John (00:10:45): Yes, yes. That's interesting, and it's interesting you talk about product onboarding here because I think of Intercom as you guys have a house view that product onboarding should be much better. And I remember a lot of the use cases you would talk about for the original Intercom—talk to your customers through the website—was that you can have personalized nurture tracks, and it's weird that you drop people into SaaS products and just expect them to be able to use them right, and you should see how people are using the product and then give them kind of specific steers based on their usage. And it sounds like you're coming to this vision again, which is: people should have better onboarding support, people should be nurtured along based on their use case, but now kind of interactively AI-pilled. Des (00:11:22): Yeah, I mean we talk a bit about the city of what is ultimately a customer agent going to be. That's what Fin will be as it grows up. It'll just become this way in which customer conversations are handled. And obviously the most direct attack here is customer service. I think every single customer touchpoint can be improved by basically immediate accurate answers available all the time. John (00:11:45): Isn't an obvious limitation—right now, you require customers to come up with a prompt. And if you look at why TikTok is so successful, it's like I would never prompt for, “I want to see videos of planes landing low over the beach in St. Martin,” but it turns out that's what I want to see. And similarly, people probably have many more things they need than they will actually come up with a prompt for and I think the product today is still mostly prompt-based. Des (00:12:10): It's reacting to what customers say. John (00:12:12): Exactly. A customer has to come along and type things into the box. Des (00:12:13): I mean today that's what customer service is. It's still kind of like, here's my problem and then we'll solve it. I think for sure there's obvious directions this will go as like, hey, what does a good customer look like? And maybe we can honestly infer that as well.But certainly “people like you should do things like this” is definitely an understandable domain, and then I just think working at the right level of interruptive help. You don't want to be too naggy or too pop-upy. It gets quite grating. But I think if you can get the first message right, you can sort of say, “Hey, if you come here, you're always going to get the thing you should do next or the thing that looks like you're stuck on.” If someone's on, I dunno, the renewal page and they have an error message, we know they're probably going to open the thing and we know they're probably going to say something that's got a lot of the context already there, so we can work out the right things to say and do. I think that's pretty doable. John (00:12:56): It feels like you could do a lot around, you train a model on what the customer is seeing on that webpage at that moment in time and use it to feed the answer and things like that. Des (00:13:06): We already do a lot of that already as in customer context, so knowing that it's John and he's on the premium plan and he's on the playlist page and there's an error on the screen, that’s all useful information when it comes to—because I think people, a lot of the YC, “I could build that in a weekend” type Hacker News crowd, I think one of the things they often they're thinking every customer support query begins with like, “Hi there, my name is blah, my user number is blah.” But actually most support conversations begin with “This is broken.” And you're like, “What's broken?” And so to solve that you need a fat reply engine that's just like, “Hey, let's chat about what's going on here.” But we realized quickly people will disengage. So any amount of extra context that when you say this is broken, and if someone says this is broken and there's a big red error box on the screen, we're like, “Well, it's probably that thing that you're talking about.” A lot of people just don't realize how deep you have to go to actually do a great job. And we do say if you install Fin today, you get 65% resolution rate after 30 days. That's shocking, but we have had to go really deep to actually get to those numbers and it involves all sorts of—every single smart thing you can think of we've had to do and then optimize and then find the right model for and all that. But one of them is customer context and that obviously answers a lot of things. John (00:14:12): What are the other smart things? Des (00:14:14): Abstraction. So I guarantee you've got no pages on your website that say “Stripe works really well for a dentistry office.” You probably don't have that in your docs. A very naive rag bot will be basically like, “Well, it doesn't say dentist and we're told not to hallucinate, so no, we don't do dentists. Sorry.” And the abstraction is in that case is well what is a dentist? It's a type of business. Does Stripe work for businesses? Can dentists be internet businesses? Well, we say we're great for internet business. So you're kind of working out what's the best risk-tolerant way to make grounded inferences without going over the cliff? That's one of, like, 27 different components of Fin. Then you've got obviously your rag and then you've got, you're like, “Hey, is an escalation appropriate at this time? Have they threatened something?” Every single type of problem, it kind of ends up you have to walk through it all to actually recreate customer service. I think a lot of times people will compare it with, “How do I reset my password? Ha! It found it!” And you're like, right, cool. That's like 0.4% of the scenarios you deal with when you're in customer service. There's a pattern. Did you ever see the movie Armageddon where—it's like Bruce Willis and Ben off or whatever—but the gist of it is they train a load of, I think it's like oil drillers to become astronauts and the comedy, the joke at the time that Ben off always says he got drunk and he recorded the voiceover for the DVD and he was like, “I always said, well, why didn't we just train the astronauts to drill oil? Surely that's an easier problem.” I think the thing that we're realizing with the AI movement is some version of what's going to happen sooner? Will AI people learn how to do CS or will CS people learn how to do AI? Thankfully, as I said, we kind of started off with CS and AI in our DNA. John (00:15:44): You say people, you mean companies in this case. Will OpenAI get better at customer service faster than Zendesk gets better at AI. Des (00:15:49): Exactly, exactly. I think we were lucky in that we kind of had already backed both horses somewhere along the way. John (00:15:55): One thing I find interesting about what you do is every company is thinking about AI right now. Every company had a board meeting in 2023 where the board is like, “Can we do a special deep dive on AI? It just feels like it's happening a lot and we need to be making sure we're on the leading edge of AI.” And then every company was like, oh, we're actually doing a lot in AI. For example, we've seen great automation wins in customer service and so it's like the joke about the bike shed versus the nuclear power plant, where everyone has opinions on how to build a bike shed. Similarly, everyone has opinions on how to do AI customer service, and so I'm curious how you sell given that I'm guessing a lot of your customers think, “Oh, we know how to do that. It's not that hard. We hooked it up to a model that we're actually very smart on this topic already.” How you sell in that environment where everyone has opinions. Des (00:16:40): Yeah, I've never seen the build versus buy thing play out more often than we do today, especially with certain— a lot of customers are like, you know that meme on Reddit, I'm not good at girls or guys, whatever. There's a lot of that where it's like, “Oh, you would never possibly understand is B2C shopping company.” And you're like, really? I've never heard of such a thing. Sometimes honestly, we just be like, “Hey look, Godspeed, you go and start building this PS here's a torture test: When you think you've got something good, run these hundred questions to us, let us know.” Oftentimes that's where they're like, “Yeah, okay, we think we need to buy your product.” But I think everyone has this idea of, in a move to AI, what can we definitely do? And we can definitely answer questions like “how do I reset my password?” And again, this is the back to the whole that's such a small amount. What they can't do is actually have conversations and all that sort of stuff. We're like, “What is your opinion on the president and how they're performing?” And a lot of times you don't want anyone to answer that question on behalf of your company. But I think a lot of times people—they dip their toes. It's almost like they fired a tracer bullet. They're like, “Yep, this seems like we're making great progress”—and every AI product has this problem where you make epic progress in the two weeks and then you hit this wall, this plateau, and two years later you're telling people, “Oh, Apple intelligence is coming in ’26” or whatever. So in this case, a lot of people start the project, feel like they definitely don't need to buy Fin. We just help them understand the difference between a good bot and a bad bot, and then they come back and they buy Fin. John (00:18:07): So where is the Fin business these days? I'm curious both just how it's performing on revenue metrics and then are you selling it to existing Intercom customers? Are you selling it to new accounts? Just how does everything work? Des (00:18:19): We're about 6,000 customers and growing quickly. Fin does about a million resolutions a week. We're charging a dollar per resolution, so you can do… John (00:18:27): Yeah so 50 million revenue run rate Des (00:18:28): Give or take. When we launched initially we sold just to our own customer base, and then as we kind of progressed we realized, hang on, moving help desk is a nightmare. You've probably done it once or twice. It's a whole ordeal, and Fin is brilliant. So we're like loads of people want this product but can't buy it. So we made the decision to launch what we internally call Fin standalone, or Fin for platforms. So now you can use Fin on top of Zendesk or HubSpot or Salesforce or any of those, as well. So we basically—Fin is available to everyone, and that's a relatively new muscle that we've been growing, but it's actually—that's kind of where we see a lot of the future growth John (00:19:05): And so is Fin like, you can connect your iPod to Windows for iTunes for Windows, but we hope that one day you buy a Mac and it's part of the whole digital hub strategy, or we're actually now all in on Fin the engine and whatever customer service platform you use is actually not a topic of huge interest for us? Des (00:19:27): This is such a core question that we kick back and forth quite a lot. John (00:19:30): This is the offsite debate that just… Des (00:19:31): Yeah, genuinely at least it's certainly one of them. The way we think about it first and foremost is the future is AI. So Fin just has to win kind of at all costs, including our help desk. Weirdly, our customers are like, they turn Fin on and are like, “Damn, this thing's good. Hey net 65% of our support volume, maybe we don't need X, Y, Z competitor and maybe we can go all in on your help desk too.” And we're like, okay, cool. That wasn't our game plan, but we're happy to help, if you know what I mean. I think the actual battleground we care most about, genuinely, has to be the AI agent. That's the one where we care about most, but it does produce a lot of demand for the actual help desk product too. John (00:20:12): Oh my god. Oh dear. Okay, Des (00:20:16): So you've actually played darts. Des (00:20:17): I've been around a dartboard. John (00:20:19): I'm curious what your AI stack looks like. Where concretely, what are the models or collection of models and prompts and everything that you are using in production? How do you handle model upgrades given that the behavior is changing so much under the hoods? How deep do you go in terms of developing the stack yourself? Then you can talk about the stack. Des (00:20:38): First thing I'd say is obviously Fin isn't one thing. It's like 27 different things or whatever. So every one of those is—whether it's the summarizer or whether it's the reranker or the retrieval engine or any of these or the direct answer, which is where we actually go and formulate the answer—every one of those is paired with the fastest, cheapest, lightest, most accurate LLM that can actually do the job reliably. Very, very high reliability. So that means there's no one particular model. So our primary partner will be Anthropic for cloud. We've architected it such that we can plug and play various different pieces, and whenever a new model comes out, or honestly a new idea for a new architecture is in—like we recently launched the ability for Fin to do complex queries, which would be, like say, go and issue the refund and update the name on utility bill or something like that—whenever we have to change the architecture, we have this kind of arduous torture test of thousands or at least a thousand CS scenarios where we have, here's the question, here's the context we're provided, here's what the current Fin answer to this question is, here's the best available human answer that we know of, and then basically here's what this new version would offer us. So whenever we say GPT-5 comes out or something like that, we're like, the reason we're not just—a lot of our competitors are quickly, “Oh, we never run a GPT-5.” And I'm like, “Ooh, I’d take a beat on that one.” You shouldn't assume everything's going to be great for your use case always. So ultimately we have to run it through this pretty expansive test to work out where the edges are. If it's scoring higher resolution, we need to understand why, because it could be just that it's trying more stuff, but that could also, the shadow side of that could be excess hallucinations or whatever. So whenever a model upgrade comes, we have to trigger this whole thing. But when we launched Fin, it was like 25% resolution. Today it's like 65. We've been increasing at roughly a percentage point a month give or take. But very little of that is actually because of the upgrades or the bumps from the models. Genuinely, I actually think—and I say this with a lot of respect and love for the CS craft—I actually think we've had enough intelligence for CS for quite a while. In fact, we published some material on this on our research blog most recently. When you look at people saying things like, “Oh, the latest, whenever, Grok can compete at mathematical Olympiad level seven or whatever,” we're like, I think you can probably do most CS. So it's often not a lack of intelligence is the reason why we're not a hundred percent… John It’s because they’re too distracted by Ani. Des Exactly. A lot of the wins come from, honestly, better architecture, better tailored models—or changing into UI can change exactly how things work. And then sometimes you will get an occasional bump here and there from a model swap. John (00:23:12): It strikes me that a lot of the, how you include the account context and the amount of account context you include is a big part of the secret sauce. Des (00:23:23): Perhaps, but it varies customer to customer. It really is one of these areas where it's a thousand lead bullets. It's not like a single silver one. It's not, if you look at a resolutionary graph, there’s no pop, give or take one or two little tweaks. It's mostly just, “Hey, we ground out through optimizing this prompt and changing this handover, we ground out another 0.7%,” and you see the AI team celebrate that on the balcony on the Friday being like, yay, 0.7 up, whatever. It's hard to work out exactly what bits and then there's obviously multiplicative benefits. You might have a win over here, and it cost you something over here as well. John (00:23:56): How deep down the stack will you go Intercom's version of Apple Silicon? Des (00:24:00): That I don't dunno for sure. We're going to chase any edge we can get. Right now I think custom models is definitely where we're going, and that's a large investment from the AI group, which is the most contested resource that we have. Every bit of work they're doing is finding a new edge in resolution rate or resolution quality. So right now it's paying out pretty well. So I think we're going to kind of place all our chips there until something changes. John (00:24:22): What does selling AI look like? Des (00:24:24): It's quite difficult. It's difficult in marketing and selling because I think… John …because it's so crowded and noisy? Des Well there's that, but it's also—it used to be the case, and for sure Intercom used to be one of these companies, where we our product looked the nicest. So all’s we had to do was to age old, blah, blah, blah, reinvented and then a big sexy screenshot. And you can still get away with that in certain domains. Linear can get away with that, because their product is the sexiest, right? I think with AI, everyone's chat bots look the same. Everyone's kind of copied our messenger. Everyone's kind of roughly converging on the certain UI paradigm and so you have to ask, then, when we say, “We are the best AI agent,” what do you think all the rest of 'em are saying, “We're the worst?” No, of course. So they're all saying this and then everyone has the same screenshots because it’s like, “Look what we do inside the chat window.” So you're like, all right, how do you actually outmarket and then how do you outsell? And one of the reasons we launched the Fin guarantee—this idea that we'll pay you a million dollars if you find some new outperforms us—was because we're trying to stress to the market this idea that we actually believe in our product to a ludicrous degree, such that you should engage with us on any sort of bake off you're doing. But I think from a marketing perspective, it's really hard to stand out. So all that you can really do is rely on backing up your claims as hard as you can, and obviously customer testimonials. Selling is harder. I think again, in the olden days—like selling SaaS in the olden days, meaning pre-2022—it was kind of like, “Our UI is nice, theirs is ugly. Here's a feature grid checkbox. We've got 24 checks, they've got 17 D7 matter. We're in.” And obviously I'm skipping over seven steps, of course; sales enablement would kill me. But you get the basic idea, right? And I think now selling AI is closer to selling infrastructure in a sense. It's more like … John (00:26:03): … our cloud is better than their cloud. Des (00:26:03): Yeah, and our performance criteria are better. It's like at times it might feel like Intel/AMD, or at times it's like it's our response times versus theirs or whatever, but you're ultimately coming into it with a battle of metrics and stuff, like our resolution rate and our CSAT versus theirs. But then people say, well, why? And then you have to then explain what's actually happening beneath the surface a little bit so that people can actually get a bit of conviction other than just trust us or please just go and try our product. Not that easy to try Fin. You have to still have to turn a lot of keys and open a lot of APIs and stuff. So the challenge genuinely becomes how do you have a sales team that's actually able to speak with a good degree of familiarity about AI? John (00:26:43): It's funny you mentioned this. We have this specific problem at Stripe, which is invariably when people switch to Stripe from a legacy processor, they see a significant revenue uplift and you think businesses are in the business of finding ways to get more revenue. You’d think they'd be really interested in this. And we have this thing that sounds shockingly good, which is if you just move over to Stripe, you start immediately getting more revenue. And it basically comes from two places. One is conversion on the actual point of payment—that if your mobile app or if your web flow is kind of janky or doesn't offer the customer's preferred payment method or something, they will abandon. And if you just look at the abandonment rates, if you're seeing a only 85% conversion rate on that form, then obviously getting it up to 90%, that's a huge deal. And those would be very high numbers. Most businesses would not see anything close to a 90% conversion rate on that form. And so there's huge improvement possible there to make the customer checkout experience as smooth as possible. And obviously things like Link then where you're not asking people to reenter their payment details, that delivers a big uplift. The second one is even crazier, because after people enter their credit card details, frequently charges are denied kind of spuriously. And so your bank thinks that it's fraudulent because they haven't heard of this merchant or whatever, and so they'll deny it or they'll think it's fraud or whatever like that. And we—through many, many years of optimization—have gotten good at ensuring that if it is a valid transaction that that is not denied. So what all that adds up to is that we can make the claim and we've seen it play out again and again—we just have Hertz switch for all their e-commerce payments to Stripe—that when people move to Stripe, they see a significant uplift in revenue. That's surprisingly hard to sell because everyone is out there saying, we are the thing that gives you more revenue. We've got the exact same thing where despite you having all the numbers in all the case studies in the world, it's hard to sell because it's… Des (00:28:35): …hard to believe. John (00:28:36): It's undifferentiated as a message. Des (00:28:37): I remember even when we switched back to Stripe, either you or Patrick was saying, “Oh, well don't forget to do Link.” And I was like, really? I was like, is this really a thing that people have? Some business has forgotten its credit card or something like that and you're going to be able to renew it. John (00:28:50): It feels unlikely. Des (00:28:50): It just feels implausible, but at the same time, the data is not really debatable. I think people like to be able to explain it to themselves, and I was talking to a guy who runs procurement and he was saying, “You have to understand I'm just getting bullshitted to all day. It's just all day relentless, absolute nonsense in my face. So if you think that you're like, ‘Ooh, 65% thing is going to stick,’ it's not. I divide it by 10 at this stage and you tell me, ‘Oh, you're going to save me 2 million in CS salaries,’ or whatever. I'm like, yeah, maybe in three years time we'll see 200 grand.” That's the kind of default posture for a lot of these people. I think it is just, it's like they've developed quite an adverse reaction to marketing. John (00:29:31): People do not have enough empathy for the procurement person who just has to endure nonstop nonsense… Des (00:29:36): Absolute garbage. John (00:29:39): That's funny. I mean this kind of gets to a topic you and I have discussed a lot, which is product marketing. How do you effectively product-market in a world of everyone making claims? One is the guarantee your guys' million dollar guarantee—has that worked? Des (00:29:53): Certainly it's worked from a point of view of, I dunno actually know how many people are in the program right now, but I could say that what has worked is… John (00:29:59): It landed the message of, “We stand behind the product.” Des (00:30:00): Being able to say, here is the reason why you can bank on us. I think that's a strong message. I mean obviously being able to point to real customers or real results and you can say, “Hey, go talk to Natalie at Nuuly,” or whatever company you want. Go talk to that person and ask them, that's their name, that's their job title. They work there, they're saying this. With 6,000 customers, it gets more believable as the numbers kind of tick up. So I guess either you can make crazy guarantee claims, you can just point to a lot of successful customers. For us, it might be different depending on your domain, but for us it's not like we can show you—we can show you, hey, here's a beautiful backend product, here's fancy reporting and all that, but that doesn't speak to—courting someone's buying, when they're buying AI off you is to some degree a replacement of work that they have to do. And the two things they care about are how much work are you going to do for me and how well are you going to do that work? And you basically need to product-market both of those things. And it's very easy to say, we're going to do all the work and we're doing really well. So you have to actually really help them understand how to appraise this scenario. Sometimes we put time into actually helping people identify when they're being lied to in a sense. We will say, like, “Hey, try this type of question” or “Ask them about this help.” You almost teach 'em to be much more conscientious buyers. Because we know the more informed the buyer—that suits us, it doesn't suit people who are just jazz-handsing their way to an AI product. But yeah, it's a difficult one. Ultimately, five years ago I would've been like, “Well, the trick John is GIFs. Have you ever considered using movies on your homepage that's really engaging.” None of that works anymore, right? I just think what you're selling is, basically, it's like an iceberg. You're selling this little bit of upfront UI here's what actually happens for your customers. Doesn't that look nice? And you're selling this gargantuan pile of work beneath the surface that is like, hey, all of the human toil goes away if you make the switch. John (00:31:47): Yeah, I can see that. What are your other pet peeves when it comes to product marketing? Actually, do you want another pint? Des (00:31:52): Yeah, sure, sure, sure. John (00:31:58): See again this one, did I not let it settle for long enough? And then you have the small head? Is this settling? Des I think your one's going to work out perfectly John Is the settling time load bearing? Okay. So what are your pet peeves when it comes to product marketing? Des (00:32:08): I think the thing that still kills me, and it's still very common, is marketers that love marketing. So rather than actually saying anything useful or specific, you'll get, like, “Forget everything about email.” You're like, okay, what am I buying? Or, like, “transformation reinvented.” And you're like, “Cool, sounds like I'm going to reinvent some transformers.” But what's actually happening here? I think there is a general still type of thing where my screensaver on my laptop is literally a typewriter where someone said, what are you actually trying to say? And I keep that there as a reminder of just nine times out of 10, the best marketing comes from just writing the thing you want to say. Because I do know how docs—do you ever get into a Google document? “Our goal is that by reading this document, the reader will know the following.” And I'm like, cool, can we just say that and say that? Why are there 2000 more words here? So I think, I guess speaking in a way that sounds great to marketers is probably the thing that kills me most because they don't market, especially in the AI era, they don't really necessarily understand the depth of what's actually happening with the AI or whatever. There's a funny stat that Ogilvy used to quote, which would say something like, of all the winners of the Cannes awards every year something like two thirds of them would lose their contract that year because the thing they won the award for, it was not actually effective in market at all. And I think there's such a repetitive pattern there where a lot of people will look at say a Stripe or a Linear and they'll be like, “All right, we should just do that.” And I think the thing that people don't get is everything means something. And this is where Eoghan, our CEO, is so differentiate. It's just like every single decision we pick here—what photo, what icon, what typeface, whatever—it all sends a message. Are we conventional or not? Are we futuristic or not? And I think whenever I see folks just copy paste somebody else's branding—and even in the sort of, oh, we will change our homework along the way—I think what they're really doing is saying, “We don't really understand what we're doing here.” That's why a classic of this is whenever an incubator spits a new batch of startups and they all basically have right-hand side screenshot, left-hand side, three bullets, sign up button, whatever—you're kind of like, okay, cool, but have you actually thought about what you're trying to say to the world? John (00:34:07): There's also a thing for startups where they probably shouldn't look at what established companies are doing because—like, Stripe for so long, we clung to making sure that we had code in the homepage and people are like, if you want to accept credit card payments for your website, we are the place to come. And at a certain point, most of the relevant people who come to your page actually know that you could do that, you can experiment a little bit more. And you know, Salesforce doesn't have to hit CRM so hard on the homepage. After 20 years, they've earned the right to talk about Einstein a bit. Des (00:34:33): Oftentimes a startup has a great idea for a great product and they pitch it, and then six months later they work on some new feature and in their heads, the new feature is the big thing that they're so impressed with, not realizing that 99.9999 of the world have not even heard about the original thing yet. But there they go destroying their original pitch by being like, “And now we've got blah, blah.” I'm like, dude, no one's even heard of the original thing yet. And here you're pitching some nuanced take on some extra feature. John (00:34:57): Speaking of David Ogilvy, you've read On Advertising, I assume. It was just, like, that's such a beautiful book where all the marketing copy in it is so good. The Rolls-Royce side of,” The only sound you'll hear at 60 miles an hour is the ticking of the clock.” But somehow that should be mandatory reading for all product marketers. Des (00:35:14): Absolutely. John (00:35:15): What kind of person succeeds in product at Intercom? Des (00:35:18): When we set up Intercom originally we were building one thing. Once we forked out to building many different areas—like sales, marketing, and support—I think we gave a lot of freedom to product leaders and sort of said, “Hey, you owned a sales product, you're marketing product.” And I think the folks who succeed there are like, they have to have decent taste, and I don't mean that in some lofty abstract way, but I mean they have to use good software, identify good software, ultimately know how to pick one out of the bunch in a sense, right? A very common interview question I ask people is what apps are on your phone? What's your favorite app? And you might have thought someone's like, “Oh, I never really thought about that.” And I'm like, oh, so what's your favorite song? I am sure you care about some things. John (00:35:57): Certainly if you're interviewing musicians, they should have a favorite song. Des (00:35:59): Yeah, you'd like to think that, but they also would be a favorite app too. So I think taste is a kind of prerequisite. And then I just think the confidence to pick a direction and then we say at Intercom, we often to say shipping is an act of confidence and humility. What that means is you have to be confident enough to put it alive and then humble enough to take the slap in the face when you've got it wrong. Totally. And react to that slap. Don't be like, “No, it's not me, it's the customers don't get it.” So I think we need high taste and then confidence and then ultimately understanding that in the market and recent sense, a product is a conversation with the market. Your launch is your opening bit, and then you have to basically adapt and react to what gets thrown back at you, which might drag you in different directions. And then you need to have, again, the confidence to prune certain things. And no, we're not building an attribution engine. Yes, we'll take on some feedback in the CRM side. But I think a lot of product managers who don't work out for us are a lot more spreadsheety and they won't take a bet. They won't take a gamble, they won't take a stand. They'll just be forever mired in like, “Well, the data suggests,” and they're just trying to hedge their bet, whereas it's just not the sort of company we are. I think we kind of believe in having an opinion about a space. John (00:37:07): Well, and the second part of what you're saying, if I'm hearing you right, is that good product managers actually can listen to the market. Des (00:37:16): They have to be able to, yeah, John (00:37:17): And hear what—I think about this a lot in the context of tech companies where—Stripe’s first operating principle is users first. We think that actually paying attention to what users tell us—tell us in every sense, via revealed preferences in the data, via just, like, when we actually have conversations with them. We start every week with a Monday morning meeting. The first thing we do is—we actually host in Intercom—we host a customer to tell us and give us a report card. And it's not an A, it's seldom an A. They always have things they want to fix, and they're very pragmatic things that they want us to improve. There's no overcomplicating. And then when we do our weekly all hands fireside, we also bring customers to that. But I feel like there's a problem of over-complexifying things and under-talking to users in Silicon Valley where, yeah, it's a bit too much celebration of the individual kind of product vision or a bit too much, as you say, trying to data your way out of it. And if you're a product manager and you're not talking to many customers each week, something's probably wrong. I bring that up because the whole original Intercom product was a way to talk to customers. This is kind of your guys’—but would you agree that diagnosis, that a lot of tech products would be better if people simply talk to customers more? Des (00:38:41): Yeah, I mean one of the ideas that stuck with me very early on—it was like 2009, 2010, I'm going to a deep cut or whatever—but there's a guy called Jared Spool who's a famous UX guy, and I was on this thread of interaction design association type people, and somebody wrote this really long like, “Hey, I've shipped X and I've shipped Y and I can't work this out. Does anyone have any speculation as to why people aren't doing this thing even though I make it really obvious on the screen?” And he just replied on and he was like, have you tried asking them? And I remember at the time I was like, right on. It felt like a revolutionary thing to say, but I find I shared this piece a while ago, which was the questions I ask in every single product review, so you can either get ready to meet me or just ideally other people can replicate. But question one is basically, “What did our users say about this when you showed it to them?” John (00:39:27): Yes. Des (00:39:28): Everyone has to have an answer to that question. When we go in, I'm like, “Hey, well what did the user say?” And I need to understand that. If you're not actually asking your users, what are you doing? The only validation we have is the market. I do think in the valley—well, I'd say the valley, but that just basically means in the tech industry opinion—there is this epidemic of highly mind your data, and what can we instrument and how many different Mixpanel dashboards can prove to me that this product should be working, just ignore the fact that it isn't or whatever. I think there's something kind of just fundamentally broken there. And interesting—you say Stripe’s value is users first. I'm just curious, does that deliberately—I mean, every word you guys say is deliberate—but that's deliberately “users” as in do you mean pointy-clicky users or do you mean customers or do you mean prospects or do you mean… John (00:40:11): Yeah, we deliberately chose users because we just meant the people using the product as opposed to customers if there's a buyer versus users. Des (00:40:21): Yeah, that's what I was going to ask. John (00:40:22): We want to focus on the people who are actually using the product, the people who are managing fraud within the business or actually responsible for increasing conversion or something like that. Des (00:40:29): Yeah, yeah. John (00:40:30): So, that was why we chose that. Des (00:40:31): Word. Yeah, no… John It's not perfect. Des It makes sense because I find oftentimes if you want to perfect a product, you talk to the users. If you want to expand your market, you talk to prospective buyers. But whenever I, even in my own portfolio, when I talk—what are users actually doing? The businesses that are, how would you say, prematurely talking to more prospects when they have a load of unhappy users are guaranteed this kind of miles wide, inches deep, messy product that doesn't actually satisfy anyone, but they'll get their one promise at a time, “Oh, we'll build that for Johnny and Johnny will sign, and we’ll build this for Jenny, and Jenny will sign,” and at no point do they have one happy customer. What they have is a marauding churn bomb of a user base. John (00:41:10): Yes, it is funny you say that. It feels like many tech companies over-rotate on sales feedback, which will by definition be from the marginal user and they're marginal in two senses. So you have all your existing users that—you're dancing with the girl that brung you over here—and then you have this future potential user who—firstly by virtue of the fact they're not already using you, maybe they're slightly outside your wheelhouse or the use case isn't perfect or something like that. So maybe they're not quite as good a fit as your existing customers. And then also by virtue of the fact that they have a whole existing way of doing things when they migrate over to your product, they'll do so in a bit worse shape of integration, where maybe they'll only use one of the four features, or not everyone in the org will be bought in. Versus the people who grew up on your product. And so maybe just restating what you're saying, I'm always struck by people are way too focused on, “We tried to win this big new shiny enterprise account and we didn't have feature X, and so therefore we're going to develop feature X,” as opposed to you've all these users who grew up in your products and really like it, but they wish you had fixed A, B, and C. And just the nature of the fact that sales gets more airtime than account management, essentially, means people really misprioritized where they spend that. Des (00:42:28): And people take NRR for granted and think that net new revenue is hard. And I think one of the things that we see a lot of is in terms of working out for your current customers, we use the phrase “permission to innovate” and “permission to expand” in Intercom, which is basically you have permission to innovate when your product's pretty good. Like, let's work on v3, but is v2 in good condition? And then permission to expand is like v3 isn't actually that exciting. Everyone's happy with v2, now I think we can try to do something new for customers, like expand our share of wallet or whatever. But I think a lot of people try to solve revenue growth with, like, aimless product expansion to just try and increase the share of wallet for the people who are stuck with you, and then they convince themselves they got PMF or that they have some sort of a good product because they're kind of like foie-gras style force feeding new features down to throats of their trapped users, and they're like, “We're doing great.” But they don't realize what they're actually doing is making their current product so messy that they're destroying the hope of future revenue. You can force your current customers into whatever upsells you have or whatever, but your product marketing along the way is getting really difficult because all these features don't make sense. And they're just like, you've tried to do this land and expanding, but you're actually just expanding and there’s no landing happening in the new product. And then you end up trying to twist yourselves in knots. And a lot of startups, Jason Lemkin said this thing: from zero to one is impossible, from one to 10 is hard, and from 10 to a hundred is inevitable. I think a lot of people, I don't think that's proven out to be true as much as it was back when he said it. I think a lot of people get stuck in some sort of glue around—somewhere around a 10 million mark—where they don't know how to get the next 10,000 logos. So they just try and milk the revenue out of the existing customers. True, just forced product adoption of new stuff. You see a lot, “Here's your copilot, I know you didn't want it, but here you go.” That type of thing. John (00:44:18): Des is describing here how they've transformed Intercom from a SaaS product to a frontier AI business. And to do so, they had to pivot not just the product but the monetization model as well. Because inference costs are so significant, AI-powered companies tend to charge based on usage rather than just allowing for unlimited plans. It gets complicated and really multidimensional very quickly. Now, fortunately, complicated and multidimensional is what Stripe Billing specializes in. Our usage-based billing engine can ingest up to a hundred thousand events a second—a hundred thousand events a second! So AI companies can monetize products based on real-time customer usage. We're powering consumption billing for companies like Figma, Cognition and tons of other leading AI applications. Our usage-based billing platform has grown 145% so far this year. So whether you're changing your business model like Intercom or starting a new product from scratch, your business strategy should dictate the billing system and not the other way around. For usage-based billing, check out Stripe Billing. As you think about the prototypical $10 million revenue B2B company, what are the common mistakes you see and what do you think the actual path that more them should follow is? Des (00:45:27): I mean, the biggest problem, mistake is not aligning your fundraising with your TAM. I think a lot of folks, we got a little bit over-convinced during the era of cloud that every business had a right to be like a unicorn. And so there's a lot of businesses whose idea was totally fine, but actually they should have gone and basecamped it more so than they did because they've raised on the assumption there's an easy path to hundreds of millions in revenue. John (00:45:53): It should be more small, profitable, $30 billion revenue companies. Des (00:45:56): Well, yeah, exactly. And I think a lot of these businesses would be great if only they didn't raise 20 and tell their investors that “you're going to easily be worth a billion” or whatever. So I think there's a genuine mismatch there where I think people have overstated how big this idea could get, as in, “Hey, I know all we do is time tracking for dentists in Delaware, but believe me, we're going to be a billion dollar company.” And you're like, okay, well one of your restrictions is going to have to break here. So that's one problem, which is more like business model and venture ambition. The other stuff I see is, it is not focusing enough on the thing the majority of your customer's value. It's easy to say—the best business in the world is one line of code that all users execute and you sell it to all users, right? They're like the sweet spot. It's hard to do in a differentiated way because obviously people learn that line of code. And that's where I think a lot of these horizontal products, they say something like a Loom or whatever: they're brilliant, they're a piece in everyone's workflow, but they're no one's end-to-end workflow. I think they can do well too. But I think the challenge is when people, rather than nailing a specific small thing, going back to the earlier point—rather than saying, “Hey, let's get really go to X before we go beyond”—when they prematurely expand, I think they forego all opportunity of being the best. And if they picked a really important area first, they don't say it outloud, but what they're saying is it's okay to not be the best at the most important thing we do. And I remember, I think it was 2012, I was in your office in FiDi, and I remember at the time it wasn't obvious to me that you weren't going to expand and do some sort of peer-to-peer and transfers and compete with PayPal. And I remember you guys had the discipline, you're like, absolutely not. We care about helping businesses charge. And there's a real harsh discipline you need to have to just basically say no to all of the surrounding opportunities. And I think for a lot of people, that discipline's the first thing to go. When you hear about competitors, you're going to hear somebody else encroaching on your space. You're going to have this really weird broad view of all the things you do. “I know we just do whatever it is, GIFs and screenshots. But actually when you think about we're a global creativity platform.” And they have this premature view of themselves as being massive and then they feel like then they go and raise off that and they need to expand into that. But I think at the core of every great business—every great SaaS business, but in the future, AI business—is something that they're just truly world-class at. And it's not some sort of 80/20 trade off. They've just basically said, we'll be better than anyone at this, right? You take Linear. It's basically they have literally the world's most efficient UI for product management and they have project management and they've just gone really deep into all of the surrounding adjacencies you would need to actually do that job really well. Figma is just an amazing creative collaboration machine. Everyone who has done really well, they've picked one thing and just gone really hard, really deep, really far on it. They haven't prematurely blown up and gone in seven different directions. John (00:48:46): And I also think there's a weird celebration in the valley of Act 2. The Valley is obsessed with finding second acts that are totally unrelated to the first business, the number of people who bring up, “Oh, and we invent an AWS.” It's like, okay, you need to use a noncliche example if you’re going to make that argument. And the flip side is you're mentioning Figma, which I think is a great example where that market proved to be way bigger than people might originally have thought. My favorite example of this is NVIDIA, where the world's largest companies, and they started in the 1990s making GPUs. If you're an investment banker trying to make a case for how NVIDIA can be a really big company, maybe you'd say, “Oh, well, and we can expand into maybe we'll actually make our own gaming rigs,” or—because it was all gaming at the time originally—“maybe we'll make gaming consoles or expand to some larger markets.” Whereas actually what transpired is, it turns out the GPU market is quite a lot bigger than people thought. And being the best at GPUs is a really valuable prize and you can't rush it. It'll emerge. But yeah, Des (00:49:55): And they could have killed themselves if they had gone in every other direction and they would've lost their edge in some sense. Figma is a great example for me of permission to expand in that, they literally nailed to a point of no credible competition, this idea of just the Photoshop killer basically, let's just say. And now they can talk about slides and text-to-app builders and every other dimension they want to go and everyone's like, “Yep, that's great.” Because you guys make great software. I think you have to first be known for, I'm trying to think—if Stripe launched a payroll product, it would carry the brand of Stripe in the sense of being, well, it's probably really good, really reliable, really fast. It probably has really nice APIs, it probably works really well, Workday, blah. You can almost impute all the ideas that would be carried into it. And I just think you have to get to that point before you have permission to make that bet. Obviously, it's a lot easier, if it's like stablecoin or whatever, but what kills me is when—I don't even want to name a weak SaaS company—but pick your favorite mediocre SaaS company and anything: is there any direction you would allow them to expand in your head? No, it's the short answer. John (00:50:56): Yeah, that's interesting: who are you really excited to adopt new products for versus who are you steering clear of the new products? Des (00:51:01): Like if Linear launched a, I don’t know, let's just say a source control tool, like, yeah, it's probably going to be really, really good. I remember Seth Godin has this hilarious point where he talks about the value of brand once it's weaponized, and he describes Nike and Hyatt hotels, and he says, “If Nike opened a hotel, you can close your eyes and see it. You know exactly what the corridors are going to look like, the vibe of the whole place, you know everything it's going to be. If Hyatt launched a sneaker, you're like, ‘What?’” And it's just, that's the difference, because Hyatt has a logo and Nike is a brand, and that's the difference. John (00:51:34): A version of this actually, maybe quite literally, is I know I saw Equinox launched a hotel—which is a pretty good idea because, the design center there for the hotel is, you just want to be able to get a good night's sleep. And it's funny how that's like a differentiated product in the hotel space of, “Don't give me any of that other shite. I just want to be able to go to my room and not have a super loud noise outside the window or weird light coming into the room.” You just want to be able to get some sleep. Des (00:52:01): Precisely. Yeah. John (00:52:02): Yeah. I thought that was funny. And you're mentioning Stripe’s expansions, and so this may be a good segue into your pricing model change. You guys are the poster child for the move from per-seat SaaS pricing—the old way of doing things—to usage-based pricing. Maybe you can describe a little bit about that and then how you implemented it and what you're doing with Stripe. Des (00:52:25): Yeah, sure. Our pricing journey is long and complex, and a lot of your listeners or viewers will know— John (00:52:32): Intercom pricing is a charged topic at Intercom. Des (00:52:35): Not anymore. We've turned a corner. Let me just back up a bit. So when we had too diverse a product strategy, we were trying to do sales software, marketing software, support software, and sales software is typically sold based on leads, creative and marketing was charged by how many contacted people you want to send, and support was sell by seats. So we had this extremely, let’s just say, detailed but unnecessarily complex pricing setup. And we lied to ourselves and said, “Don't worry because there's always going to be a human to help people navigate this, because you're never going to have to self-serve this.” But ultimately, people were just like, “I have been refreshing this for seven minutes and I can't understand a word of it.” And that was just one of the few things we got wrong in our first move up market. When Eoghan returned, one of the decisions he made was just like, “Hey, we need to sort out pricing.” And we handed back, truly handed back, I think about $50 million of revenue. I think— John (00:53:22): Was that controversial, like with the board, with investors? Des (00:53:23): We had support for it. I think it was like, people don't really—people massively underestimate what it means to have a really happy customer base. It’s because word of mouth doesn't have an attribution or a UTM code, if you know what I mean. So they don't understand how to think about happy customers. So making the decision to basically standardize on an easy-to-understand pricing that's fair, transparent, predictable, et cetera—that was the first decision that we made. This was before AI, right, and that was us returning to Stripe, was a large part of that. In fact, as a small segue, I think a couple years prior, I had said to you or to Patrick, “Hey, you guys should actually build as part of your product offering a pricing page creator.” And I think at the time though, I probably got one of those thumbs up replies or something like that. Was like, “Yeah, whatever Des.” John (00:54:13): It's on the list. Des (00:54:15): I think you've done it since though. John (00:54:16): Yeah! Des (00:54:17): But my thinking at the time was, basically some version of this: you need to not let your customers go wild with pricing. You need to actually put some sort of guardrails onto how they think about pricing. Otherwise, they're going to go and invent stuff that you guys don't support and then you're going to move all your business logic across— John (00:54:31): Pricing is writing checks. Des (00:54:32): Yes. Yeah, exactly. And I remember sitting in a Stripe mini all hands, or whatever, explaining that, “Hey, now none of our business logic runs through Stripe.” And either you or Patrick was saying some version of, “That's not a good thing.” I think, generally speaking, my advice to any software company is, “Don't afford yourself too many degrees of freedom here. You'll actually cripple yourselves in a quagmire of complexity that'll take you many years and ultimately many tens of millions of dollars to get out of.” John (00:54:56): It's a weird failure mode that every single company falls into, which is, you start signing deals that have some super creative pricing structure and the customer negotiates A, B, C, D and E. And it's like, it is built in Microsoft Word, but it is actually—it’s just not built practically in code because it just exists with this one customer. And it may not even be possible to build in codes. Sometimes it's kind of ambiguous. And if in the subsequent year this happens, then we go back to the prior year and we do an adjustment— Des (00:55:25): I call Rick and he decides what the discount is— John (00:55:27): There's like a time travel component to the whole thing. And then they obviously have this, again, we see all customers running into it, these kind of manual billing issues where there is a guy who has to deal with all these contracts that were agreed during the sales process. And so as you're saying, an opinionated billing engine is actually pretty important, assuming you believe that billing should be automated. If you're happy manually getting out the calculator for every single customer month, then that's fine. Des (00:55:54): And probably got a large deal desk function and doing all the work behind the scenes. So anyway, that was the first piece of our pricing. And then the second piece was obviously when we launched Fin, then it was like, “Hey, how do we charge for this,” because we're replacing seats. And at the time—it has improved out this way fully—but at the time Fin looked like it was going to be pretty cannibalistic to Intercom. It was like, “Hey, if we're automating—at the time, what we thought was 25% of your revenue—we assume that means 25% less seats in the future.” Or at the very least what it would likely mean is the growth rate or the NRR of the seats model will be affected by the fact that Fin’s doing all the work, and now at 65%, you'd expect it to be even further true. So it was like, “Hey, how do we charge in a way that makes sense?” And then also how do we be aggressive? As in, we really wanted to put a mark on the market that sort of says we're very AI forward. I think what Eoghan and then Darragh came up with was just like, “Hey, let's just charge per literal resolution, like every time we do work, we charge them. When we don't, we don't.” And this is at a time when AI was margin negative and all that sort of stuff, we were still working out how that whole world plays out. Today, we're really happy with our margins, but at the time, it was like, “Hey, this could— John (00:56:58): It was a bet. Des (00:56:59): Yes, it was definitely a bet. We made the decision, and I think the market responded really well, because I think it was very clear that the single statement of, “We only get paid when we do work; we don't get paid when we don't do work,” it's from the same vein as the guarantee—which is just like, that's how you know that we believe in our product and that's how our product works. And it's been copied a million times since. But I think the actual decisions that we made in the run-up there were really, really important from a point of view of backing up our claims. And then obviously for a lot of our competitors, they were like, “We don't really have a great way to respond to that, because either our product doesn't work or we're kind of hooked on these really expensive resolutions.” And we were totally the cat amongst the pigeons there, which has been really well received by the market and our customers generally do love it. It is funny though, you still get people being like 99¢ is ridiculously expensive and you're like, “Why do you think this?” And the answer is because, some version of, we don't know how to calculate COGS. John (00:57:52): Yeah, how much are the humans costing you? Des (00:57:55): Yeah, exactly. And how much is their office costing you and all the other stuff. John (00:57:58): But people like certainty. How do you get them okay with the variable component? Des (00:58:03): Obviously you can contract out whatever you want, but what we offered people is, hey, most of the time people have at least one or two years look back on, with customers who spike for tax season or customers who spike for Christmas or whatever, and we can basically say, “Hey, let's contract your base rate and let's talk about overages for the months you need it.” And that totally works. What we're basically saying is, “Yes, you don't have predictability in the sense of it not being fixed, but you can model it based on what's happened in your history,” and it's only really brand new startups that don't have a clue what's going to happen, but they're not usually worried about this. John (00:58:37): So you're using a relatively new product, Stripe usage-based billing, for this. How has that—you migrated from Zuora for that. How has that process been? Des (00:58:44): Yeah, I mean, I would say just to go back to your earlier, we afforded ourselves too much complexity. We kind of codified that complexity in Zuora. I guess the best way to describe it is we just twisted ourselves in knots, and it got to a place where we actually, we ended up like Ciaran Lee, who was our CTO here, he ended up actually returning to the company with one mission, which was like, “I am going to, one, John (00:59:08): Fix billing. Des (00:59:08): Yeah, exactly. To fix this, right? And it worked, right, but it was a substantial amount of work to unwind so much and then to deconstruct so many of these à la carte Microsoft Word style deals into something that was a go forward, acceptable whatever, and then obviously moving towards a clean, transparent seat-based pricing, and then just layering on a usage base on top was actually pretty simple in the greater scheme of things. And all the stuff that we needed, you guys were ahead of us on discounts for volume, et cetera, all the sort of obvious stuff people would push for. John (00:59:41): Yeah. Is this just where pricing in this new world goes? Because obviously no one buys labor on an unlimited basis, and at least for the moment, the inputs of AI do actually scale with usage for a significant basis. And so it feels like you have to have some usage-based pricing. This is certainly the bet we are making where again, the reason that billing, the top thing they're thinking about is making billing work well in the usage-based world is it just feels like many products are becoming much more expensive to serve and therefore have to have a usage-based component. But is this permanent or I dunno, does the AI get cheap enough that maybe we go back to unlimited plans or Des (01:00:22): I dunno, I dunno if unlimited plans will ever, well, I dunno. Here's how I think about it. I think ultimately all AI has two vectors, there’s how much work you are doing and how well are you doing it? John (01:00:32): Yes. Des (01:00:32): And the volume of work you're doing, it's almost, well actually both of 'em are going to be proportional to how many tokens you're burning or whatever. So you're going to want to factor that in, especially if you're a consumer app as well, we're going to go nuts. So I think you have to have some, I'm not a fan of cost plus pricing, but it does place a kind of lower bound on what you can do here, which is just like, hey, unlike SaaS, you are actually sending money out the back door as well. So I think you have to have something that's proportionate to how much work you're doing. And then I think aside from that, you have to charge consistent with how much work are you displacing. I think that's where you can say, hey, for us anyway, if you take an average person who sits in a seat to do customer service, if they do, let's just say they do 20 conversations a day, that's what 400 conversations a month, when we were thinking about how we charge, we're like, hey, well if that person does 400 a month, Fin does 65% of that seat, we're still up. We're only charging, whatever, $90 for the seat. So from our point of view, it was an obvious, an easy swap. I think for a lot of businesses it might not be if your AI doesn't work or it's spurious or its value can't be articulated. Isn't it cool that you can now dynamically summarize a GitHub issue or something like that? You're like, cool, I don't know how much people will pay for that. They don't know either. Or, hey, you can now generate random graphics in your newsletter tool. You're like, okay… John (01:01:50): It's like vitamins versus painkillers, AI pricing. Des (01:01:54): And it's specifically in this case, the painkillers are very strict, if we don't do it, a human's going to do it and we know exactly what they cost, and the vitamins don't have anything approximating that. So not only is it nice to have, it's like I don't even know what it's worth. I saw a while ago someone said when Studio Ghibli came out and everyone was using that, someone said, hey, the fiverr.com equivalent of all these things would've been trillions of dollars. You're like, right, but no one was ever going to spend that. So there's no sane way to actually talk about what actually happens here. I think it was Byrne Hobart who said that when you're tied to business outcome, that business outcome is usually done by humans. I think it's going to be really, really easy to make a business case for saying swap this over to AI. It's better, faster, cheaper. I think when your AI is not tied to business impact or is debatable in quality or whatever, I think you end up with these people who are just like, oh, let's just stick a tenner on the seat and see what happens. So it's like you can have a normal seat or an AI seat and then you're kind of like, I hope no one uses the AI too much. You're permitting yourself to build weak AI stuff if you do that because you're not pushing yourselves to say, hey, we need to articulate the value of each incremental usage here. John (01:02:57): Well, when you talk about this AI pricing dynamic, one thing that really strikes me is just how fast AI companies grow from a revenue perspective. So I just saw Mati from ElevenLabs. We actually had a great session at our customer event in London, but he tweeted that they've just passed 200 million in ARR and that's 2 years after founding it, maybe 3 years after founding. But in my day, businesses didn't do that. And it's really striking for me how somehow they seem to climb the revenue ramps much quicker. Des (01:03:28): I know, I mean… John (01:03:28): You guys would Fin is another example… Des (01:03:29): Yeah, for sure. We forecast like Fin will be 100 million probably early next year or whatever and back at… John (01:03:37): Yeah, from when? Starting from… Des (01:03:38): I dunno, probably about two years, something like that. Yeah. John (01:03:41): So two years to 100million in ARR… Des (01:03:43): When we started and probably when you guys started it was like that was the threshold to go public. John (01:03:48): Exactly. It used to take a long time to get to 100 million in ARR. Des (01:03:50): It was like seven years. John (01:03:48 ): It used to be a lot of money. Des (01:03:49 ): Yeah, exactly. Back in the day. But yeah, the acceleration is more there. Mati's Eleven's a fantastic product and it's a great example of there's four if you like, horsemen of AI products that I observe whenever I'm investing, and it's rare you see all four. But the things you want in an AI startup is one is the revenue backed by usage. And that's why I love usage-based revenue as opposed to shelfware or pilot ware.Hey, we sold it to two guys in the corner and they're going to put it live someday. So you want revenue backed by usage. You want the usage tied to a real business impact. So as you mission-critical, as in if you're building a phone product on top of 11, if that doesn't work, that's really bad. So it's critical. The third one is obviously you want deep AI, deep differentiated AI, it can't be a thin wrapper. And then the fourth one is you actually want positive unit margins and they all just, or at least a clear path to positive unit margins if you're not there already. And I think when you look at so much of the AI landscape, you'll see so few businesses that have all four, it's such a rare sort of air to be in, to be like actually, hey, we're doing a real thing with real differentiated AI. It really matters to businesses and we're making money off it. Most of the time when you hear about these, “we went from zero to six million overnight,” it's kind of like to generate JPEGs of a smurf or whatever and you're like, all right, cool. I'm not sure that's going to renew. John (01:05:09): That's the simplest AI investing framework I've heard. Des (01:05:12): I'll tell you why it's simple because you're going to make me write no checks. So I guess I'd say most of the AI companies I’ve invested in probably three or four. Three of the four I'd say. John (01:05:23): The only one I might quibble with there, I think, that's very good for staying out of trouble. And this is where I tend to push back when people are saying, oh, it's an AI bubble. It's like, I dunno, I think people are happy with the tokens they're buying. I think a lot of tokens are happening and just generally they seem to be delivering useful outcomes… because they're actually delivering value on the customer service side or people enjoy their Midjourney adventure, but people are getting value from the products. Des (01:05:46): So it's a pushback that doesn't… John (01:05:49): I was going to push back on number four, which is positive unit margins because just aren't the underlying costs… Again, when you guys started Fin, it sounds like you were underwater… Des (01:05:58): Yeah, we were losing money on it… John (01:06:00): But then just pretty quickly it right-sizes as you optimize it, and so couldn't one be too focused on the current implementation? Des (01:06:06): Yeah, I mean this is a conversation we have internally with our CFO quite a bit actually because, we’re good, John (01:06:12): I can imagine it’d be the kind of thing a CFO would want to… Hey, Des, do you have five minutes? Des (01:06:16): That's exactly, yeah, “quick chat.” I can't help but notice the team have done this preemptive loading or whatever. It's causing this shitload of money. So what's my counter? I guess I prefer it if the path towards profitability isn't just OpenAI is going to figure this out for me, right. An interesting way I'd say this, with Fin for example, obviously our profit goes up when we are firing less dead tokens. A dead token being we've generated an answer and it wasn't right, so we can't charge money for it. If you're, say, guessing the next line of code or tab to autocomplete the next line of code, if five of six of those is wrong, I don't know if you're ever going to get bailed out because you’re basing five-sixths of your costs is not something you can resell. So there's a questionnaire of how much of your tokens are actually generating a thing that a user wants independent of what you charge for as long as the user wants it. I think you're always in good condition, whereas if you're burning a million tokens to find one, and that one, you're never going to be able to recoup your costs or at least I'd love to see your telemetry to make sure that you actually have thought this true. I suspect you haven't. Des (01:07:25): 45, that’s not bad. John (01:07:26): Yeah, I was very impressed by the multiple, clearly deliberate twenties. I'm curious about the cofounder dynamic you guys have across all the cofounders where, lemme try this out. My sense is that people want to have a subject-matter-area-based explanation for cofounder collaboration where, I'm the technical guy, and they're the business guy, whatever. And in my experience, or at least with me and Patrick, it's much more personality-tension-based where I would say he's more visionary and expansionary, and I'm more, well, we have food at home already. You've got to finish the products that you're already doing or I'm more frugal and he always wants to spend all our money or whatever the tension you're describing is. And then there's a useful, well, one, it's useful to have someone to be able to, you’d go mad by yourself trying to solve all these fairly knotty problems, but also a good company strategy probably exists at the intersection of those tensions. Does that describe your relationship with your cofounders, and what would you describe as the personality tensions? Des (01:08:27): I mean we're definitely all different. A lot of key things we all agree on, Eoghan would be like a, first and foremost, he's a very strong CEO. He is very decisive and he is very brave, is the best way I could describe it. An interesting thing, when he returned to Intercom, one of the things he did was basically rebuild the culture, and one of the things he focused on was resilience and open-mindedness. We didn't know AI was coming. He didn't know AI was coming. But to be able to react to AI requires a lot of manic pivots, zero certainty, and ultimately conviction bets. And I can't think of somebody better to do it. That wouldn't have been me, not in a million years. I would be even being as AI-pilled as I am, I still would've … And I even look back at my own performance in that period and I'm like, you know what? I wasn't brave enough. One of the things I pushed through was the city of creating the team Fin, which is like, “Hey, let's just build a new startup.” Let's isolate them for everyone else. Different floor, different section in the office. No one else is in there, it's just their own Slack channel, their own everything. They're entirely secluded. And how do you not push for that? I don't know if we would've had the clarity and the focus that we needed. John (01:09:32): People might be offended. Des (01:09:33): Yeah, of course. All of the things, all of the downsides you'd possibly guess are all there. I also think that there's no path to … there's no way. The phrase I've settled on when I look back, is, sometimes you have to go too far to know you've gone far enough. And I think a lot of the mistakes I see in people who are trying to adapt to AI, for example—I'll come back to the co-founder here in a sec is they tell themselves that they've done enough because, oh, a few sparkly buttons. The merge features AI and we're happy. John (1:10:01): And we have an AI assistant in the product. Des (1:10:01): Yeah, exactly. And we've updated our homepage and say we're AI first, so we're good. And I think you need to be willing, genuinely willing to make brave hard to undo bets. And I think you need obviously having this sort of moral authority of a founder and being CEO kind of gives you some of that, but still it's a huge decision to make. And I think I am much more of an, my default DNA is I'm more of an operator in the sense of, “All right, what are we doing? OK, well I'll make it work.” Whatever it is. And I think if it was a company of people like me, what you'd see is probably predictable, reliable, sustainable performance or whatever, but probably not enough actual, well, definitely not enough brave big swings, which is actually where you need to get to. But I mean it is a cocktail as in somebody needs to go and actually do the thing, once we've decided what we're doing as well. The way it ended up, I was leading Team Fin after I decided it and that was ultimately what led to the creation of the whole Fin initiative, whatever. John (01:11:00): You've now worked with so many different companies externally, you've seen a lot. What is predictive of success and what is predictive of failure? Des (01:11:08): The biggest thing I'll always come back to when I'm talking to anyone who's trying to pitch me to invest, or me to induce John to invest, is it's always some version of: Do you have a real product that solves a real problem that really exists and people are really already trying to solve by paying money or time somewhere? It sounds so trivial, but you'll be shocked how many times you'll fail or you'll get some sort of jazz hands-type routine somewhere along the way where it's like, “Don't look too much at this, but just trust me.” The areas that I end up being blind to is the extremely market-expanding type things. As in if someone said to you, “Hey, all companies are going to have a chatroom and they're going to all hang out in an all day and have unproductive conversations, it's going to be big.” I'd be like, “Oh, I don't see it.” Whereas that’s how you would've missed out on Slack, or whatever. But I think I can almost hear from the what are you building and why and who's it for, and show me what the product does if it's not a real solution to a real problem, I'm kind of already out. And then the other big, I'd say, prediction is just one of the things that's happened in the last 10 years—I'm sure you've seen this a lot—it was a lot easier to invest when being a founder was uncool. And I think a combination of, I blame genuinely the “Social Network.” I blame just kind of the entrepreneurial lifestyle. I blame TikTok. I blame all these things. John (01:12:21) Soho House. Des (01:12:21) Exactly all of that, right? To some degree, like remote working, into the mix as well. But I think the amount of people who are chasing the trinkets of being a founder of a startup, even if they're quite smart and they can actually kind of go and build something, if their actual motivation isn't the problem or isn't just some deep desire to be quite successful, but it is instead to be perceived the whole, “I could have been a contender. Rather than I could have contended.” If you don't really, really want to actually play the game, instead you just want to be seen to be playing the game. I think that's probably the single biggest thing that tells you you're probably, best-case scenario you'll sell at 5 million. But more likely you'll still be alive in seven years, all your investors will wonder what you're doing, and you'll be basically sending one investor update every now and then. John (01:13:09): Yes, I have noticed investor updates with metrics don't predict success, but investor updates without metrics that tell a really fancy story, but don't have metrics, are actually quite predictive of failure. Those companies always fail. Des (01:13:26): I basically 100% agree. And honestly, you can even tell where the metrics are in the update because oftentimes my favorite updates, I mean this company actually probably should … John (01:13:35): And no o investor updates are fine. There's a bunch of successful companies that just never … We never sent investor updates. I'm sorry for all the investors, but we're bad communicators. But if you go to the trouble of writing an investor update and then make a proactive decision to not say how your business is doing, that suggests some deep denial about what we're running a business means. Des (01:13:57): So there's one company—I can't say—but we're both an investor in it, but their updates, just one of the most recent ones was just like, “Here's performance ARR plus 17%, blah plus this, blah plus that.” Something like, “I hope you can see from the numbers we're doing great. Best of luck to see you, see you next quarter.” And I was like, “Yep, brilliant archive. I'll mark it up.” Something like that. I think in general, the degree to, I think it was Paul Graham said, the ratio of numbers to words is usually the actual thing you're looking for, which is if the numbers speak, then the words don't have to. John (01:14:29): What else is predictive of success? Numbers is one. Des (01:14:33): I almost want to say the inverse of all the things that I hate seeing. I hate seeing founders who invest massively in their personal brand instead of their company brand. I hate seeing people who are obsessed about … if the first three or four updates I get are begs for retweets and tweets and all that sort of stuff, that's always not a great sign because it sort of says to me you haven't worked out how to market or whatever. Anything around what are the customers saying? Whenever I reply and say, “What do customers think of this feature?” They're like, “Oh, we're going to ask them.” John (01:14:58): Maybe what you're describing is there's a very boring playbook, boring and unglamorous playbook, for making products work. About writing code, talking to customers, running that iterative loop, and avoiding distractions and people who seem incurious about that playbook or just are failing to execute on it is kind of a warning sign. Des (01:15:19): Yeah, I would say that's definitely true. So it's basically if you got a decent, decent product, decent area, real solution, and are you just willing to work on the boring stuff that needs doing to actually make that whole thing? And then will you get bored in a year or are you still excited by it? And I think to some degree, even successful founders can get distracted by glamorous opportunities, whether it's like, “Oh, there's a new wave of whatever, like crypto or NFTs, or whatever. You'll see people get their head turned quite a bit and I think if you're genuinely married to the problem and married to the solution, you'll tend to not be as distractible. And then so many of these businesses just need time. They just need time and execution. John (01:15:59): Yeah, totally. Last question because I've had a lot. In what ways is Intercom itself AI-native? Des (01:16:07): The biggest initiatives we've driven recently has been around how we actually do R&D. So I think we launched this initiative, Darragh launched it, I think, about four months ago called “2x." Where we basically said, “Hey, we're going to double the productivity of R&D before February 1.” That means we measure this and everyone's going to poke holes in this, and that's grandgrand, but it doesn't really matter. The measurement is, I think it's deployments to production involving code that has to execute regularly. So it's a hard thing to fake. If you fake it, it's like we should probably fire you. Now. Interestingly, is not just the engineers going really hard. No, there's so many different elements to this, but one of the biggest ones that we saw recently, which really has been awesome in Emmet, our head of design, he basically said, “Every designer by August 1 needs to be shipping code.” This is the end to the discussion of should designers code, right? Designers should code. So we basically said, “Hey, all designers can now ship code.” Weirdly, the win there is that the amount of engineering distraction has just gone away. So every paper cut in your UI used to result in a GitHub issue that gets filed and triaged and gets picked up on a Friday in between breaks, or whatever. And now all that just goes. So now you're like, “Oh, I want to fix this button and fix that padding, fix that thing, change that radius, change that color.” All that shit just happens automatically and it's getting to much meatier stuff like, redesign this entire UI, redesigned this flow, change this wizard. All of that's now being managed entirely by our design team. What that has resulted in is engineers who are now doing far more, staying in the zone, far greater, using whether it's called code or Codex or any of those or Augment or any of tools to actually just be far more productive. And there's some real wins here. One reason we had was Fin works in Slack. But when we were building that, it was built very firmly from, “How do we use AI here?” So it was like, let's build one perfect Slack solution. Let's document all of our principles and then let's have cloud code, right? The Microsoft Teams, the Discord, the WhatsApp, it's every other solution. So we went live to production with Slack, but I think we have everything else now in public beta and it'll all go live. We're finding all these, it's a lot of 1.2x wins, but then every now and then there's a 50x productivity boost that we're finding. I think that's probably the biggest way in which the product is being built from an AI native way. On the go-to-market side of the house, we've been slower, but I think we're looking at, we've trained a GPT, if you like, on all of our marketing copy art principles, our content, our visuals, etc. And we've been using that to produce a lot of stuff like event invites and things like that. John (01:18:36): Well, Des, thank you. Cheers. Des (01:18:36): Thanks very much. Yeah.