Intercom Shifts to Usage Pricing, Pivots to AI Support Bot Fin

Intercom转向按量计费，全面押注AI客服机器人Fin 传统的SaaS销售模式已经消亡，取而代之的是一场基础设施之战；在这场竞争中，AI产品必须证明其在现实场景中的问题解决率并展现对人力的替代效果，才能促成交易。 在这次内容广泛的对话中，Des Traynor阐述了Intercom向AI优先型客户服务提供商的战略转型。这一转型由其自动化智能体Fin推动，目前Fin每周处理超100万次问题解决，成功率达65%。讨论深入剖析了B2B销售机制的演变：传统的功能清单已被围绕解决率与实际人力成本的严苛基础设施及指标之争所取代。Traynor坦诚分享了产品管理理念，强调深刻的用户同理心重于孤立的数据指标，同时详细介绍了Intercom如何通过舍弃5000万美元收入来彻底重构复杂的定价结构，以实现极致简洁。最后，他强调了公司内部激进的生产力变革，例如要求设计师交付代码，并借助Claude Code等AI工具实现非线性的工程效率提升。 Intercom经历了两次自我重塑：最初从具备核心CDP功能的实时访客消息工具，演变成为覆盖销售、营销和支持的综合套件；随后在2022年生成式AI爆发前夕再次转型，全面聚焦于客户服务。 (3:01) 在ChatGPT发布后，Intercom迅速构建并推出了AI智能体Fin；其解决率已从上线之初的25%增长至65%，目前每周处理超过100万次对话，单次解决收费0.99美元。 (5:53) Fin的架构是一个由约27个不同组件组成的复杂流水线（包括自定义模型、检索系统和重排器），绝非单纯套用第三方API的基础单模型包装壳。 (5:53) 评估AI客服时一个常见的误区，是将其与理想化、毫无瑕疵的人工客服进行对比，却忽视了现实中人类客服存在的语言障碍、响应慢以及服务时间受限等实际制约。 (7:54) AI要成功解决问题，需要深入的用户上下文和扎实的推理能力来应对现实中诸如“这个坏了”等模糊提问，这使其成为一项复杂的基础设施任务，打破了批评者所谓“周末就能做出来”的轻率预设。 (12:28)

Intercom co-founder Des Traynor argues that winning in the generative AI era requires software companies to abandon superficial feature additions in favor of deeply integrated, labor-replacing infrastructure backed by usage-based pricing and disciplined product execution. Pivoting Intercom to AI-First Customer Service (3:01) Facing declining net new revenue in 2022, Intercom narrowed its focus from a broad sales, marketing, and support suite exclusively to customer service shortly before ChatGPT debuted. Recognizing the shift, Intercom established a dedicated 50-person AI lab to build Fin, an AI customer service agent launched alongside GPT-4 in March 2023. Unlike basic wrappers, Fin relies on a pipeline of approximately 27 specialized components—including proprietary models, custom retrieval, rerankers, and summarization engines—to resolve edge cases through abstraction and deep customer context. Fin's Performance and Cannibalistic Pricing Model (5:53) Fin's resolution rate climbed from 25% at launch to 65% across roughly 6,000 customers, handling over one million weekly conversations and totaling 40 million end-to-end resolutions. Traynor notes this progress stems not from raw model upgrades, but from incremental system and interface optimizations. To align incentives and disrupt legacy competitors, Intercom prices Fin at $0.99 per successful resolution. While cannibalistic to the company's traditional per-seat revenue model, Traynor notes Fin achieved an approximate $50 million revenue run rate and is on track to reach $100 million in ARR roughly two years post-launch. Selling Labor Replacement Over SaaS Features (25:13) Traynor explains that marketing AI has shifted from visual UI showcases to infrastructure metrics focused on resolution rates, response speed, and customer satisfaction. Because procurement teams discount vendor claims, Intercom offers a 'Million Dollar Guarantee' against competing tools and provides prospects with evaluation suites to reveal the fragility of internal builds and basic bot wrappers. In Traynor's view, AI products must act as 'painkillers' that displace measurable human labor rather than unmeasurable 'vitamins' sold via flat per-seat add-ons, making usage-based monetization permanent. Product Discipline and User-Led Iteration (37:50) Traynor critiques startups that stall around $10 million in ARR by force-feeding unrequested add-ons to existing customers instead of expanding their user base. He advocates for 'permission to innovate,' where teams must ensure core software is loved before building adjacent features. Drawing parallels to Stripe's strict user-first philosophy, Traynor warns product managers against hiding behind analytical dashboards without directly interviewing end users, stressing that real-world feedback must guide software iteration over speculative investor trends. Internal AI Workflows and the 2X Initiative (76:10) To increase development velocity, Intercom introduced the '2X' initiative to double R&D deployment frequency. As part of this push, design leads mandated that designers write and ship their own production code, removing minor UI paper cuts from engineering backlogs. Engineers rely on AI-assisted coding tools like Claude Code to automate repetitive work; for example, the team authored a reference Slack integration and used AI agents to automatically generate subsequent integrations for Microsoft Teams, Discord, and WhatsApp. Traynor's investment framework prioritizes startups demonstrating deep AI differentiation, usage-backed revenue tied to critical business impact, and viable unit margins. However, managing model costs and eliminating 'dead tokens'—processing spent on failed or inaccurate answers—remains an ongoing engineering and margin challenge.

Intercom联合创始人Des Traynor阐述了公司如何从多功能SaaS收缩并转型为AI优先的客服平台，指出在生成式AI时代，企业必须依靠深层上下文架构、基于结果的定价机制以及务实的产品与工程纪律来建立防御壁垒。 从全功能SaaS到聚焦客服的AI转型 (3:01) Intercom于2011年创立初期作为通用工具起步，随着采用“待办任务”（Jobs-to-be-Done）框架，业务曾一度扩张到销售、营销和支持三大领域。然而到了2022年，面对净新增收入下滑的压力，重返公司的联合创始人Eoghan McCabe主导了战略收缩，决定完全聚焦于客户服务。就在该转型决定落地约十周后，ChatGPT正式发布，公司迅速调整研发重心。2023年3月，在GPT-4发布的同一天，Intercom推出了AI客服智能体Fin。 Traynor指出，Fin的表现摆脱了早期规则机器人的笨拙体验，其解决率从上线初期的25%提升至当下的65%，每周处理超过100万次对话，累计完成约4000万次端到端场景解决。目前Intercom已组建由50人组成的专职AI实验室，不再局限于对第三方API的简单包装，而是自建涵盖检索、重排与总结的定制模型。此外，为打破客户迁移客服底座的阻力，Intercom还推出了Fin Standalone，支持其直接接入Zendesk、HubSpot和Salesforce等竞品系统运行。 深层上下文架构与千颗铅弹的渐进优化 (12:28) 针对外界常认为周末就能拼凑出客服机器人的观点，Traynor指出真实的客户支持极度依赖用户上下文。大部分用户提问往往极其模糊（如直接说“坏了”），若没有底层的客户数据平台（CDP）预先捕获用户的身份、套餐等级及屏幕报错信息，对话系统便无法直接提供有效排查。此外，高解决率依赖于抽象推理能力，例如系统需要理解牙医诊所属于互联网企业范畴，从而基于Stripe文档作出准确判断，而非机械性地否认支持。 在底层架构上，Fin并非依赖单一模型，而是串联了约27个微型组件，针对检索、重排和最终应答等各环节分别匹配最快、最廉价且最轻量的LLM。Intercom虽以Anthropic的Claude Sonnet为核心模型伙伴，但整体设计具备可插拔性。Traynor强调，将解决率从25%推向65%的动力并非单一模型升级，而是由上千项微小改进组成的“一千颗铅弹”，涉及Prompt调优、UI交互改进及严格的千例评估测试集，单次改动通常仅带来约0.7%的边际提升。 按解决付费：AI时代的商业模式重构 (55:58) 引入Fin给Intercom带来了直接挑战：若AI解决了大部分咨询，客户对人工客服坐席的需求必然缩减，这直接冲击了传统的按坐席收费模式。为了与客户利益绑定，Intercom打破惯例，采取按每次实际成功解决收取0.99美元（或称1美元）的计费模式，即“仅在实际完成工作时代收费”。目前Fin已拥有约6000家客户，年化收入运行率（ARR）约为5000万美元，并预计在推出约两年后的明年初突破1亿美元ARR。 在经历早期功能混杂导致定价过于复杂后，Intercom曾主动放弃约5000万美元收入以清理和简化价格体系。Traynor将AI产品分为两类：直接替代高昂人工成本的“止痛药”型产品，其商业价值量化明确；而缺乏清晰人力成本对照的“维生素”型辅助功能，则往往被迫使用低价值的固定坐席加价。此外，针对AI部署初期的单位经济模型，Traynor强调必须通过工程手段压缩因生成错误答案而无法计费的“死Token”（Dead Tokens），而非被动等待基础模型降价。 B2B销售新范式与反思虚荣指标 (25:13) 在AI时代，B2B销售已从过去的UI美观度竞争与功能复选框罗列，转变为类似Intel与AMD之间的底层基础设施算力与指标竞争。采购方对常规营销宣称持严重怀疑态度，促使Intercom推出“百万美元保证”来证明产品实力。Traynor指出，向客户推销AI实际上是在销售对人工劳动的替代，买家最关注的是处理工作量与完成质量，因此销售团队需指导买家建立辨析能力，识别竞品的虚假宣传。 在初创企业运营方面，Traynor批评了行业内盲目追求估值与不切实际第二曲线的倾向。他认为许多企业在达到1000万美元ARR时陷入停滞，往往是因为未能将融资规模与实际总可寻址市场（TAM）对齐，随后便试图向存量客户硬塞不需要的附加功能来伪造增长。在内部产品评审中，Traynor坚持将“用户对此怎么说”置于数据看板之前，警告企业警惕因只顾满足边际销售线索定制而忽视核心用户体验，最终引发产品大规模流失的风险。 2X效能变革与AI原生工程实践 (76:10) 为从根本上提升组织交付速度，Intercom内部推行了代号为“2X”的生产力计划，目标是将研发部门的生产率翻倍。衡量该指标的核心标准是向生产环境部署具有实际业务逻辑的代码频次。在此计划下，设计团队出台硬性要求，规定全体设计师必须直接编写并提交代码，接管按钮调整、间距修改及完整向导流程等界面改造，从而消除了以往频繁提单打断工程师的琐碎摩擦，释放工程团队专注深层架构。 在系统工程方面，Intercom探索了AI原生开发路径。在研发Fin的多平台集成时，团队首先针对Slack场景构建并沉淀出高度标准化的架构规范，随后利用Claude Code等AI工具自动化编写针对Microsoft Teams、Discord及WhatsApp的适配集成，实现了非线性的研发效能飞跃。类似工作流也逐步渗透到市场端，团队利用基于内部准则微调的GPT模型，实现活动物料与营销文案的自动化生成。 在技术成熟度与客户预期不断博弈的生成式AI演进期，真正的壁垒不在于调用通用基础模型的速度，而在于企业能否将专有上下文、严格的单位经济模型与重塑后的工程文化深度整合；这种转变既要求初创企业克制盲目扩张的冲动，也对遗留SaaS厂商的自我颠覆决心提出了考验。

[0:00] Des Traynor: Back home, drinking is a very serious activity. I wouldn't make a game out of it. [0:00] Des Traynor: they were literally giving it away at the end and still couldn't get rid of it. [7:54] Des Traynor: A human artisanal handcrafted answer... won't be there in seven seconds. It'll probably be 18 minutes [14:14] Des Traynor: Will AI people learn how to do CS, or will CS people learn how to do AI? [25:13] Des Traynor: We've got 24 checks, they've got 17, these seven matter, we're in. [31:48] What are you actually trying to say? [31:48] Des Traynor: 99.4% of the world has still never heard of your product [35:14] Des Traynor: shipping is an act of confidence and humility. [37:50] Jared Spool: Have you tried asking them? [37:50] Des Traynor: What did our users say about this when you showed it to them? [40:00] Des Traynor: marauding churn bomb [45:17] Des Traynor: not aligning your fundraising with your TAM [51:05] you just want to be able to get a good night's sleep. [54:56] John Collison: I call Ricky and he decides what the discount is [55:58] Des Traynor: we only get paid when we do work [59:42] John Collison: In my day, businesses didn't do that, [71:00] Des Traynor: 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? [76:10] Des Traynor: If you fake it, it's like we should probably fire you.

[0:00] Did you ever work in a bar? [0:01] I've never worked in a bar. [0:02] I have a Guinness keg at my house. [0:03] Actually. Do you want another pint? [0:04] Yeah, sure, sure. [0:05] I was talking to a guy who runs procurement [0:06] and he was saying like, [0:07] you have to understand I'm just getting bullshit to all day. [0:10] You know, it's just all day relentless, [0:12] absolute nonsense in my face. [0:14] People do not have enough empathy [0:15] for the procurement person who just has to endure, [0:17] nonstop, nonsense. Absolute garbage. [0:21] ChatGPT launched, I think on a Thursday. [0:23] I had a call with, Fergal, our head of AI on a Friday. [0:25] Eoghan made the decision, like, [0:26] on the Sunday I think it was, [0:28] and we started working on like the AI version [0:30] of Intercom on the Monday. [0:31] It was a lot easier [0:32] to invest when being a founder was uncool. [0:34] I blame genuinely The Social Network. [0:36] I blame just kind of the entrepreneurial lifestyle. [0:39] I blame TikTok, I blame all these things— [0:42] Soho House. Yeah, yeah, exactly. [0:45] You just show me your technique [0:46] and I will learn from you. [0:48] This will be the first time [0:49] I've ever poured a pint to get a Guinness in America. [0:51] My fellow Irishman, Des Traynor, [0:52] is the cofounder of Intercom, [0:53] the customer service giant turned AI company. [0:56] He's also a prolific blogger [0:57] and one of the most respected voices on product strategy. [0:59] Cheers. Cheers. [1:02] Okay. [1:05] You are my first Irish guest [1:08] and so I actually have a critically important question [1:10] that has been burning this entire time. [1:13] Whenever we release these episodes, [1:15] all the YouTube commenters are obsessed with people [1:17] splitting the G. [1:18] This was never a thing for me growing up, [1:21] a thing I never saw in Ireland. [1:22] It's like an invasive species maybe from TikTok. [1:25] Is this a thing? [1:27] No, it's not amongst anyone who you would respect. [1:30] It's very much [1:31] an actual TikTok thing. [1:32] It's a slight bit of a tourist thing because of that. [1:35] But it basically means like you're drinking, [1:36] I think it's a quarter a pint in your first mouthful. [1:38] And it's like, I don't know, I don't [1:39] A waste of good Guinness. [1:40] Yes. Yeah, okay. [1:41] It's a little like when I first came to America, [1:43] drinking games, [1:44] it's like back home drinking is a very serious activity. [1:45] Exactly. [1:46] I wouldn't make a game out of it. [1:47] Exactly. [1:48] Yeah, yeah, yeah. I was very confused by that in America. [1:50] Okay, and then my other stout related question is, [1:54] I saw you opining on like Beamish, Murphy's, [1:56] everything like this. [1:57] I don't know, what is your, [1:59] what is your view on the stout landscape [2:01] in particular? [2:02] I probably default to Guinness. [2:04] There's been, like, there's been moments when I go Kilkenny [2:06] Beamish, Murphy's. [2:07] It kind of, usually it's like when you're over there [2:09] at their respective hometowns. [2:10] I see. But for the most part, [2:11] I default to Guinness [2:12] and for like one brief day I tried Island's Edge. [2:15] I don't know if you remember [2:16] when Heineken launched, it was Heineken. [2:18] Heineken launched— Launched a Guinness Killer [2:20] like, and it lasted [2:20] all of like six months or something like that. [2:22] They were literally like, I think literally [2:23] they were giving it away, [2:24] at the end and still couldn't get rid of it. Yeah. [2:26] Have you heard all the Guinness stats about, you know, [2:27] the Guinness used to be a majority [2:29] of Ireland's stock market. [2:31] Obviously the canal system was built [2:33] for Guinness distribution, [2:34] but sometimes when you tell people [2:35] that Guinness used to be a very significant part [2:37] of Ireland's economy. [2:38] You know, they don't believe you. [2:40] But the stats are really there [2:41] And it's weird. [2:42] Like it trickles into modern day like [2:43] where I live by Castleknock [2:45] like a lot chunks of the Phoenix Park like are owned still— [2:49] By the Guinness family. [2:50] Or exactly or used to be there [2:51] and it's like a, they don't, [2:53] and then maybe they donated it to OPW [2:55] or something like that but it's very much still [2:56] kind of carries forward. [2:57] Yes. Okay. [2:59] We're not actually going to talk about Guinness [3:00] the whole time. [3:01] We should also talk about Intercom. [3:02] And I was thinking in preparing for this, [3:04] I'm very impressed by businesses [3:07] that can reinvent themselves [3:12] or maybe even reinvent themselves multiple times. [3:13] So you think about Netflix, [3:14] they started with the original DVD by mail business [3:18] and then, oh my god, the internet's coming [3:19] and they had a few abortive attempts at streaming, [3:22] like remember the whole Qwikster debacle [3:24] but then really cracked streaming movies. [3:27] And so, you know, you can watch "The Godfather" [3:29] or you can watch whatever movie you want on Netflix [3:32] and then of course as they got more squeezed [3:34] by the rights holders, [3:35] you actually can't go watch "The Godfather" anymore [3:37] or you type any movie into Netflix [3:39] and it's not there [3:40] because now Netflix is all first party content [3:43] that they have developed themselves. [3:44] And so they've reinvented the platform once again to be, [3:47] rather than watching other people's content, [3:48] watching Netflix content, [3:50] they've twice reinvented the company from DVD by mail [3:52] to streaming third-party content [3:54] to streaming first-party content. [3:56] Intercom strikes me as a business [3:57] that has similarly been reinvented twice [4:00] where you guys got started with the Intercom feature of, [4:03] you know, you can talk to your customer through the website [4:05] and then you guys became a customer service company, [4:08] which is actually different for reasons we can talk about. [4:10] And now you're becoming an AI customer service company. [4:13] That's my theory of Intercom. [4:14] Is that actually an accurate theory? [4:16] Yeah, that's true. [4:19] There's a bit of like extra context I'd give it. [4:21] When we started it wasn't really, [4:23] it was like about like, [4:24] this is like literally when we started [4:26] and our initial plan was like, hey, like, [4:29] you remember internet [4:30] because like Stripe was in its early days back then as well, [4:32] but like there was no tooling [4:33] to run a SaaS business. [4:35] There was like literally nothing. [4:35] It was like you were using, [4:36] you were kind of abusing like PayPal for payments [4:39] and you're using like MailChimp [4:40] for like talking to your customers and stuff. [4:42] You know, we had this idea of like talking [4:43] to your customers is really important [4:45] so someone should work on that, you know. [4:47] And like there was so much stuff [4:49] in the early releases of Intercom. [4:50] Like it was the first live CDP, [4:52] like you can actually see [4:52] who's live in your product right now [4:54] and what they're doing [4:55] and you could store data against them. [4:56] You could say show me all my premium customers [4:57] and weirdly like use cases like that still exist today. [4:59] I was in Denmark last week and I was like, [5:00] oh I should message all my Copenhagen customers and see you. [5:03] And like all that stuff is like you're still [5:04] kind like who else is doing this? [5:06] So we sort started out like, [5:07] with a very, very general purpose idea, [5:10] which was like let internet businesses talk [5:12] to their customers [5:13] and then we kind of fell in love with this jobs [5:15] to be done methodology [5:17] and like as one of the things you do in that [5:18] is you look at how your product is actually used [5:20] and then you iterate on it from that point of view. [5:22] And that kind of led us down this path [5:24] of sales, marketing and support. [5:25] And then I guess to skip a load of years here, [5:28] let's just say 2020 things were great, [5:30] 2022 things weren't as great [5:32] and then it was like we need to focus [5:34] A concise— Yeah, exactly. [5:36] Let's just cut a lot of the reasons why or whatever, [5:38] some sort of disease as well along the way. [5:41] But yeah, so 2022, like it was like hey, you know, [5:44] Eoghan return, Eoghan left in 2020 I think, [5:47] Eoghan returned, you know, [5:48] business had been like in declining net new revenue. [5:50] He said like we need to focus, [5:51] we're going to focus on customer service. [5:53] And then a short time later, [5:54] I think maybe like 10 weeks later, [5:56] AI happened. ChatGPT launched I think on a Thursday. [5:59] I think, I had a call with Fergal, our head of AI, [6:01] on a Friday. [6:02] I spoke with Eoghan all through the weekend [6:03] and like we made the decision, [6:06] Eoghan made the decision, like, [6:07] on the Sunday I think it was. [6:09] And we started working on like the AI version [6:11] of Intercom on the Monday and that was like 2022. [6:13] And I think, like, were it not for that, [6:14] it's hard to say exactly [6:16] where things would've gone but certainly [6:17] that's the reason I'm sitting here, yeah. [6:18] And obviously people are naturally wired to be skeptical, [6:21] you know, when they hear the AI version of Intercom, [6:23] you have, you know, everyone and their mother out there [6:25] saying we're an AI company now. [6:27] Yeah but you actually are an AI company now. [6:28] Mm-hmm. And so, [6:29] describe what the AI version [6:30] of Intercom actually means. [6:32] The biggest thing it means for us is our product Fin. [6:34] So we launched Fin in March, I think 2023. [6:38] We launched a few little AI features. [6:40] We were the first people to actually build anything [6:41] on, like, the GPT-3.5 [6:43] and then we launched Fin in March, GPT-4 launch day. [6:46] And Fin was, like, basically, like, [6:49] the first chatbot that worked, [6:50] was the best way you could think about it. [6:51] What that really meant [6:52] was it could actually have conversations [6:53] and answer questions. [6:54] And when we launched it, [6:55] it was doing I think 25% resolution rate [6:57] and that was like crazy numbers. [6:59] Today, it's like 65% [7:00] and today Fin's resolving about a, [7:02] I think it's over a million conversations a week. [7:04] It's handled about 40 million [7:06] actual end-to-end customer service scenarios to date. [7:08] It's you know, growing like over 200% year over year. [7:11] It's like, you know, it's, we charge a dollar per answer [7:14] so you can work the revenue out or 99 cent even. [7:16] It's becoming like just this, you know, [7:18] AI sort of growth story inside Intercom, [7:21] which is already like a sort of, you know, [7:23] a mature SaaS business into hundreds of millions of revenue. [7:25] But I think when we think about, [7:26] like what does it mean to be AI, [7:27] it's like first of all, [7:28] what is the future growth of your business? [7:30] And the answer is AI. [7:31] And then over the last, say, six months [7:32] we've been going hard on being, like, [7:34] a kind of properly deep AI company. [7:36] We're now at a point where we're like, you know, [7:37] we're using our own models inside Fin, [7:39] we're using custom reranker [7:40] or custom retrieval, summarization, et cetera. [7:43] And we're doing a lot of this work. [7:44] We have like an AI lab of 50 people [7:46] and we really just kind of have gone all in, [7:47] on the idea that like, you know, like obviously, [7:49] Intercom still has a help desk product [7:51] but like the entire future of CS is clearly going to be AI [7:53] and that's what we're all in on. [7:54] Yes, yes, maybe there's the repeated pattern in tech [7:57] where the enthusiasm for technologies [7:59] comes before the technology being ready. [8:01] And so people are excited about computer games [8:03] before the wave of like good computer games. [8:05] Yeah. Yes. Or you know, [8:06] people were pitching mobile internet [8:08] and you know, you'll buy cinema tickets via WAP it's like— [8:10] And J2ME and all that, yeah. [8:11] Exactly. Yeah, yeah. [8:12] And you know, we actually had our version of that, [8:13] like we had a product called Resolution Bot. [8:15] It was originally called Answer Bot. [8:16] But I think Zendesk tried to sue us [8:17] Because they had a competitive product [8:18] but like Resolution Bot was actually a good AI product, [8:21] at the time, but it was just AI wasn't there. [8:23] And so that was the actual reason [8:24] why we had such a headstart. [8:25] Because we actually already [8:26] had a little AI group ready to go. [8:27] We'd already built a RAG engine ready to go. [8:29] Yeah. So we were able to jump [8:30] a lot quicker than a lot of folks. [8:31] But yeah, like it's, I think a lot of these products, [8:33] you're right, like have kind of two or three stabs [8:35] before they go mainstream. [8:36] Yeah, and there was this whole enthusiasm cycle [8:38] for bots in 2017 I want to say. [8:41] And the tech just wasn't there at all. [8:42] It was a horrible experience for customers. [8:44] It was also quite clunky to set up for businesses. [8:46] Yeah, yeah, yeah. And at some point [8:47] there was a genuine question at times [8:50] where it was just like, [8:50] is a web form not just better? [8:53] I think in a lot of cases it was. [8:54] Yeah, yeah, yeah. [8:55] Whereas I think now people are starting [8:56] to have the experience of, you know, [8:58] the classic thing is you know, [8:59] you're talking to a bot, [9:00] it's like please will you please just connect me to a human? [9:02] Whereas now it's like can you just connect me [9:04] to a bot and like can we— [9:05] We see that a lot. [9:06] We see like, oh hey Jenny, sorry to bother you. [9:08] Can you put back onto Fin? [9:09] It was actually doing a great job. [9:10] It just thought it wasn't, you know, [9:11] Because I asked a few too many questions. [9:12] But I think like there's a general pattern [9:14] we're noticing which is like, [9:15] a lot of experiences are just better digitized [9:18] Because partially because of like human considerations. [9:21] Like one of the reasons people go [9:22] to like the the kiosk in McDonald's, I suppose, [9:24] as opposed to their actual counter [9:27] is because they don't want to have to think out loud [9:28] in front of a human. [9:29] They're like, oh give me a second, [9:30] do I want fries? Yeah. [9:31] A lot of the reason why people prefer Waymo, [9:33] for some people, [9:34] it's like I just don't want to have the conversation. [9:35] I don't know the awkwardness around tipping [9:36] or whatever it might be. [9:37] And I just think with what we see a lot is like, [9:40] once Fin answer one question well, [9:42] people are like oh this thing's paying out. [9:43] Let me, now that you're doing [9:44] that I'm going to ask you all the things [9:45] I was wondering about. [9:47] Yeah, yeah, yeah. Whereas I think [9:48] they'd feel or probably nearly weird unloading all that [9:50] on one poor CS rep, you know? [9:52] Ah, okay, so you're seeing a lot of induced demand [9:53] where people use interactive customer service more? [9:57] Which is great, which is interesting, [9:57] because it turns Fin into not just being a kind of a, [10:00] you know, a cost takeout [10:02] but also it's like how much better would your business be [10:04] if everyone knew how to do everything they wanted to do? [10:06] Yes. And like the answer is all, [10:07] you know, a lot of times it's a lot better [10:08] and it's not just the whole, like, [10:10] I don't want burden a human, [10:11] it's also like people often, [10:12] one of the biggest fallacies in AI [10:13] is people compare it with, like, [10:14] with this perfect human that does not exist, [10:17] like the the driver that never crashes. [10:19] Or like in our case, [10:20] it's like well a human artisanal handcrafted answer, [10:23] and I'm like, yeah, like let's pretend [10:24] that would be there in seven seconds. [10:26] It won't be, it'll probably be 18 minutes. [10:27] It's also not going to be perfect. [10:29] It might not speak that language. [10:29] And let's presume you're doing those handcrafted answers, [10:31] which you're not. Yes, exactly. [10:32] Someone is like trying to close the case [10:34] before they move on to the next one. [10:35] So like I think yeah, comparing it like, [10:36] I know we have this thing where we expect our AI products [10:38] to be flawless and we're totally tolerant [10:40] of like human showing up hung over only [10:41] speaking one language [10:42] and only working six hours or whatever, [10:43] you know, it's just, it's a funny contrast. [10:46] Yes, yes. That's interesting. [10:47] And it's interesting you talk about product onboarding here [10:51] because I think of Intercom as you guys are very, [10:54] you guys have a house view [10:55] that product onboarding should be much better. [10:56] And I remember a lot of the use cases you would talk about [10:58] for the original Intercom taught your customers [11:00] through the website, was that, [11:02] you can have personalized nurture tracks [11:04] and like it's weird that you drop people into SaaS products [11:06] and just expect them to be able to use them right. [11:08] And you should see how people are using the product [11:09] and then give them kind of specific [11:11] steers based on their usage. [11:12] And it sounds like you're coming to this vision again, [11:15] which is people should have better onboarding support, [11:17] people should be nurtured along based on their use case, [11:20] but now kind of interactively AI powered? [11:22] Yeah, I mean we talk a bit about like, [11:26] what is ultimately a customer agent going to be. [11:28] Like that's what Fin will be when it, as it grows up, [11:30] it'll just become like this way [11:32] in which customer conversations are handled. [11:34] And obviously the most direct attack here [11:36] is like customer service. [11:37] But you know, I think every single customer touchpoint [11:41] can be improved by like, [11:42] by basically immediate accurate answers available [11:45] all the time. [11:46] Isn't an obvious limitation, [11:47] like right now you require customers [11:49] to come up with a prompt [11:50] and if you look at why TikTok is so successful, [11:52] it's like I would never prompt for, you know, [11:54] I want to see videos of planes landing low [11:56] over the beach in St. Maarten, [11:58] but like it turns out [11:59] that's what I want to see. That's what [12:00] Yeah, yeah, yeah. [12:01] And similarly people probably [12:03] have many more things they need [12:05] than they will actually come up with a prompt for. [12:07] And I think the product today is still mostly prompt based. [12:10] It's reacting to a customer. [12:11] Exactly, a customer has to come along [12:12] and like type things into the box. [12:13] I mean today that's what customer service is. [12:16] It's still like kind of like, [12:17] here's my problem and then we'll solve it. [12:18] I think, you know, for sure there's obvious directions [12:21] this will go like, you know, [12:22] hey what does a good customer look like? [12:24] And maybe we can honestly infer that as well. [12:26] But certainly like people like you [12:28] should do things like this. [12:29] It's definitely an understandable domain [12:32] and then I just think working out the right level [12:33] of interruptive help, [12:34] you know, like you don't want to be too naggy [12:36] or too pop-upy like, you know, [12:37] it gets kind of quite grating. [12:38] But I think if you can get the first message right, [12:40] you can sort of say, hey, if you come here [12:42] you're always going to get the thing you should do next. [12:44] Or like the thing that looks like you're stuck on. [12:45] Like if someone's on the, I know the renewal page [12:47] and they have an error message, [12:48] we know they're probably going to open the thing [12:50] and we know they're probably going to [12:51] say something that's got, [12:52] you know, a lot of the context already there [12:53] so we can work out the right things to say and do, [12:55] I think that's pretty doable. [12:56] It feels like you could do a lot around, yeah. [12:59] You train a model on what the customer is seeing [13:02] on that webpage at that moment in time [13:04] and use it to feed the answer and things like that. [13:06] We already do a lot of that already, [13:08] like as in customer context. [13:08] So like, so knowing that they're knowing that it's John [13:11] and he's on the premium plan and he's on the playlist page [13:14] and there's an error on the screen, [13:15] is all useful information when it comes to, [13:17] because like, in people, a lot of like the, [13:20] you know, YC I could build that [13:21] in a weekend type Hacker News crowd. [13:23] I think one of the things they often, you know, [13:26] they're thinking every customer support query [13:27] begins with like, hi there, [13:29] my name is blah and my user number is blah. [13:30] But actually like most support conversations begin [13:32] with like, this is broken and you're like, [13:34] what's broken? [13:35] And like, so to solve that, [13:36] you need like a reply engine [13:38] that's just like, hey, let's chat [13:39] about what's going on here. [13:40] But like we realize quickly, [13:41] like people will kind of disengage. [13:42] So any amount of extra context [13:44] that when you say this is broken [13:45] and if someone says this is broken [13:47] and there's a big red error box on the screen, [13:48] we're like, well it's probably that thing [13:49] that you're talking about. [13:50] A lot of people just don't realize how deep you have to go [13:52] to actually do a great job. [13:53] And like we're like, we do like say, [13:55] if you install Fin today, [13:56] you get like 65% resolution rate after 30 days, [13:58] like that's shocking. [13:59] But like we have had to go really deep to actually get [14:02] to those numbers and it involves like all sorts of, [14:04] every single smart thing you can think of we've had to do [14:07] and then optimize [14:07] and then find the right model for all that. [14:09] But one of them is customer context [14:10] and that obviously answers a lot of things. [14:12] Yes. What are the other smart things? [14:14] Abstraction. [14:15] So like, you know, like I guarantee you've got like no pages [14:18] on your website that say Stripe works really well [14:20] for a dentistry office, right? [14:21] You probably don't have that in your docs. [14:23] A very naive RAG bot will be basically like, [14:25] well it doesn't say dentist [14:26] and we're told not to hallucinate. [14:28] So no we don't do dentists. Sorry. [14:30] And like the abstraction is like, you know, in that case [14:32] as well, what is it dentist, it's a type of business. [14:34] Does Stripe work for businesses? [14:35] Can dentist be internet businesses? [14:37] Well we say we're great for internet business. [14:39] You know, so you're kind of working out what's the best, [14:41] what's the best sort of risk tolerant way [14:44] to make grounded inferences without going over the cliff. [14:47] You know, that's one of like 27 different [14:49] sort of components of Fin. [14:50] Then you've got obviously your RAG [14:52] and then you've got like, you're like, [14:53] hey is this is an escalation appropriate at this time? [14:55] Like, hey, have they threatened something [14:57] or have they, you know, [14:58] like you have every single type of problem, [15:00] it kind of ends up you have to walk through it all [15:01] to actually recreate customer service. [15:03] I think a lot of times people will compare it [15:05] with like, how do I reset my password? [15:07] Huh, I found it and you're like, right. Cool. [15:09] That's like 0.4% of the scenarios you deal with [15:12] when you're in customer service. [15:13] There's like the, there's a pattern. [15:14] Did you ever see the movie "Armageddon" where like, [15:17] it's like Bruce Willis and Ben Affleck or whatever. [15:19] But the gist of it is they train a load of, [15:20] I think it's like oil drillers to become astronauts [15:22] and the comedy, the joke at the time [15:24] that Ben Affleck always says he got drunk [15:25] and he recorded the voiceover for the DVD [15:27] and he was like, I always said, well why didn't we just [15:29] train the astronauts to drill oil? [15:31] Like surely that's an easier problem. [15:32] I think the thing that we're realizing [15:34] with the AI movement is some version [15:35] of what's going to happen sooner? [15:37] Will AI people learn how to do CS [15:38] or will CS people learn how to do AI? [15:40] Thankfully as I said, we kind of started off [15:42] with CS and AI in our DNA. [15:45] Oh you say people, you mean companies in this case? [15:46] Exactly. Will OpenAI get better [15:47] as customer service faster [15:48] than Zendesk gets better at AI. [15:49] Yes, exactly. [15:50] Exactly, then I think we just, [15:51] we were lucky in that we kind of had already backed [15:53] both horses somewhere along the way. [15:55] So yeah. [15:56] One thing I find interesting about what you do [15:57] is every company is thinking about AI right now, [16:00] you know, every company had a board meeting in 2023 [16:02] where the board is like, [16:03] can we do a special deep dive on AI? [16:04] because it just feels like it's a lot happening [16:06] and we need to be making sure we're, [16:07] you know, on the leading edge of AI. [16:09] And then every company was like, oh, [16:11] we're actually doing a lot in AI, for example, [16:13] we've seen great automation wins in customer service. [16:15] Yeah, yeah. And so, [16:16] it's kind of like, you know, [16:17] the joke about the bike shed, you know, [16:18] versus the nuclear power plant [16:19] where everyone has opinions on how to build a bike shed. [16:22] Similarly kind of, everyone has opinions [16:24] on how to do AI customer service. [16:26] And so I'm curious how you sell, given that, [16:31] I'm guessing a lot of your customers [16:32] think, oh we know how to do that. [16:33] It's not that hard. [16:34] We hooked it up to a model that, [16:36] you know, we're actually very smart on this topic already, [16:38] how you sell in that environment [16:39] where everyone has opinions? [16:40] Yeah. there's like, I've never seen the like build versus [16:44] buy thing play out more often than we do today. [16:46] Especially with like certain, [16:48] like a lot of customers are like, [16:48] you know that meme on Reddit, [16:50] but I'm not like good or girls or guys whatever. [16:51] Like there's a lot of that where it's like, [16:52] oh you would never possibly understand is [16:55] B2C shopping company. [16:56] And you're like, really, I've never heard of such a thing. [16:58] Sometimes honestly we just be like, hey look, you know, [17:02] Godspeed, you go and start building this. [17:04] PS here's a torture test. [17:05] When you think you've got something good, [17:07] run these a hundred questions to us, let us know. [17:09] Oftentimes that's where they're like, yeah, okay, [17:11] we think we need to buy your product. [17:13] But like, I think there is a, everyone has this idea [17:17] of like, in a move to AI, what can we definitely do? [17:20] And we can definitely answer questions like [17:22] how do I reset my password? [17:23] And again, this is the back to the hole [17:24] that's such a small amount. [17:25] What they can't do is actually have conversations [17:27] and all that sort of stuff. [17:29] We're like what is your opinion on the president [17:31] and how they're performing? [17:32] And like a lot of times, well you don't want, you know, [17:34] you don't want anyone to answer that question [17:36] on behalf of your company. [17:37] But I think a lot of times people like they dip their toes. [17:41] It's almost like they fired a tracer bullet. [17:43] They're like, yep, [17:43] this seems like we're making great progress. [17:45] And every AI product has this problem [17:47] where you like make epic progress in the first two weeks. [17:49] And then you hit this wall, this plateau [17:52] and like you know, two years later you're telling people, [17:54] oh Apple Intelligence is coming in 26 or whatever, right? [17:56] So like in this case a lot of people start the project [17:59] feel like they definitely don't need to buy Fin. [18:01] We just help them understand the difference [18:03] between a good bot and a bad bot [18:05] and then they come back and they buy Fin. [18:07] So where is the Fin business these days? [18:09] I'm curious both just how it's performing on revenue metrics [18:13] and then are you selling it to existing Intercom customers? [18:16] Are you selling it to new accounts? [18:17] Just how does the whole thing work? [18:19] We're about 6,000 customers and growing quickly. [18:22] Fin does about a million resolutions a week. [18:24] We're charging a dollar per resolution. [18:26] So you can do- [18:27] That's a 50 million revenue run rate, [18:28] Give or take. [18:29] When we launched initially we sold to just [18:32] to our own customer base. [18:33] And then as we kind of progressed, [18:35] we realized, hang on, moving help desk is a nightmare. [18:39] Like you've probably done it once or twice, right? [18:41] We've done— It's a big job. [18:42] One to another, right? [18:42] It's a whole ordeal and Fin is brilliant. [18:46] So we're like loads of people want this product [18:47] but can't buy it. [18:48] So we made the decision to launch like [18:49] what we internally call Fin standalone or Fin for platforms. [18:52] So now you can use Fin on top of Zendesk or HubSpot [18:56] or Salesforce or any of those as well. [18:58] So we basically Fin is available to everyone [19:00] and that's a relatively new muscle that we've been growing, [19:02] but it's actually, you know, [19:03] that's kind of where we see a lot of the future growth. [19:05] And so do you, is Fin like [19:11] you can connect your iPods to Windows for you know, [19:15] iTunes for Windows, [19:16] but we hope that one day you buy a Mac [19:17] and it's part of the whole digital hub strategy [19:19] or we're actually now all in on Fin the engine [19:22] and whatever customer service platform you use is actually [19:25] not a topic of huge interest for us. [19:27] This is such a core question [19:28] that we kick back and forth quite a lot. [19:30] This is the offsite debate that just [19:31] currently had. Yeah, genuinely like, or [19:33] at least it's certainly one of them. [19:34] The way like we think about it, first and foremost, [19:36] is like the future is AI. [19:37] So like Fin just has to win kind of like at all costs, [19:40] including our help desk. [19:42] Weirdly our customers are like, [19:43] they turn Fin on and are like, damn, this thing's good. [19:46] Hey net like 65% of our support volume, [19:49] maybe we don't need X, Y, Z competitor [19:52] and maybe we can go all in on your help desk too. [19:54] And we're like, okay cool. [19:56] That wasn't our game plan. [19:57] But we're happy to help if you know what I mean. [19:59] Right, I think like the actual, [20:00] the battleground we care most about genuinely [20:02] has to be the AI agent. [20:04] That's the one where like we care about most, [20:06] but it does produce a lot of like, you know, demand [20:09] for the actual help desk product too. [20:12] Oh my God. [20:14] Oh dear. [20:15] Okay, so you've actually played darts. [20:17] Well I've been around at dartboard. [20:19] I'm curious what your AI stack looks like. [20:21] Where concretely, you know, what are the models [20:25] or collection of models and prompts [20:26] and everything that you are using in production. [20:29] How do you handle model upgrades given that you know, [20:31] the behavior is changing so much underneath the hood? [20:34] How deep do you go in terms [20:36] of developing the stack yourself? [20:37] And you can talk about the stack. [20:38] First thing I'd say is obviously Fin isn't like one thing. [20:40] It's like 27 different things or whatever. Right? [20:42] So every one of those is like, [20:44] whether it's the summarizer [20:45] or whether it's like the reranker or the retrieval engine [20:48] or any of these, or the direct answer, [20:49] which is where we actually go and formulate the answer. [20:51] Every one of those is paired with the like the fastest, [20:54] cheapest, lightest, most accurate LLM [20:56] that can actually do the job reliably. [20:58] Like very, very high reliability. [21:00] So like that means like there's no one particular model. [21:03] So our primary partner will be Anthropic [21:07] for Claude Sonnet. [21:08] We've architected it such that like, [21:09] we can plug and play various different pieces [21:11] and whenever a new model comes out, [21:13] or honestly a new idea for a new architecture is in, hey, [21:16] we recently launched the ability for Fin [21:17] to do complex queries, [21:18] which would be like say go and issue the refund [21:20] and update the name on utility builder [21:22] or something like that. [21:23] And whenever like we have to change the architecture, [21:25] we have this kind of like arduous torture test [21:28] of like thousands or at least a thousand CS scenarios [21:32] where like we have like, here's the question, [21:34] here's the context we're provided, [21:36] here's what like the current Fin answer to this question is, [21:39] here's the best available human answer that we know of. [21:41] And then basically here's what this new version [21:43] would offer us. [21:44] And then so like whenever we say like, [21:46] oh GPT-5 comes out [21:47] or something like that, we're like, you know, [21:50] the reason we're not just a lot of our competitors [21:51] are kind of quickly, oh we never run a GPT-5. [21:53] And I'm like, ooh, I'd take a beat on that one, [21:56] you know, like you shouldn't assume [21:57] everything's going to be great [21:58] for your use case always, right? [22:00] And so we ultimately, we have to run it [22:02] through this pretty like expensive test to work out [22:04] where the edges are, [22:05] if it's scoring higher resolution, right. [22:07] We need to understand why, [22:08] because it could be just [22:09] that it's like trying more stuff. [22:10] But that could also, that the shadow side [22:12] of that could be like excess hallucinations or whatever. [22:14] So whenever a model upgrade comes, [22:15] we have to trigger this whole thing. [22:16] But when we launch Fin it was like 25% resolution. [22:19] Today it's like 65. [22:21] We've been increasing at roughly a percentage [22:23] point a month give or take. [22:24] But like very little of that is actually [22:25] because of the upgrades or the bumps from models. [22:29] Genuinely, I actually think, [22:30] and I say this with a lot of respect [22:32] and love for the CS craft, [22:33] I actually think we've had enough intelligence [22:34] for CS for quite a while. [22:36] In fact we published some material on this [22:38] on our research blog read most recently. [22:40] Like when you look at like, you know, [22:41] you know people are saying things like, oh, like the latest [22:43] whenever you know, Grok can compete [22:45] at mathematical Olympiad like level seven or whatever. [22:48] We're like, right, I think it can probably do most CS. [22:52] You know like, [22:52] so it's often not a lack of intelligence [22:55] is the reason why we're not a hundred percent. [22:57] It's because they're too distracted by Ani. [22:59] Exactly. [23:00] A lot of the wins come from, like, [23:02] honestly better architecture, [23:03] better like tailored models [23:05] or like changing in the UI can change exactly [23:06] how things work. [23:09] And then sometimes you will get an occasional bump [23:11] here and there from a model swap. [23:12] It strikes me that a lot of the, [23:14] how you include the amount, the account context [23:18] and the amount of account context you include [23:21] is a big part of the secret sauce. [23:23] Perhaps. [23:23] But there like it kind of, it varies customer to customer. [23:26] It really is one of these areas [23:27] where like it's a thousand lead bullets. [23:28] It's not like a single silver one. [23:30] Like it's not, if you look at a resolutionary graph, [23:32] there's no pop like, give or take one or two little tweaks. [23:35] Like it's mostly just, hey we ground out through like, [23:39] you know, optimizing this prompt and changing this handover. [23:42] We ground out another 0.7% and like, you know, [23:44] and you see AI team celebrate [23:46] that on the balcony on the Friday being like, [23:47] yay, 0.7 up, whatever. [23:49] It's hard to work out exactly what bits [23:51] and then there's obviously multiplicative benefits. [23:53] Like you might have a win over here, [23:54] it cost do something over here as well. [23:55] Yeah. How deep down the stack will you go? [23:57] Like what's Intercom's version of Apple Silicon? [24:00] That I don't know for sure. [24:01] I mean like we're going to chase any edge [24:02] we can get right now. [24:04] I think custom models is definitely where we're going [24:07] and that's like a large investment from the AI group, [24:09] which is like the most contested resource that we have. [24:11] Every bit of work they're doing is finding a new edge [24:14] in resolution rate or resolution quality. [24:16] So right now it's paying out pretty well. [24:18] So I think we're going to kind of place all our chips there [24:21] until something changes. [24:22] What does selling AI look like? [24:24] It's quite difficult. [24:24] It's difficult in marketing and selling. [24:26] Because I think like, [24:27] because it's so crowded and noisy? [24:28] Well there's that, but it's also like, it used [24:31] to be the case and for sure Intercom used to be [24:33] one of these companies [24:34] where we our product looked the nicest. [24:36] So all we had to do was to age old, you know, [24:39] blah blah blah, reinvented and in a big sexy screenshot. [24:42] And you can still get away with that in certain domains, [24:44] Linear can get away with that [24:45] because their product is the sexiest, right? [24:47] I think with AI, everyone's chat bots look the same. [24:49] Everyone's kind of copied our messenger. [24:50] Everyone, you know, everyone's kind of like roughly like [24:53] converging on a certain UI paradigm. [24:55] And so you have to ask like then when we say like, you know, [24:58] we are the best AI agent, [25:00] like what do you think all the rest of them are saying, [25:01] we're the worst? [25:02] Like, no, of course. [25:03] So they're all saying this [25:04] and then everyone has the same screenshots. [25:05] Because it's like, look what we do inside the chat window. [25:07] So you're like, all right, how do you actually [25:08] outmarket and then how do you outsell? [25:10] And one of the reasons we launched like the Fin guarantee, [25:12] this idea that, we'll, like, [25:13] we'll pay you a million dollars [25:14] if you find some new outperforms us [25:15] because like we're trying to stress to the market this idea [25:19] that like, we actually believe in our product [25:21] to a ludicrous degree such that you should engage with [25:24] on any sort of bake off you're doing. [25:26] But I think from a marketing perspective, [25:28] it's really hard to stand out. [25:29] So all that you can really do is like rely on like [25:31] backing up your claims as hard as you can. [25:33] And obviously customer testimonials, selling is harder. [25:36] Because I think again, in the olden days, [25:38] like selling SaaS in the olden days, meaning like pre 2022, [25:42] it was kind of like our UI is nice, theirs is ugly. [25:45] Here's a feature grade checkbox. [25:46] We've got 24 checks, [25:47] they've got 17, these seven matter, we're in. [25:50] You know, like, and that was like that, [25:52] like obviously I'm skipping over seven steps. [25:53] Of course sales enablement would kill me. [25:55] But you get the basic idea, right? [25:57] And I think now selling AI [25:59] is closer to selling like infrastructure in a sense. [26:01] It's more like, it's— Our cloud [26:03] is better than their cloud. [26:04] Yeah. And our performance criteria are better. [26:06] It's like at times it might feel like, you know, Intel AMD [26:09] or at times it's like, you know, [26:10] it's our response times versus theirs or whatever. [26:12] But like you're ultimately coming into it like a, [26:13] with a battle of metrics [26:15] and sort of like our resolution rate [26:16] and our CSAT versus theirs. [26:18] But then people say, well why? [26:19] And then you have to then explain what's actually happening [26:22] beneath the surface a little bit. [26:23] So that people can actually get a bit of conviction. [26:26] Other than just like, trust us [26:26] or like please just go and try our product. [26:28] Because like it's not that easy to try Fin, [26:30] you have to still have to like turn a lot of keys [26:32] and open a lot of APIs and stuff. [26:34] So like the challenge genuinely becomes like [26:37] how do you have a sales team that's actually able [26:39] to like speak with a good degree [26:41] of familiarity about like AI. [26:43] It's funny you mentioned this. [26:44] We have this specific problem at Stripe, [26:46] which is invariably when people switch [26:48] to Stripe from a legacy processor, [26:50] they see a significant revenue uplift. [26:53] And you think like businesses are in the business of, [26:56] you know, finding ways to get more revenue. [26:58] You think they'd be really interested in this. [26:59] And we have this thing that sounds shockingly good, [27:03] which is if you just move over to Stripe, [27:05] you start immediately getting more revenue. [27:08] And it basically comes from two places. [27:11] One is conversion on the actual point of payment [27:14] that if your mobile app or if your Webflow is kind of janky [27:17] or doesn't offer the customer's preferred payment method, [27:20] or something they will abandon. [27:21] And if you just look at the abandonment rates, you know, [27:23] if you're seeing a kind of only 85% conversion rate [27:27] on that form, then obviously getting it up to 90%, [27:29] that's, you know, that's a huge deal. [27:30] And those would be very high numbers. [27:31] Most businesses would not see anything close to a 90% [27:34] conversion rate on that form. [27:35] And so there's huge improvement possible there [27:37] to make the customer kind of checkout experience [27:39] as smooth as possible. [27:40] And obviously things like Link. [27:42] Then where you're not asking people [27:43] to reenter their payment details, [27:45] that delivers a big uplift. [27:46] The second one is even crazier is [27:48] after people enter their credit card details, [27:51] frequently charges are denied kind of spuriously. [27:55] And so you bank thinks that it's fraudulent [27:58] because you know, they haven't heard of this merchant [28:00] or whatever and so they'll deny it [28:02] or they'll think it's fraud or whatever like that. [28:04] And we, through many, many years of optimization [28:08] have gotten good at ensuring [28:10] that if it is a valid transaction, [28:11] that is not denied. [28:12] So what all that adds up to is that we can make the claim [28:16] and we've seen it play out again and again. [28:17] You know, we just had Hertz switch [28:18] for all their e-commerce payments to Stripe [28:20] that when people move to Stripe, [28:21] they see a significant uplift in revenue. [28:24] That's surprisingly hard to sell. [28:25] Because everyone is out there saying, you know, [28:27] we are the thing that gives you more revenue. [28:29] And we've had the exact same thing [28:30] where despite you can have all the numbers in all [28:32] the case studies in the world. [28:33] It's just, it's hard to sell [28:35] because it's hard to believe [28:36] It's undifferentiated as a message. [28:37] I remember even when we switched back to Stripe, [28:40] either you or Patrick was saying like, [28:41] oh well don't forget to do Link. [28:42] And I was like, really? [28:43] I was like, is this really a thing that like, you know, [28:45] people have what some businesses forgotten it's credit card [28:48] or something like that and you're going to [28:49] be able to renew it. [28:50] Right. It feels unlikely. [28:50] Yeah. It just feels implausible. [28:52] But like at the same time the data is not like, [28:54] not really debatable. [28:55] I think people like to be able to explain it [28:58] to themselves and like not like, you know, I think yeah, [29:01] wouldn't, I was talking to a guy who runs procurement [29:03] and he was saying like, [29:04] you have to understand I'm just getting bullshitted [29:05] to all day. [29:07] You know, it's just all day relentless, [29:09] absolute nonsense in my face. [29:11] So like if you think that like you're like, ooh, [29:13] 65% thing is going to stick. [29:15] It's not, it's just like I take, I divide it by 10 [29:17] at this stage, you know, [29:17] and you tell me, oh, you're going to save me 2 million [29:19] in CS salaries or whatever. [29:20] I'm like, yeah, maybe in three years time, [29:23] we'll see 200 grand, you know? [29:24] Yeah. Yeah. Like that's the kind [29:25] of the default posture for a lot of these people. [29:27] I think it is just, it's like they've developed [29:29] quite an adverse reaction to like, marketing. [29:31] People do not have enough empathy [29:33] for the procurement person who just [29:34] has to endure nonstop nonsense. [29:36] Absolute garbage. [29:39] That's funny. [29:40] I mean, this kind of gets to a topic [29:40] you and I have discussed a lot, which is product marketing. [29:44] How do you effectively product market [29:47] in a world of everyone making claims? [29:49] Like one is the guarantee that you guys, [29:51] Million Dollar Guarantee, has that worked? [29:53] It's certainly, it's worked from a point of view of like, [29:54] I don't know actually know how many people are actually [29:56] are in the program right now. [29:57] But I could say that what has worked [29:58] is like— It's landed the message [29:59] of we stand behind the product. [30:01] Totally, being able to say like, here is the reason [30:02] why you can bank on us. [30:03] I think that's a strong message. [30:06] The other, I mean obviously like being able to point [30:07] to real customers or real results sort of like, [30:10] and you can say, Hey, go talk to Natalie at Nuuly [30:13] or who you know, or whatever company you want, [30:15] go talk to that person and ask them. [30:16] Because like that's their name, that's their job title. [30:18] They work there. [30:19] They're saying this, you know, [30:20] with, with 6,000 customers, it's kind of, [30:22] it gets more believable as the numbers kind of tick up. [30:25] But I so like, I guess either, you know, [30:27] you can make crazy guarantee claims. [30:29] You can like just point to a lot of successful customers. [30:32] For us, it might be different depending on your domain, [30:34] but for us, like, you know, it's not like [30:35] we can show you, [30:37] we can show you like, hey, [30:38] here's a beautiful backend product, [30:39] here's fancy reporting and all that. [30:40] But that doesn't speak to the, [30:41] the courting someone's buying, [30:43] when they're buying AI off you [30:44] is to some degree a replacement of work [30:46] that they have to do. [30:47] And the two things they care about [30:48] are how much work are you going to do for me? [30:49] And how well are you going to do that work? [30:51] And you basically need to product market [30:52] both of those things. [30:54] And it's very easy to say, [30:55] we're going to do all the work and we're doing really well. [30:58] So you have to actually really help them understand [31:00] how to appraise the scenario. [31:02] Like sometimes we put time [31:03] into actually helping people identify [31:06] when they're being lied to in a sense. [31:07] So you we will say like, hey, try this type of question [31:10] or ask them about this, you know? [31:12] Help you almost kind of teach them [31:14] to be much more conscientious buyers. [31:16] Because we know the more informed buyer, [31:17] like that suits us. [31:18] It doesn't suit people who are just kind [31:19] of like jazz handsing their way to an AI product. [31:22] But like yeah. It's a difficult one. [31:24] It ultimately, like five years ago I would've been like, [31:26] well the trick John is gifs. [31:27] You know, have you ever considered [31:28] using movies on your homepage? [31:30] You know, that's really engaging. [31:31] None of that works anymore, right? [31:32] Because I just think like what you're selling is [31:34] basically it's like an iceberg. [31:35] Like you're selling this little bit of like upfront UI [31:38] of like, here's what actually happens for your customers. [31:39] Doesn't that look nice? [31:40] And you're selling this like gargantuan pile of work [31:43] beneath the surface that is like, [31:44] hey, all of the human toil goes away if you make the switch. [31:47] Yeah, I can see that. [31:48] What are your other pet peeves [31:50] when it comes to product marketing? [31:51] Actually, do you want another pint? [31:52] Yeah, sure, sure, sure. [31:58] See again this one, [31:59] did I not let it settle for long enough? [32:00] And then you have the small head, like is the settling— [32:02] I think your one's going to work out perfectly. [32:03] Is the settling time load bearing? [32:05] Okay, so what are your pet peeves [32:07] when it comes to product marketing? [32:08] I think the thing that still kills me, [32:10] and it's still very common, [32:11] is marketers that love marketing. [32:13] So like you'll, rather than actually saying anything useful [32:15] or specific, you'll get like, you know, [32:16] forget everything you know about email. [32:18] You're like, okay, what am I buying? [32:20] Or like, you know, transformation reinvented [32:22] and you're like, cool. [32:23] Sounds like I'm going to reinvent some transformers. [32:25] Like, well, like what's actually happening here. [32:27] I think there is a general still type, [32:29] a type of thing where my screensaver [32:30] on my laptop is like literally a typewriter [32:32] where someone said, [32:33] what are you actually trying to say? [32:34] And I keep that there as a reminder of like, [32:36] just nine times out of 10, [32:38] the best marketing comes from [32:39] just writing the thing you want to say [32:40] because like the amount of docs, [32:41] do you ever get into a Google document? [32:42] Like our goal that is that [32:44] by reading this document the reader will know the following. [32:46] And I'm like, cool. [32:47] Can we just say that instead of this? [32:49] Why is there 2000 more words here? [32:50] So like, I think I guess like speaking in a way [32:53] that sounds like great to marketers [32:57] is like probably the thing that kills me most [32:59] because they don't market, especially in the AI era, [33:01] they don't really necessarily understand the depth [33:03] of what's actually happening with the AI or whatever. [33:05] There's a funny stat that Ogilvy used to quote, [33:07] which would say something like, of all the winners [33:10] of the Cannes Awards every year something like, [33:12] two thirds of them would lose their contract that year [33:14] because the thing they won the award for, [33:16] was not actually effective in market at all. [33:18] And I think there's such a repetitive pattern there. [33:20] You, like, where a lot of people, [33:23] like they will look at say a Stripe or a Linear [33:26] and they'll be like, all right, we should just do that. [33:27] And you don't really like, [33:28] The thing that people don't get [33:30] is like everything means something. [33:31] And this is like where our Eoghan, [33:33] our CEO is like so differentiated. [33:34] He just like every single decision we pick here, [33:37] what photo, what icon, [33:38] what typeface, whatever. [33:40] It all sends a message. [33:41] Are we conventional or not? [33:42] Are we futuristic or not? [33:44] And I think whenever I see folks [33:45] just like copy paste somebody else's branding [33:48] and even in the sort of like, [33:50] oh, we'll change our homework along the way. [33:52] I think what they're really doing is saying, [33:53] we don't really understand what we're doing here. [33:55] And, like, that's why a classic of this is, [33:58] you know, whenever an incubator spits [33:59] out a new batch of startups [34:00] and they all basically have right hand side screenshot, [34:02] left hand side, three bullets, sign up button, [34:05] whatever, you're kind of like, okay, cool, [34:05] but have you actually thought about [34:06] what you're trying to say to the world? [34:07] There's also a thing for startups [34:09] where they probably shouldn't look at [34:10] what established companies are doing. [34:12] Because like Stripe for so long, [34:13] we clung to making sure that we had code in the homepage [34:17] and people are like, you know, [34:18] if you want to accept credit card web payments [34:18] for your website, [34:19] like, we are the place to come. [34:21] And at a certain point, most of the relevant people [34:24] come to your page actually know that you could do that. [34:26] They experiment a little bit more [34:27] and you know, Salesforce doesn't have to hit CRM [34:30] so hard on the homepage. [34:31] Because after 20 years they've earned the right [34:33] to talk about Einstein a bit. [34:34] Oftentimes like a startup has a great idea [34:35] for a great product and they pitch it [34:37] and then like six months later, [34:39] they work on some new feature. [34:40] And in their heads, [34:41] the new feature is the big thing [34:42] that they're so impressed with. [34:43] Not realizing that 99.49 of the world [34:46] have not even heard about the original thing yet. [34:48] But there they go destroying their original pitch [34:49] by being like, and now we've got blah blah. [34:51] I'm like, dude, no one's even heard [34:52] of the original thing yet. [34:53] And here you're pitching some nuanced [34:54] take onto some extra feature. [34:56] Yeah, yeah, yeah. [34:57] Speaking of David Ogilvy, [34:58] you've read On Advertising [34:59] I assume. Yeah. Yeah. [35:00] Well it was just like, [35:02] that's such a beautiful book [35:03] where all the marketing copy in is so good. [35:05] Like the, the Rolls Royce ad of, you know, [35:06] the only sound you'll hear at 60 miles an hour [35:08] is the ticking of the clock. [35:10] But somehow that, yeah, [35:12] should be mandatory reading for all product marketers. [35:14] Absolutely. What kind of person [35:16] succeeds in product at Intercom? [35:18] When we set up Intercom originally, [35:21] like we were kind of like building one thing, [35:23] once we forked out to building many different areas [35:26] like sales, marketing and support, [35:27] I think we gave a lot of freedom to product leaders [35:29] and sort of said, hey, you own a sales product [35:31] or a marketing product. [35:32] And I think the folks who succeed there are like, [35:36] they have to have like decent taste. [35:38] And I don't mean that in some lofty abstract way, [35:40] but I mean they have to like, you know, use good software, [35:42] identify good software, [35:43] ultimately know how to like pick one out of the bunch [35:46] in a sense, right? [35:46] Like a very common interview question I ask people [35:48] is like, what apps are on your phone? [35:49] What's your favorite app? [35:50] And you might have thought someone's like, oh you know, [35:51] I never really thought about that. [35:52] And I'm like, ooh, so like, like what's your favorite song? [35:55] Like, you know, like I'm sure you care about some things. [35:56] Yeah, yeah, yeah. [35:57] Certainly if you're interviewing musicians, [35:59] they should have a favorite song. [36:00] Yeah, yeah. You'd like to think that. [36:01] But they also would be a favorite app too. [36:02] So I think like taste is a kind of prerequisite. [36:04] And then I just think like the confidence [36:06] to pick a direction and then the, you know, [36:08] we say like at Intercom we often to say shipping [36:10] is an act of, like, confidence and humility. [36:13] And what that means is like, you know, [36:15] you have to be confident enough to put it live [36:16] and then humble enough to take the slap in the face [36:18] when you've got it wrong totally. [36:19] And react to that slap, [36:20] don't like kind of be like, [36:21] no, it's not me, it's the customers or don't get it, right? [36:24] So I think like we need like high taste [36:26] and then like confidence [36:27] and then like ultimately understanding [36:29] that like in the Marc Adreessen sense, [36:31] like a product is a conversation with the market, [36:33] your launch is like your opening bit [36:35] and then you have to basically adapt [36:36] and react to what gets thrown back at you, [36:38] which might drag you in different directions. [36:39] And then you need to have the, again, the confidence [36:41] to prune certain things and like, no, [36:43] we're not building an attribution engine. [36:44] Yes, we'll take on some feedback in the CRM side, [36:46] but I think like the, you know, a lot of product managers [36:51] who don't work out for us are like a lot more spreadsheety [36:54] and like, you know, they won't take a bet, [36:57] they won't take a gamble, they won't take a stand. [36:58] They'll just be like forever mired in like, [37:00] well the data suggests [37:01] and they're just trying to hedge their bet. [37:02] Whereas it's just not the sort of company we are. [37:04] I think we kind of believe in having [37:06] an opinion about a space. [37:07] Well, and the second part of what you're saying, [37:10] if I'm hearing you right, [37:11] is the good product managers [37:14] actually can listen to the market. [37:16] They have to be able to, yeah. [37:18] And hear what, Yeah, [37:20] like I think about this a lot in the context [37:22] of tech companies where Stripe's first operating principle [37:26] is usually first. We think that actually paying attention [37:29] to what users tell us in every sense, you know, [37:33] via revealed preferences in the data [37:35] via just like when we actually have conversations with them, [37:38] we start every week with Monday morning meeting, [37:41] the first thing we do is we actually host a, you know, [37:42] in the Intercom recently, we host a customer to tell us [37:46] and give us a report card. [37:47] And you know, it's not an A, you know, [37:48] No. but, [37:49] it's seldom an A, you know, [37:50] they always have things they want to fix [37:52] and they're very pragmatic things [37:53] that they want us to improve. [37:55] There's no over complicating. [37:56] And then when we do our weekly All Hands fireside, [37:58] and we also bring customers to that. [38:00] But I feel like there's a problem [38:02] of over complexifying things [38:04] and under talking to users in Silicon Valley where yeah, [38:09] it's a bit too much celebration [38:12] of the individual kind of product vision [38:15] or a bit too much, as you say, [38:18] trying to data your way out of it. [38:20] And if you're a product manager [38:22] and you're not talking to many customers each week, [38:25] something's probably wrong. [38:26] I bring that up [38:27] because like the whole original Intercom product [38:29] was a way to talk to customers. [38:31] Like, this is kind of your guys. [38:33] Absolutely. But, [38:34] would you agree with that diagnosis, [38:35] that a lot of tech products would be better [38:38] if people simply talked to customers more? [38:40] Yeah, I mean like one of the ideas [38:44] that, like, stuck with me very early on, [38:46] it was like 2009, 2010, [38:47] I'm going this like a deep cut or whatever. [38:49] But there's a guy called Jared Spool [38:50] who's like a famous UX guy. [38:52] And I was on this thread [38:53] of like interaction design association type people. [38:55] And somebody wrote this really long, like, you know, [38:57] hey, I've shipped X and I've shipped Y [38:59] and I can't work this out. [39:00] Does anyone have any speculation [39:01] as to why people aren't doing this thing? [39:03] Even though I make it really obvious on the screen [39:05] and he just like replied on [39:05] and he was like, have you tried asking them? [39:08] And I remember like at the time I was like, right on. [39:11] Like, it was like it's, it felt like a revolutionary thing [39:13] to say, but I find like, you know, [39:15] I shared this piece a while ago, [39:16] which was like the questions I ask [39:18] in every single product review, [39:19] kind of like, so you can kind of either get ready [39:21] to meet me or like, just ideally other people can replicate. [39:24] But question one is basically [39:25] what did our users say about this [39:26] when you showed it to them? [39:28] And everyone has to have an answer to that question, [39:30] when like when we go in, [39:31] like I'm like, hey, well what did the user say? [39:33] And like, I need to like understand that. [39:34] Because if you're not actually asking your users, [39:36] what are you doing? [39:37] Like, you know, the only validation we have is the market. [39:39] I do think in The Valley does [39:40] and like, well I, I say The Valley, [39:42] but like that just basically [39:43] means in the tech industry. [39:44] There is this epidemic of like hiding behind your data [39:47] and like, you know, what can we instrument [39:49] and how many different, you know, [39:50] Mixpanel dashboards can prove to me [39:51] that this product should be working. [39:53] Just ignore the fact that it isn't or whatever. [39:54] I think there's something kind of [39:55] just fundamentally broken there. [39:58] And interesting, you say Stripe's, [40:00] first value is like users first. [40:01] I'm just curious, is that deliberately the, [40:04] I mean every word you guys say is deliberate, [40:06] but that's deliberately users [40:07] as in do you mean like pointy click-y users [40:08] or do you mean customers [40:09] or do you mean prospects [40:10] or do you mean like? [40:11] Yeah, we deliberately chose users [40:14] because we just meant the people using the product [40:16] as opposed to customers, you know, [40:19] if there's a buyer versus users. [40:21] Yeah, that's what I was going to ask. [40:22] We want to focus on the people [40:23] are actually using the product, [40:24] people who are, you know, [40:25] managing fraud within the business. [40:27] Or actually responsible for increasing conversion [40:29] or something like that. [40:30] So that was why we chose that word. [40:31] Yeah. No. It's not perfect. [40:32] It makes sense. [40:33] I often, because I just, I find oftentimes [40:36] if you want to perfect a product, you talk to the users. [40:39] And if you want to expand your market, [40:40] you talk to like prospective buyers. [40:42] But whenever I, like even in my own portfolio, [40:44] when I talk like, what are you actually doing? [40:45] Like the businesses that like are, how would you say, [40:47] like prematurely talking to more prospects [40:50] when they have a load of unhappy users [40:52] are guaranteed this kind of miles wide inches deep [40:55] like messy product that doesn't actually satisfy anyone. [40:58] But like they'll get their kind of, like, [40:59] one promise at a time. [41:01] Oh we'll build that for Johnny and Johnny will sign [41:02] and we'll build this for Jenny and Jenny will sign [41:04] and like at no point do they have one happy customer. [41:07] What they have is like a marauding churn bomb [41:09] of a user base. [41:10] Yes. It's funny you say that. [41:12] It feels like many tech companies [41:14] over rotate on sales feedback, [41:17] which will by definition be from the marginal user [41:20] and they're marginal, in two senses. [41:22] So you have all your existing users, you know, [41:23] that you dancing with the girl that brung you over here. [41:26] And then you have this future potential user [41:30] who firstly by virtue of the fact, [41:32] they're not already using you, [41:33] maybe they're slightly outside your wheelhouse [41:36] or like the use case isn't perfect or something like that. [41:38] So maybe they're not quite [41:39] as good a fit as your existing customers. [41:41] And then also by virtue of the fact [41:43] that they have a whole existing way of doing things [41:46] when they migrate over to your product, [41:47] they'll do so in a bit worse shape of integration [41:52] where maybe they'll only use one of the four features [41:55] or you know, not everyone in the org will be bought in [41:59] versus the people who grew up on your product. [42:01] And so maybe just restating what you're saying, [42:03] I'm always struck by people are way too focused [42:06] on we tried to win this big new shiny enterprise account [42:09] and they didn't have, you know, we didn't have feature X [42:11] and so therefore we're going to develop feature X [42:12] as opposed to you've all these users [42:14] who grew up in your products [42:15] and really like it, but they wish you would fix A, B and C. [42:20] And just the nature of the fact [42:21] that sales gets more airtime [42:23] than account management essentially— [42:25] Absolutely. Means [42:26] people really misprioritize [42:27] where they spend their time. [42:28] Yeah, it's, and people take ARR for granted [42:30] and think that like [42:31] that Net New Revenue is that is hard, right? [42:33] Like, and I think one of the things [42:34] that we see a lot of [42:35] is like in terms of like, you know, [42:39] working out for your current customers, [42:40] like we use the phrase permission to innovate [42:43] and permission to expand in Intercom, [42:44] which is basically like you permission to innovate [42:46] when your product's pretty good. [42:47] Like, as in yeah okay let's work on V3. [42:49] But like is V2 in good condition? [42:51] Like, or you know, and then permission to expand [42:53] is like V3 isn't actually that exciting. [42:55] Everyone's happy with V2, [42:57] now, I think we can try to do something new for customers, [42:59] like expand our share of wallet or whatever. [43:01] But I think a lot of people try [43:03] to solve revenue growth with like aimless product expansion [43:07] to just try and increase the share of wallet for the people [43:10] who are stuck with you. [43:11] And then they convince themselves they got PMF [43:13] or like, you know, that they have, like, [43:14] some sort of a good product because they're kind of, [43:17] like foie gras style force feeding new features [43:19] down the throat of their trapped users. [43:21] And they're like, you know, we're doing great, [43:23] but they don't realize what they're actually doing [43:24] is making their current product so messy [43:26] that like they're destroying the hope of future revenue. [43:28] Because any like, yeah, [43:29] you can force your current customers [43:30] into whatever upsells you have or whatever, [43:33] but your product marketing along the way [43:34] is getting really difficult [43:36] because all these features don't make sense. [43:37] And they're just kind of like, [43:38] you've tried to do this like land and expand thing, [43:40] but you're actually just expanding [43:41] and it's no landing happening in the new product. [43:44] And then you end up [43:45] trying to twisting yourselves in knots [43:46] and a lot of startups like, you know, [43:48] Jason Lemkin came up with this thing like, [43:49] from zero to one's impossible, [43:50] from one to 10 is hard [43:51] and from 10 to 100 is inevitable. [43:53] Like I think a lot of people, [43:54] I don't think that's proven out to be true [43:56] as much as it was back when he said it. [43:58] I think a lot of people get stuck in some sort of glue [44:01] around somewhere around the 10 million mark [44:03] where they don't know how to like get the next 10,000 logos. [44:06] So they just try and milk the revenue [44:07] out of the existing customers. [44:09] True, like just forced product adoption of new stuff. [44:11] Like, you know, and that you see a lot, [44:12] here's your Copilot, I know you didn't want it, you know, [44:14] but here you go, [44:15] like that type of thing. [44:18] Des is describing here [44:19] how they've transformed Intercom from a SaaS product [44:21] to a frontier AI business. [44:23] And to do so they had to pivot not just the product [44:25] but the monetization model as well. [44:28] Because inference costs are so significant. [44:30] AI powered companies tend to charge based on usage [44:32] rather than just allowing for unlimited plans. [44:34] It gets complicated and really multidimensional, [44:37] very quickly. [44:38] Now fortunately, complicated and multidimensional [44:41] is what Stripe Billing specializes in. [44:44] Our usage based billing engine can ingest up [44:47] to a hundred thousand events a second, [44:49] a hundred thousand events a second. [44:50] So AI companies can monetize products [44:53] based on real time customer usage. [44:55] We're powering consumption billing [44:56] for companies like Figma, Cognition [44:58] and tons of other leading AI applications. [45:01] Our usage-based billing platform has grown [45:02] 145% so far this year. [45:05] So whether you're changing your business model like Intercom [45:07] or starting a new product from scratch, [45:08] your business strategy should dictate the billing system [45:11] and not the other way around. [45:13] For usage-based billing, check out Stripe Billing. [45:17] As you think about the prototypical $10 million [45:20] revenue B2B company. [45:22] What are the common mistakes you see [45:24] and what do you think the actual path [45:26] that more of them should follow is? [45:27] I mean the biggest mistake [45:30] is like not aligning your fundraising with your TAM. [45:33] I think a lot of folks, [45:35] like we got a little bit over convinced [45:38] during the era of cloud [45:39] that every business had a right [45:40] to be like a unicorn. [45:42] And so there's a lot of businesses whose idea [45:44] was like totally fine, [45:45] but actually they should have gone on [45:47] and base camped it more so than they did [45:49] because they've raised on the assumption [45:51] there's an easy path to like [45:52] hundreds of millions of revenue. [45:53] There should be more small, [45:54] profitable 30 billion revenue companies. [45:56] Well, yeah, exactly. [45:57] And I think a lot of these businesses [46:00] would be great if only they didn't raise 20 [46:02] and tell their investors [46:03] that you're going to like easily [46:05] be worth a billion or whatever. [46:06] So think like there's a genuine mismatch there [46:08] where I think people have like overstated, [46:11] like how big this idea could get. [46:12] You know, as in, hey I know we, all we do [46:14] is like time tracking for dentists in Delaware, [46:16] but like, believe me, [46:17] we're going to be a billion dollar company. [46:18] And you're like, okay, well one of your restrictions [46:20] is going to have to break here. [46:21] So that's one problem, [46:22] which is more like kind of business model like [46:24] and venture ambition. [46:25] The other stuff I see is like, [46:26] it is kind of like not focusing enough [46:29] on the thing you're the majority of your customer's value. [46:31] Like it's easy to say, [46:33] the best business in the world is like one line of code [46:35] that all users execute and you sell it to all users, right? [46:38] They're like, you know, the sweet spot. [46:40] It's hard to do it in a differentiated way. [46:41] Because you know, obviously people learn [46:42] that line of code and that's where you, [46:44] I think a lot of these like horizontal products, [46:46] they say something like illume or whatever, [46:48] they're like a brilliant, [46:49] they're a piece in everyone's workflow, [46:51] but they're no one's like end to end workflow. [46:53] I think they can do well too. [46:55] But I think the challenge is like when people, [46:58] rather than nailing a specific small thing, [47:01] come back to kind of the earlier point, [47:02] like rather than like saying, [47:03] hey let's get really good to X [47:04] before we go beyond, [47:05] when they kind of prematurely expand, [47:08] I think they forego all opportunity of, like, [47:10] kind of being the best. [47:11] And if they picked a really important area first, [47:14] then they don't say it loud [47:15] what they're saying is, [47:16] it's okay to not be the best [47:17] at the most important thing we do. [47:19] And I remember, like I remember, [47:20] I think it was 2012, I was in your office on, [47:22] you know, in and I remember at the time [47:25] it wasn't obvious to me that you weren't going to expand [47:27] and do some sort of like, you know, [47:28] peer to transfers and compete with PayPal. [47:31] And I remember like you guys had the discipline, [47:32] you're like, absolutely not. [47:33] We care about helping businesses charge [47:35] and like there's a real harsh discipline you need to have [47:38] to like just basically say no [47:39] to all of the surrounding opportunities. [47:42] And I think a lot of people, [47:43] that discipline's the first thing to go, [47:45] when you hear about competitors, [47:47] you're going to hear somebody else encroaching [47:47] on your space. [47:48] You're going to have this really weird broad view [47:50] of all the things you do. [47:51] Like I know we just do like whatever it is, [47:53] gifs and screenshots. [47:54] But actually when you think about, [47:56] we're a global creativity platform. [47:57] And like they have this premature view of themselves [47:59] as being massive and then they feel like [48:01] then they go and raise off that [48:02] and they need to expand into that. [48:03] But I think like at the core of every great business, [48:07] every great SaaS business, [48:09] but in the future AI business, [48:11] is something that they're just truly world class at. [48:12] And it's like, it's not some sort of 80/20 trade off. [48:15] They've just basically said, [48:16] we'll be better than anyone at this, right? [48:19] And like you take like Linear, [48:20] it's basically like they have literally [48:22] the most world's most efficient UI for like, [48:24] for product management [48:25] and they have all sort project management [48:26] and they've just gone really deep into, [48:28] like all of the surrounding adjacencies [48:30] you would need to actually do that job really well. [48:32] Figma is just an amazing creative collaboration machine. [48:35] Like everyone who like has done really well, [48:38] they've picked one thing and just gone really hard, [48:40] really deep, really far on it. [48:42] They haven't like prematurely blown up [48:44] and gone in seven different directions. [48:46] And I also think there's an, [48:48] a weird celebration in The Valley of [48:52] Act Two, like The Valley's obsessed with finding [48:57] second acts that are totally unrelated [48:59] to the first business. [49:01] Like the number of people who bring up like, you know, oh [49:03] and we like invent like an AWS. [49:04] It's like okay, you need to use a non cliche example [49:07] if you're going to make that argument. [49:08] And the flip side is, you know, [49:09] you're mentioning Figma, which think is a great example [49:11] where that market proved to be way bigger [49:14] than people might originally have thought, [49:17] you know, my favorite example of this is NVIDIA, [49:20] where they're the world's largest companies [49:22] and they started in the 1990s making GPUs. [49:25] If you're an investment banker trying to, you know, [49:27] make a case for how NVIDIA can be a really big company, [49:30] maybe you'd say, oh well [49:31] and we can expand into maybe we'll actually, you know, [49:34] make our own gaming rigs [49:35] or you know, it was all gaming at the time originally. [49:37] You know, maybe we'll make gaming consoles [49:38] or you know, we'll expand to some larger markets. [49:40] Whereas actually what transpired [49:42] is it turns out the GPU market [49:44] is quite a lot bigger than people thought. [49:46] And being, you know, the best at GPUs [49:49] is a really valuable prize. [49:51] And you can't rush it, you know, [49:53] it'll emerge and it's, but yeah. [49:54] And they could have killed themselves [49:55] if they had gone in every other direction like, right. [49:57] And they would've lost their edge in some sense. [50:00] Figma is a great example for me of like permission to expand [50:02] and not like they literally nailed to a point [50:05] of like no credible competition. [50:07] This idea of like just, you know, [50:10] the Photoshop killer basically, let's just say. [50:12] And now they can talk about like slides [50:13] and text to app builders [50:15] and like every other dimension they want to go. [50:17] And everyone's like, yep, that's great. [50:18] Because you guys make great software, right. [50:20] I think you have to first be known for, like, [50:22] I'm trying to think like if Stripe launched [50:23] a payroll product, [50:24] it would carry the brand of Stripe in the sense [50:27] of like being, well it's probably really good, [50:29] really reliable, really fast. [50:30] It probably has really nice API, [50:32] it probably works really well with Workday. [50:33] You can almost impute like all the ideas [50:35] that will be carried into it. [50:36] And I just think like you have to get to that point [50:38] before you have permission to make that bet. [50:40] Like obviously it's a lot easier, [50:41] it's like stablecoin or whatever. [50:42] But like what kills me is when I'm like, you know, [50:45] I don't even want to like name a weak SaaS company [50:47] but like pick your favorite like, you know, [50:49] mediocre SaaS company [50:50] and any think like, is there any direction you would [50:53] allow them expand in your head? [50:54] No, [50:55] like this is your short answer. [50:56] Yeah, that's interesting. [50:57] Who are you really excited [50:58] to adopt new products for [51:00] versus who are you steering clear of the new products? [51:02] Like if Linear launched a, I don't know, [51:03] let's just say a source control tool, be like, [51:05] yeah it's probably going to be really, really good. [51:07] And like I remember Seth Godin has this hilarious point [51:10] where he talks about like the value [51:11] of brand once it's, once it's like weaponized [51:14] and he describes like Nike and Hyatt hotels [51:17] and he says if Nike opened a hotel, [51:19] you can close your eyes and see it. [51:20] You know exactly what the corridors are going to look like. [51:21] You know, the vibe of the whole place, you know everything. [51:24] It's going to be, if Hyatt launched a sneaker, [51:27] you're like, what? [51:29] You know, and it's just, that's the difference though. [51:30] Because if Hyatt has a logo [51:31] and Nike is a brand and that's the difference. [51:34] A version of this actually maybe quite literally is, [51:36] I know saw Equinox launched a hotel would a pretty good idea [51:39] because the design center there for the hotel [51:43] is you just want to be able to get a good night's sleep. [51:46] And it's funny how that's like a differentiated [51:49] product— Exactly. [51:50] In the hotel space of like, [51:51] don't gimme any of that other shite. [51:53] I just want be able to go to my room [51:55] and not have like a super loud noise outside the window [51:58] or like weird light coming into the room. [52:00] You just want to be able to get sleep. [52:01] Precisely. Yeah. [52:02] Yeah. I thought that was funny. [52:04] And you're mentioning kind of Stripe's expansions [52:06] and so this may be a good segue into your [52:10] pricing model change. [52:11] You guys are the poster child [52:13] for the move from per seat SaaS pricing, [52:16] the old way of doing things to usage based pricing. [52:20] Maybe you can describe a little bit about that [52:23] and then how you implemented it [52:23] and what you're doing with Stripe. [52:25] Yeah, sure. [52:26] Our pricing journey is long and complex [52:29] and a lot of your listeners are, [52:31] viewers will know. [52:32] Intercom pricing is a charged topic? [52:33] Yeah. At Intercom? [52:34] Not anymore. [52:35] You know, we've turned a corner, let me just back up a bit. [52:38] So like when we had like two diverse [52:41] or product strategy, we were trying to do sales software, [52:42] marketing software, support software. [52:44] And like sales software is typically sold [52:46] based on leads created, [52:47] marketing was charged by like [52:48] how many contracted people you want to send [52:50] and support was sold by seats. [52:51] So we had this extremely, [52:53] let's just say detailed [52:54] but like unnecessarily complex like pricing setup. [52:57] And we kind of, we lied to ourselves and said don't worry [52:59] because there's always going to be a human [53:00] to help people navigate this [53:02] because you're never going to have to self-serve this. [53:03] But like ultimately people were just like, [53:04] I have been refreshing this for like seven minutes [53:06] and I can't understand the word of it. [53:07] And that was just one of the few things we got wrong [53:09] in our kind of first move up market. [53:10] When Eoghan return, one of the decisions he made [53:12] was just like, hey, we need to like, you know, [53:14] sort our pricing. [53:15] And we handed back, [53:17] like truly handed back I think about 50 million of revenue. [53:21] I think it's— [53:22] Was that controversial like with the board, [53:23] with investors? [53:25] Like we had support for it. [53:26] I think it was like people don't really under... [53:29] people massively underestimate what it means [53:31] to have a really happy customer base. [53:34] It's because word of mouth doesn't have like an attribution [53:36] or a UTM code, if you know what I mean. [53:38] So like they don't understand [53:39] how to think about like happy customers. [53:41] So making the decision to basically like, you know, [53:43] kind of standardize on an easy to understand pricing [53:46] that's like fair, transparent, predictable, et cetera. [53:49] That was the first decision that we made. [53:50] This was before AI, right. [53:52] And that was like us returning [53:54] to Stripe was a large part of that. [53:56] In fact, as a small segue, like [53:58] I think like a couple years prior I'd said to you [54:00] or to Patrick like, hey you guys should actually build [54:02] as part of your product offering a pricing page creator. [54:06] I think at the time though, [54:07] I probably got one of those like thumbs up replies [54:10] or something like that. [54:11] I was like, yeah, whatever Des. [54:13] But I think— It's on the list. [54:14] I think you've done it since though. [54:15] We have— Yeah, yeah, yeah. [54:17] But like my thinking at the time [54:18] was basically some version of this, right? [54:20] You need to not let your customers [54:21] go wild with pricing, right? [54:23] You need to like actually put some sort [54:25] of guardrails onto how they think about pricing. [54:27] Otherwise they're going to go and invent stuff [54:29] that you guys don't support. [54:30] And then they're going to [54:31] move all your business logic across— [54:31] Pricing is writing checks. [54:32] Yes, yeah. Exactly. [54:34] Once, like, and I remember like sitting [54:36] in a Stripe mini All Hands [54:37] or whatever explaining like that like, [54:38] hey, now none of our business logic runs through Stripe. [54:41] I think either you or Patrick was saying like, [54:42] some version of like, that's not a good thing. [54:43] I think, you know, generally speaking, [54:45] my advice to any software company is like, [54:46] don't afford yourself too many degrees of freedom here. [54:48] Because you'll actually cripple yourselves in a quagmire [54:51] of complexity that will take many years [54:53] and ultimately like many tens of millions of dollars [54:55] to get out of. [54:56] It's a weird failure mode [54:57] that every single company falls into, [54:59] which is you start signing deals [55:01] that have some super creative pricing structure [55:04] and the customer negotiates A, B, C, D and E. [55:06] And it's like it is built in Microsoft Word, [55:09] but it is actually, it's just not built practically in code [55:13] because it just exists with this one customer. [55:14] And it may not even be possible [55:16] to building in code. [55:17] Like sometimes the, you know, [55:18] it's kind of ambiguous or it's like, [55:20] and if in the subsequent year this happens, [55:23] then we go back to the prior year and we do adjust. [55:25] I call Ricky and he decides with the discount is. [55:27] Exactly, there's like a time travel component [55:29] to the whole thing. [55:29] And then they obviously have this, again, [55:32] we see all customers running into it, [55:33] these kind of manual billing issues [55:35] where there is a guy who has to deal [55:37] with all these like contracts, [55:38] that were agreed during the sales process. [55:40] And so as you're saying, [55:42] an opinionated billing engine is actually pretty important, [55:46] assuming you believe that billing should be automated. [55:48] If you're happy, like manually getting out the calculator [55:51] for every single customer month [55:53] then that's fine. [55:54] And probably got a large deal desk function [55:56] and doing all the work behind the scenes. [55:58] So anyway, that was like the first piece of our pricing. [55:59] And then the second piece was obviously when we launched Fin [56:01] then it was like, hey, how do we charge for this? [56:02] Because we're replacing seats and like at the time [56:05] it hasn't improved out this way fully, [56:07] but like at the time Fin looked like it was going to [56:08] be pretty [56:09] cannibalistic to Intercom. [56:10] Because it was like, hey, if we're automating at the time [56:12] what we thought was like 25% of your revenue, [56:14] we assume that means like 25% less seats in the future. [56:17] Or at the very least what it would likely mean [56:19] is the growth rate or the AONR of the seats model [56:22] will be affected [56:23] by the fact that Fin doing all the work [56:24] and now at 65%, you'd expect it to be even further true. [56:27] So it was like, hey, how do we charge [56:28] in a way that makes sense? [56:29] And we're, and then also like, [56:30] how do we be aggressive? [56:31] Like as in we want really wanted to like, [56:35] like put a mark on the market that sort of says like, [56:37] you know, we're like very AI forward. [56:40] And I think what Eoghan and then Darragh came up with [56:42] was just like, hey, like let's just charge [56:44] per literal resolution. [56:45] Like every time we do work, we charge, [56:47] every time we don't, we don't. [56:47] And this is at a time when like AI was like, [56:49] you know, margin negative and all that sort of stuff. [56:50] We were still working out [56:51] how that whole world plays out. [56:53] Today, we're, you know, we're, already, [56:55] really happy with our margins but at the time, [56:57] it was like, hey, this could— [56:58] It was a bet. Yes. [56:59] It was deFinitely a bet. [57:00] We made the decision [57:02] and I think the market responded really well [57:04] because I think it was very clear that like, [57:07] the single statement of like, [57:09] we only get paid when we do work. [57:10] We don't get paid when we don't do work. [57:12] It's from the same vein as the guarantee, [57:14] which is just like, [57:15] that's how you know that we believe in our product [57:17] and that's how, you know, our product works. [57:19] And it's been copied a million times since. [57:20] But I think like the actual decisions [57:23] that we made in the run up there [57:24] were really, really important from a point of view [57:25] of backing up our claims. [57:27] And then obviously for a lot of our competitors, [57:29] they were like, ugh, you know, [57:30] we don't really have a great way to respond to that [57:32] because either our product doesn't work [57:33] or we're kind of hooked on these really [57:35] expensive resolutions. [57:36] And we were totally the cat amongst pigeons there [57:38] which has been like really well received by the market [57:40] and our customers generally do love it. [57:42] It is funny though, like you still get people being like [57:44] 99 cents ridiculously expensive. [57:46] And then, you know, you're like, why do you think this? [57:48] And the answer is because some version [57:50] of we don't know how to calculate cogs, you know? [57:52] Yeah, yeah. [57:53] Yeah, how much of the humans costing you? [57:55] Yeah, exactly, and how much is their office costing you [57:57] and all the other stuff, you know? [57:58] Yeah, yeah, yeah, but people like certainty, [58:00] how do you get them okay with the variable component? [58:02] You know, you can, obviously, [58:03] you can contract out whatever you want, right? [58:07] Like, but what we offered people is like, hey, like, [58:09] you know, most of the time people have like [58:11] at least one or two years look back on like what, [58:12] you know, like with customers who spike for tax seasons [58:15] and our customers who spike for Christmas or whatever. [58:17] And we can basically say like, hey let's, [58:19] like let's contract your base rate [58:21] and let's talk about overages for the months you need it. [58:23] And that's like, that totally works. [58:24] What we're basically saying is like, yes, [58:25] you don't have predictability in the sense [58:27] of it not being fixed, [58:28] but you have like, you can model it based on [58:30] what's happened in your history. [58:31] And it's only really like brand new startups [58:33] that don't have a clue what's going to happen, [58:34] but like they're not usually [58:35] worried about this. [58:36] Yeah. Yeah. [58:37] So you're using a relatively new product, [58:38] Stripe usage based billing for this, [58:39] how has that, you migrated from Zuora for that? [58:42] How has that process been? [58:44] Yeah, I mean I would say, just to go back to your earlier, [58:47] like we afforded ourselves too much complexity. [58:51] We kind of codified that complexity in Zuora. [58:53] I guess the best way to describe it's, [58:55] we just twisted ourselves in knots, you know, [58:56] and it got to a place where we actually, [58:58] we ended up like Ciaran Lee who is our CTO, here, yeah. [59:02] He ended up actually returning [59:03] to the company one mission, which was like, [59:05] I am going to— Fix billing. [59:08] Yeah. Exactly. [59:08] To fix this. Right. [59:09] And like it worked, right. [59:10] But it was like, it was a substantial amount of work [59:13] to like unwind so much. [59:15] And then like they were like to kind of like deconstruct [59:18] so many of these like ala carte Microsoft Word style deals [59:22] into something that was a go forward, acceptable whatever. [59:24] And then obviously moving towards like a clean, [59:27] transparent seat based pricing [59:29] and then just layering on a usage base on top, [59:31] was actually pretty simple in the greater scheme of things [59:33] like, and like all the stuff that, you know, [59:35] we needed, you guys were ahead of us on like, you know, [59:38] discounts for volume, et cetera. [59:39] All the sort of obvious stuff people would push for. [59:41] Yeah, yeah, yeah. [59:42] Is this just where pricing in this new world goes? [59:45] Because obviously no one buys labor [59:48] on an unlimited, [59:51] you know, basis and at least for the moment, [59:54] the inputs of AI do actually scale with usage, you know, [59:57] for a significant basis. [59:58] And so it feels like you have to have [60:00] some usage based pricing. [60:01] This is certainly the bet we are making [60:02] again, the reason that billing, kind of, [60:05] the top thing they're thinking about [60:06] is kind of making billing work well in the usage based world [60:08] is it just feels like many products [60:10] are becoming much more expensive to serve [60:12] and therefore have to have usage based component. [60:14] But is this permanent? [60:17] Or I don't know, does the AI get cheap enough [60:19] that maybe we go back to unlimited plans or? [60:22] I don't know, I don't know if unlimited plans will ever, [60:24] well I don't know, I, here's how I think about it. [60:27] Like I think ultimately all AI has like two vectors, [60:30] just like how much work are you're doing [60:31] and how well are you doing it? [60:32] And the volume of work you're doing it, [60:35] it's almost, well actually both of them [60:37] are going to be like proportional [60:38] to how many tokens you're burning, whatever. [60:40] So like you're going to want to factor that in, [60:42] especially if you're a consumer app as well, [60:43] we're going to go nuts. [60:44] So I think like you have to like have some, you know, [60:47] I'm not a fan of like cost plus pricing, [60:49] But like, but it does place a kind of a lower band [60:52] on what you can do here, [60:53] which is just like, hey, you know, [60:54] unlike SaaS you are actually sending money [60:56] out the back door as well. [60:58] So I think you have to have something [60:59] that's proportionate to how much work you're doing. [61:01] And then I think aside from that, [61:03] you have to charge consistent with like, [61:04] how much work are you displacing? [61:06] I think that's where you can sort of say, hey, like, [61:08] you know, for us anyway, like if you take an average person [61:11] who sits in a seat to do customer service, [61:13] if they do, like, let's just say, [61:15] they do 20 conversations a day, [61:16] that's what, 400 conversations a month. [61:19] When we were thinking about how we charge, [61:21] we're like, hey, well if that person does 400 a month, [61:24] you know, Fin does 65% of that seat, [61:26] we're still up because we're only charging like whatever, [61:27] $90 for the seat. [61:29] So from our point of view, [61:30] it was like an obvious, an easy swap. [61:32] I think for a lot of businesses it might not be, [61:34] if your AI doesn't work or it's spurious [61:36] or like its value can't be articulated. [61:38] Like, isn't it cool that you can now dynamically [61:40] summarize a GitHub issue or something like that. [61:41] You're like, cool, [61:42] I don't know how much people will pay for that. [61:43] Because they don't know either. [61:45] Or like, hey, you can now generate random graphics [61:48] in your, like in your newsletter tool, you're like, okay— [61:50] It's like vitamins versus painkillers. [61:51] Yeah. Exactly. AI pricing. [61:53] Yeah, and like, and it's specifically like in this case [61:56] the painkillers have a very strict, like if we don't do it, [61:59] a human's going to do it and we know exactly what they cost. [62:01] And the vitamins doesn't have anything approximating that. [62:04] So not only is it a nice to have, [62:05] it's like I don't even know what it's worth, [62:06] like, you know, I saw a while ago, [62:08] like someone said, like when, [62:09] you know, when Studio Ghibli came out [62:10] and everyone was like using like that someone said like, [62:13] hey, the fiverr.com equivalent of, [62:15] of all these things would've been like trillions of dollars. [62:18] And you're like, right. [62:18] But like no one was ever going to spend that. [62:19] So there's no sane way to actually talk about [62:21] what what actually happens here. [62:23] I think it was Byrne Hobart who said that, you know, [62:24] when you're tied to business outcome, [62:26] that business outcome is usually done by humans. [62:27] I think it's going to be really, really easy to join [62:30] to make a business case for saying swap this over [62:31] to AI, it's better, faster, cheaper. [62:33] I think when your AI is not tied to business impact [62:37] or is like debatable in quality or whatever, [62:39] I think, you know, you end up with these people [62:40] who are just like, oh, let's just stick a tenor [62:42] on the seat and see what happens. [62:43] You know, so it's like, hey, [62:44] you can have a normal seat or an AI seat. [62:46] And then you're kind of like, [62:47] I hope no one uses the AI too much. [62:49] You're permitting yourself to build weak AI stuff [62:52] if you do that. [62:53] Because you're not pushing selves to say, [62:54] hey, we need to articulate the value [62:56] of each incremental usage here. [62:57] Well, when you talk about this AI pricing dynamic, [63:00] one thing that really strikes me is just [63:02] how fast AI companies grow from a revenue perspective. [63:05] So I just saw Mati from ElevenLabs. [63:09] We actually had a great session at our customer event [63:11] in London, but he tweeted [63:13] that they've just passed 200 million in ARR [63:15] and that's two years after founding on it, [63:18] maybe three years after founding. [63:20] But you know, in my day, businesses didn't do that. [63:22] And it's really striking for me, [63:24] how somehow they seem [63:26] to climb the revenue ramps much quicker. [63:28] I know, like, I mean. [63:29] You guys with Fin is another example. [63:30] Yeah, yeah, for sure. [63:32] We forecast like Fin will be a hundred million [63:34] probably early next year, whatever. [63:35] And like back at home— [63:37] Yeah. From when, like what? [63:38] Starting from? [63:39] I don't know, probably about two years, something that like. [63:41] Yeah. Yeah. So two years to a hundred million in ARR. [63:42] Like when we started [63:44] and probably when you guys started, [63:45] like it was like that was the threshold to go public. [63:48] Exactly. [63:48] It used to take a long time to get to one million ARR [63:50] Thought it was [63:51] like seven years. It used to get [63:52] a lot of money. [63:53] Yeah, exactly, back in the day. [63:54] But yeah, it's the acceleration is more there. [63:57] There's like Mati's, you know, [64:00] Eleven's a fantastic product, right? [64:01] Like, and it's a great example It's very impressive. [64:04] Of like, there's like four kind of, [64:05] if you like, horsemen of AI products that I observe, [64:08] whatever I'm investing. [64:08] And it's rare you see all four. [64:10] But the things you want in an AI startup [64:12] is kind of, one, is like, is the revenue backed by usage? [64:15] Like, and that's why I love usage based revenue, right? [64:17] As opposed to like, you know, shelfware or pilotware. [64:20] You know, like, hey, we sold it [64:21] to two guys in the corner [64:22] and they're going to put it live someday. [64:23] So like, you want, you want revenue backed by usage. [64:25] You want the usage tied to a real business impact. [64:27] So it has to be mission critical, you know, as in like, [64:29] like if you're building a phone product on top of Eleven, [64:32] like if that doesn't work, that's really bad. [64:34] So they, so it's critical. [64:36] The third one is obviously you want deep AI, [64:38] deep differentiated AI can't be a thin wrapper. [64:40] And then the fourth one is like you, actually, want [64:41] positive unit margins and all this, [64:43] or at least a clear path to positive unit margins [64:45] if you're not there already. [64:46] And I think when you look at so much of the AI landscape, [64:49] you'll see so few businesses that have all four, [64:51] it's such a rare sort of area to be in, [64:54] to be like actually like, hey we're doing a real thing, [64:56] with real differentiated AI, [64:57] it really matters to businesses [64:58] and we're making money off it. [64:59] Most of the time when you hear about these, [65:01] well we went from zero to 6 million overnight, [65:03] it's kind like, you know, to generate JPEGs of a smurf [65:06] or whatever and you're like, all right cool. [65:07] I'm not sure that's going to renew, you know? [65:08] Yeah, that's the simplest [65:10] AI investing framework I've heard. [65:12] I'll tell you why it's simple. [65:13] Because you're going to make me write no checks. [65:16] So like, I guess I'd say most of my, [65:19] most of the AI companies I've invested [65:21] in probably three of four, I'd say. [65:23] The only one I might quibble with there. [65:24] I think that's very good for staying out of trouble. [65:28] And this is where, you know, I tend to push back [65:30] when people are saying, oh it's an AI bubble. [65:31] It's like, I don't know, I think people are happy [65:33] with the tokens they're buying. [65:34] You know, I think like there's a lot of tokens happening [65:36] and just generally they seem [65:37] to be delivering useful outcomes that customers [65:40] because they're actually delivering value [65:41] on the customer service side [65:42] or people enjoy their Midjourney adventure, [65:44] you know, but like people are getting value [65:46] from the products. [65:47] I used it, so it's a pushback [65:48] that like doesn't like it. [65:49] I was going to push back on number four, [65:50] which is— Positive unit margins. [65:51] Positive unit margins [65:52] because just aren't the underlying costs, [65:54] like again, when you guys started Fin it sounds like [65:58] We were losing— Underwater money on it, [65:59] Yeah, yeah. Right. [66:00] But then just pretty quickly it right sizes [66:02] as you optimize it. [66:03] And so couldn't one be too focused [66:05] on the current implementation. [66:06] Yeah, I mean, [66:08] this is a conversation we have internally [66:09] with our CFO quite a bit actually. [66:11] Like, we're like, because we're good. [66:12] You can imagine be the kind of thing the CFO would like. [66:14] Yeah, yeah. Hey Des, [66:15] you have five minutes, [66:17] money is losing— That's exactly what's last. [66:18] Quick chat. [66:19] I can't help but notice the team has done this like [66:21] preemptive like loading or whatever. [66:23] It's costing us a shitload of money. [66:25] So like what's my counter to that? [66:27] I guess like, I prefer if the path towards [66:31] profitability isn't just like the, you know, [66:34] OpenAI is going to figure this out for me, right? [66:36] Like an interesting way I'd say this, [66:37] like with Fin, for example, obviously [66:40] our profit goes up when we are firing less dead tokens. [66:43] A dead token being we've generated an answer [66:46] and it wasn't right. [66:46] So we can't charge money for it. [66:47] Like if you're like say like guessing [66:49] the next line of code, right? [66:50] Or like tab to auto complete the next line of code, [66:52] if like five of six of those is wrong, [66:55] I don't know if you're ever going to get bailed out by, [66:57] because you're basically five, sixths of your costs [67:00] is like, you know, it's not something you can resell. [67:02] So like there's a question there of like [67:04] how much of your tokens are actually generating a thing [67:06] that a user wants, [67:07] independent of what you charge [67:09] for as long as the user wants it, [67:10] I think you're always in good condition, [67:11] whereas if you're like burning a million tokens to find one [67:14] and that one, you're never going to be able [67:15] to recoup your cost. [67:16] Or at least, you know, I'd love to see your telemetry [67:18] to make sure that you actually have thought this through. [67:20] I suspect you haven't, you know. [67:25] 45, that's not bad. [67:27] Yeah, but I was very impressed by the multiple, [67:29] clearly deliberate twenties. [67:31] So I'm curious about the co-founder dynamic you guys have [67:34] across all the co-founders where, let me try this on. [67:37] My sense is that people want [67:38] to have a subject matter area based explanation [67:43] for co-founder, you know, collaboration where, you know, [67:45] I'm the technical guy [67:46] and they're the business guy, whatever. [67:47] And in my experience, or at least, with me and Patrick, [67:50] it's much more personality, tension based [67:53] where I would say he's more visionary and expansionary [67:57] and I'm more, well, you know, we have food at home already, [67:59] you know, you gotta finish the products [68:00] that you already doing. [68:01] Yeah. Yeah. Or you know, I'm more frugal [68:03] and he always wants to spend all our money [68:05] or you know, whatever the tension you're describing is. [68:07] And then there's a useful, well one, it's useful [68:10] to have someone to be able to, you go bad mad by yourself. [68:13] Trying to solve all these fairly knotty problems, [68:15] but also a good company strategy probably exists [68:19] at the intersection of those tensions. [68:22] Does that describe your relationship with your co-founders [68:24] and what would you describe as the personality tensions? [68:27] I mean, we're definitely all different. [68:29] A lot of like key things we all agree on, [68:32] Eoghan would be like a, like first and foremost, [68:35] I mean he's like, he's a very strong CEO. [68:37] He is very decisive and he is very brave [68:40] is the best way I could describe it. [68:41] An interesting thing, like when he returned to Intercom, [68:43] one of the things he did was make, [68:44] like basically rebuild the culture. [68:46] And one of the things he focused on was like [68:47] resilience and open-mindedness. [68:49] You know, we didn't know AI was coming. [68:51] He didn't know AI was coming. [68:52] But like to be able to like react to AI [68:55] requires a lot of manic pivots. [68:59] Zero certainty and ultimately conviction bets. [69:01] And I can't think of somebody better to do it. [69:04] That wouldn't have been me. [69:05] Like, not in a million years. [69:06] I would be like, like even being as AI pilled as I am, [69:10] I still would've like, and I can even, [69:12] I even look back at my own performance in that period [69:14] and I'm like, you know what, I wasn't brave enough. [69:16] Like of the things Eoghan pushed for [69:17] was the idea of creating the team Fin, which is like, [69:19] hey, let's just build a new startup. [69:21] Let's isolate them from everyone else. [69:22] Different floor, different section in the office. [69:24] No one else is in there. [69:25] It's just their own Slack channel, their own everything. [69:27] They're entirely secluded. [69:29] And how you not push for that? [69:30] I don't know if we would've had the clarity [69:31] and the focus that we needed. [69:32] People might be offended. [69:33] Yeah, yeah. Of course. [69:34] Like all of the things, [69:35] all of the downsides you'd possibly guess are all there. [69:38] I just, I also think like that there's no path to like, [69:41] there's no way, you know, [69:43] the phrase I've settled on when I look back [69:45] on this is like, sometimes you just, you have [69:47] to go too far to know you've gone far enough. [69:49] You know, and I think a lot of the mistakes I see in people [69:51] who are trying to adapt to AI, for an example, [69:53] I'll come back to the co-founder here a sec, is like, [69:55] they tell themselves that they've not, [69:57] they've done enough because they, oh, you know, [69:58] a few sparkly buttons, [69:59] the merge features AI. [70:01] And we're happy— [70:02] We have an AI assistant to the product. [70:03] Yeah, exactly. And we've updated our homepage. [70:05] You say we're AI first, so like we're good. Yeah. [70:06] Like, and I think you need to be willing, genuinely willing [70:10] to like make brave hard to undo bets. [70:14] And I think you need like sort of obviously, you know, [70:17] having this sort of moral authority of a founder [70:18] and then being CEO kind of gives you some of that. [70:20] But still it's a huge decision to make. [70:23] And I think like I am much more of an, [70:26] my default DNA is like, [70:28] I'm more of an operator, in the sense of like, [70:30] all right, what are we doing? [70:30] Okay, well I'll make it work. [70:31] You know, whatever it is. [70:32] And I think if it was a company of like people like me, [70:34] like what you'd see is probably like, you know, predictable, [70:39] reliable, sustainable performance or whatever. [70:40] But like, but probably not enough actual like sort of, [70:43] well definitely not enough kind of brave big swings, [70:45] which is actually where you need to get to. [70:47] But I mean it is a cocktail like as in you, [70:49] somebody needs to go and actually do the thing [70:50] once we've decided what we're doing as well, [70:52] the way it ended up, like I was leading Team Fin [70:54] after Eoghan decided it. [70:56] And like that was what ultimately what led to the creation [70:58] of like the whole Fin initiative or whatever. Yeah. [71:00] You've now worked with so many different companies, [71:04] externally you've seen a lot. [71:05] What is predictive of success [71:07] and what is predictive of failure? [71:08] The biggest thing I'll always come back to [71:10] when I'm talking to anyone [71:11] who's trying to pitch me to invest [71:12] or pitch me to introduce John to invest [71:14] is it's always some version of [71:17] do you have a real product that solves a real problem [71:20] that really exists [71:22] and people are really already trying to solve [71:23] by paying money or time somewhere. [71:25] It sounds so trivial, [71:26] but you'll be shocked how many times you'll fail. [71:28] Or you'll get some sort of jazz hands type routine [71:30] somewhere along the way [71:31] where it's like, don't look too much at this, [71:32] but just trust me. [71:33] The areas that I end up being blind to in that is like, [71:36] you know, the extremely market expand-y type things, [71:39] like as in, if someone said to you, hey, [71:41] like all companies are going to have a chat room [71:43] and they're going to all hang out in an all day [71:44] and have unproductive conversations, it's going to be big. [71:46] I'd be like, oh, I don't see it, [71:48] like, you know, whereas like the, [71:49] so you would've missed out on Slack or whatever. [71:51] But I think like I can almost, you know, [71:54] hear from the like what, you know, what are you building [71:56] and why and who's it for? [71:57] And show me what the product does. [71:59] If it's not a real solution to a real problem, [72:01] I'm kind of already out. [72:02] And then the other big, [72:03] I'd say prediction is just like, [72:05] there's one of the things [72:06] that's happened in the last 10 years, [72:08] I'm sure you've seen this a lot, is like, [72:10] it was a lot easier to invest [72:11] when being a founder was uncool. [72:15] And I think like combination of like, [72:16] I blame genuinely The Social Network, [72:19] I blamed it just kind of the entrepreneurial lifestyle. [72:22] I blame like TikTok and blame all these things. [72:24] Soho House. [72:25] Yeah, yeah, exactly. [72:26] All of that, right. [72:28] To some degree like remote working, [72:30] I'd throw into the mix as well. [72:32] But I think the amount of people [72:34] who are chasing the trinkets [72:35] of being a founder of a startup, [72:37] even if they're quite smart [72:38] and they can actually kind of go and build something [72:40] if their actual motivation isn't the problem [72:42] or isn't just some deep like desire to be quite successful, [72:45] but it is instead to be perceived [72:47] like the whole kind of, I could have been a contender [72:48] rather than I could have contended, [72:50] like if you don't really, really want to be like [72:53] to actually like play the game [72:55] instead you just want to be seen to be playing the game, [72:57] I think that's probably the single biggest thing [72:58] that tells you like, you know, you're probably, [73:00] you're like best case scenario you'll sell at 5 million, [73:02] but more likely you'll still be alive in seven years. [73:04] All your investors will wonder what you're doing. [73:06] And you'll be basically [73:07] sending one investor update every now and then, you know? [73:09] Yes, I have noticed investor updates with metrics [73:13] don't predict success, [73:16] but investor updates without metrics [73:18] that tell a really fancy story [73:21] but don't have metrics [73:23] are actually quite predictive of failure. [73:25] Yeah. Those companies always fail. [73:26] I basically 100% agree [73:28] and honestly you can even tell [73:29] where the metrics are in the update, [73:31] like, because oftentimes like my favorite updates, [73:33] I mean this company I actually probably should— [73:35] And no investor updates are fine. [73:36] There's like a bunch of successful companies [73:38] that just never bought, [73:38] we never sent investor updates like, I'm sorry [73:41] for all the investors but we're bad communicators. [73:43] I was curious actually. [73:44] Yeah, yeah, but if exactly, but if you... [73:45] I owe you an email. [73:47] and if you go to the trouble [73:50] of writing an investor update, [73:52] and then make a proactive decision to not say [73:54] how your business is doing [73:55] that suggests some deep denial [73:57] about what running a business means. [73:58] So there's one company, I can't say [73:59] but like we're both an investor in it, [74:00] but like their updates just one of the most recent ones [74:03] was just like, here's performance ARR plus 17% [74:07] blah plus this blah plus that, [74:09] you know, something like, you know, [74:11] I hope you can see from the numbers we're doing great. [74:13] Best to luck to see you next quarter. [74:15] And I was like, yep. Brilliant. [74:16] Archive, I'll mark it up, you know, like, [74:18] so something like the, I think in general, [74:20] if you know the degree to, [74:22] I think it was Paul Graham said like the ratio of numbers [74:24] to words is usually the actual thing you're looking for. [74:26] Like, which is like, [74:26] if the numbers speak then the words don't have to. [74:28] Yes. Yes. [74:30] What else is predictive of success? [74:32] Numbers is one. [74:33] I almost kind of want to say the inverse [74:34] of all the things that I hate seeing. [74:36] I hate seeing founders who invest massively [74:37] in their personal brand instead of their company brand. [74:40] I hate seeing like people who are like obsessed about like, [74:43] if the first three or four updates I get [74:44] are like begs for retweets and quote tweets [74:46] and all that sort of stuff, that's always, [74:48] not a great sign, because it sort of says to me, [74:50] you haven't worked out how to market or whatever. [74:51] Like anything around like, you know, [74:53] what are the customers saying? [74:54] Like that's like, you know, whenever I reply [74:56] and say like, what do customers think of this feature? [74:58] They're like, "Oh, we're going to ask them." [74:59] Maybe what you're describing [75:00] is there's a very boring playbook, [75:03] boring and unglamorous playbook [75:04] for making products work about writing code, [75:07] talking to customers, running that iterative loop, [75:10] and avoiding distractions [75:12] and people who seem incurious about that playbook [75:15] or just are failing to execute on it [75:18] is kind of a warning sign. [75:19] Yeah. I would say that's definitely true. [75:21] Like, so it's basically like, [75:22] if you got like a decent product, [75:24] decent area, real solution, [75:27] and are you just willing to like work on the boring stuff [75:30] that needs doing to actually make that whole thing? [75:32] And then will you get bored in a year? [75:34] Or are you still excited by it? [75:35] Like, and I think to some degree, [75:37] even successful founders can get distracted [75:41] by glamorous opportunities whether it's like, [75:43] oh, there's a new wave of whatever, [75:45] like crypto or like NFTs or whatever. [75:47] you'll see people get their head turned quite a bit [75:49] and I think if you're like genuinely married to the problem [75:51] and married to the solution, [75:52] you'll tend to like sort of not be as distractible. [75:54] And then like so many of these businesses just need time, [75:56] you know, like they just need time and execution. [75:59] Yeah. Totally. [76:01] Last question, because I've had a lot. [76:04] In what ways is Intercom itself AI native? [76:07] The biggest initiatives we've driven recently has been [76:10] like around how we actually do R&D. [76:13] So I think we launched this initiative, [76:15] Darragh launched it, I think about four months ago, [76:17] called 2X, where we basically said, [76:19] hey, like we're going to double the productivity [76:21] of R&D before February 1st. [76:23] That means the, and we measure this [76:24] and like everyone's going to poke holes in this [76:26] and that's grand, but it doesn't really matter. [76:28] The measurement is, I think, [76:30] it's deployments to production involving code [76:34] that has to execute regularly right. [76:35] So it's a hard thing to fake. [76:38] If you fake it, it's like we should probably fire you, [76:40] you know, because like, so. [76:41] Now interestingly, everyone says like, [76:43] is not just the engineers going really hard? [76:44] No, like, there's like so many different elements to this, [76:46] but one of the biggest ones that we saw recently, [76:48] which really has been like awesome as Emmet, [76:51] our head of design, [76:52] he basically said, every designer by August 1st, [76:54] needs to be shipping code. [76:56] This is like the end to the discussion [76:57] of should designers code, right? [76:58] Designers should code. [77:00] So we basically said, hey, like, [77:02] all designers can now ship code. [77:04] Weirdly, the win there is that like, [77:07] the amount of engineering distraction has just gone away. [77:09] So every paper cut in your UI used to result [77:12] in a GitHub issue that get filed and triaged [77:15] and get picked up on a Friday in between breaks or whatever. [77:17] And now all that just goes. [77:18] So now you're like, oh, I want to fix this button [77:20] and fix that padding, fix that thing, change that radius, [77:22] change that color, all that just happens automatically. [77:24] And like it's getting to much meatier stuff [77:26] like redesign this entire UI, [77:28] redesigned this flow, change this wizard. [77:29] All of that's now being managed entirely by our design team. [77:32] What that has resulted in is like engineers [77:34] who are now doing far more, like, [77:35] who are staying in the zone for far greater [77:37] using like what it's called, [77:38] Claude Code or Codex or any of those, [77:39] or Augment or any of any of those tools to actually just, [77:42] you know, be far more productive. [77:43] And there's some real wins here, [77:45] like one reason one we had was like Fin works in Slack, [77:48] but when we were building that, [77:50] it was built very firmly from like, [77:51] how do we use AI here? [77:53] So it was like, let's build one perfect Slack solution. [77:56] Let's document all of our principles [77:57] and then let's have Claude Code, write [78:00] Microsoft teams, the Discord, the WhatsApp, [78:02] it's every other solution. [78:03] So we went live to production with Slack, [78:06] but I think we have like everything else [78:07] now in like public beta, like, and it'll all go live. [78:09] So we're finding all these, like, [78:10] it's a lot of like 1.2x wins, [78:13] but then every now and then [78:14] there's like a 50x productivity boost that we're finding. [78:16] So I think that's probably the biggest way [78:17] in which like the product is been built [78:19] from an AI native way. [78:20] On the go to market side of the house, [78:21] we've been slower, but like, [78:22] I think we're looking at like, [78:25] we've trained a GPT, if you like, [78:27] on like all of our marketing copy principles, [78:30] our content, our visuals, et cetera. [78:31] And we've been using that to produce a lot of like, [78:33] you know, stuff like event invites and things like that. [78:35] Hmm, well, Des, thank you. [78:38] Cheers. Thanks very much. Yeah.