AI 对你的产品战略意味着什么 | Paul Adams(Intercom 首席产品官)
AI 对你的产品战略意味着什么 | Paul Adams(Intercom 首席产品官)
AI 对你的产品战略意味着什么 | Paul Adams(Intercom 首席产品官)
文字稿
Paul Adams: 这就像一颗朝你飞来的流星。它将彻底改变社会。我认为,如果人们没有好好探索 AI,它就会把你甩在身后。我会从你的产品所做的事情开始思考:“它背后的核心理念是什么?人们为什么使用它?它为他们解决了什么问题?“诸如此类。回到基本面。然后问自己:“AI 能做到这些吗?“对很多产品来说,答案是”是的,它可以。“对有些产品来说,可能是”它能部分做到。“而对另一些,可能是”它做不到,至少目前还做不到。“然后,在某些场景下,AI 会是替代——它直接替你完成。而在其他地方,它会是增强——辅助你、帮助你。但无论如何,我认为你必须把你的产品、AI 目前能做的事以及未来将能做的事放在一起对照,然后问自己:“好,我们打算怎么做?”
Lenny: 今天的嘉宾是 Paul Adams。Paul 是 Intercom 的首席产品官,这个职位他已经担任了超过十年。在此之前,他曾任 Facebook 全球品牌设计负责人、Google 用户研究员、Dyson 产品设计师,而他的第一份工作是汽车内饰设计师。在这次对话中,Paul 分享了一些精彩的失败故事,包括他曾在一场大型演讲中在台上僵住、不得不走下台的经历,以及他从这些失败中学到的东西。之后我们深入探讨了如何将 AI 纳入你的产品战略,包括 Intercom 全力投入 AI 过程中大量精彩的实例。Paul 还分享了他最喜欢的一些框架、产品经验,以及更多内容。
这是我第一次不在自家录音室、而是在酒店房间里录制的节目。对我们所有人来说,这都是一次有趣的尝试。
戛纳演讲的故事
Paul,非常感谢你来参加节目,欢迎来到播客。
Paul Adams: 谢谢你,Lenny。很高兴来到这里。
Lenny: 很高兴你能来。我从很多不同的人那里听到过关于你的好评,所以真的很高兴我们终于做了这期节目。而且你有爱尔兰口音,根据我的经验,这对收听率总是有加成的,所以感谢你把口音也带来了。
Paul Adams: 是啊,听起来不错。
Lenny: 我想先聊几个故事。第一个是你在戛纳做主题演讲的故事,能讲讲当时发生了什么吗?
Paul Adams: 好的。工作中有些事情在当时非常令人难忘,但不会真正给你留下心理创伤。但这件事属于那种会留一辈子伤疤的。长话短说,那是十多年前,我在 Facebook。当时我很喜欢那里,我觉得那是个很好的工作场所。在旧金山期间,我为 Facebook 做了大量内部和外部的演讲。Facebook 在戛纳——全球最大的广告节——一直有一个主题演讲的席位。前一年,Zuck 做了那场演讲,他是主讲人,接受了访谈。他在隐私问题上受到了刁难,效果没有他们预期的那么好。
所以第二年他们让我来做。也许是我的爱尔兰口音让这个机会落到我头上。是的,我登上了那个舞台,全球最大的广告舞台。我记得大概讲到三四分钟的时候——那个演讲我已经讲过很多次了——我突然僵住了。我想不起接下来该说什么。那是我人生中第一次逐字逐句地背诵演讲稿。通常我只准备几个要点,内容会灵活调整,比较随意。但那次经过了媒体训练,被告知”绝不能说错话。“然后我就是想不起来该说什么了。我有点像经历了一次恐慌发作,走下了台,麦克风还没摘,骂了句脏话。台下所有人都开始笑。我当时心想,“天哪,他们在笑我吗?我的天,这……”
但我设法扭转了局面,重新走了出去。我内心那道防线反而被卸下了。之后大部分演讲进行得很顺利。那天晚上我出了名。后来在戛纳的海滨大道上,到处都是玫瑰酒,是的,我因为那场演讲既出了名也出了丑。
Lenny: 我觉得你经历了每个人在想到要做演讲时最害怕的噩梦。我觉得有意思的是,你挺过来了。这是一个很有意思的教训——你可以在数千人面前僵住,走下台,但最终一切都会好起来。
Paul Adams: 是的。我想这一切都是自然发生的。但从那以后,每次我走上会议演讲的舞台,即使是今天,我脑海中仍然会有那一丝疑虑。这种事之后再也没发生过。但我觉得,面对这些事情,当生活给你抛来这样那样的曲线球时,你必须去适应,而且事情并没有那么严重。说到底,这些事都没那么严重。你继续前行,从经历中学习。所以是的,但我仍然希望它不要再发生。
Lenny: 我也很讨厌公开演讲,而且总是担心这恰好就是会发生在我身上的事。所以我觉得听到这个故事挺好的——即使可能发生的最糟糕的事情真的发生了,事情也能挺过去。
Paul Adams: 你能扭转局面。是的。
Lenny: 我想听的第二个方面是你在 Google 的经历。你在 Google 参与了几个产品,这两个都算不上什么大成功。之后转到 Facebook 的过程也挺波折的。能分享几个那个时期的故事吗?
在 Google 与 Facebook 的岁月
Paul Adams: 是的。和上台演讲那件事一样,都是在经历中学习。我在 Google 待了四年,在 Facebook 待了大约两年半。在那两家公司期间,正是社交科技浪潮的顶峰。Google 对 Facebook 构成的生存威胁感到非常恐惧。而 Facebook 则非常自信,认为自己能推出一种新的社交广告单元,做出一个类似 AdWords 的东西,从内部瓦解 Google 的收入。所以在那个时期身处其中是非常迷人的,尤其是跳槽到新公司。
在 Google,我参与了很多失败的社交项目,就像你提到的——Google Buzz、Google Wave(编注:原文 Ventilator 疑为口误,实指 Google Wave)、Google Plus。我认为,这些项目很大程度上是出于恐惧来驱动的。出发点不是”让我们为用户做出一个好产品”,不是”让我们真正理解人们在和家人朋友沟通时遇到的困难”,也不是”让我们真正去创造一些美好的东西”。出发点是恐惧。
在那段时间里,我学到了很多关于如何不去做产品的方法。顺便说一下,应该提一下,当时的 Google 同时还在做很多了不起的事情——Google 在做 Google Maps,一个不可思议的产品,也是我最喜欢的产品之一,我认为是有史以来最好的产品之一。他们还在做 Android。我当时在移动团队和移动应用团队,Android 正好发布。所以 Google 完全有能力做出极其出色的产品。只是我恰好身处社交这一块,而这一块做得并不好。Google Buzz 是一场隐私灾难,Google Plus 也差不多。
在这过程中,我发表了关于”群组”(Groups)的研究,做了大量的调研。一个有趣的插曲是,当时我在 UX 团队做研究员,被要求做很多战术性的研究,比如可用性测试之类——用户能不能使用这些产品?但我同时也在同一批用户访谈中做了很多形成性研究。我会跟团队说:“我来做研究,回答你们的问题。但我也要做点别的事,我会额外花 20 分钟做这个。“我当时的做法是,和用户一起画出他们的社交网络——里面所有的人,家人、朋友,他们的沟通方式。我们把所有沟通渠道都标出来,讨论哪些好用、哪些不好用。我们在大约 18 个月里和几十个人做了这件事。每次都出现同样的模式:人们需要更好的方式来与家人和朋友的小群体沟通。
现在回头看,我会说”WhatsApp”。或者如果大家都用 Apple 的话,那就是 iMessage。事后看来显而易见。但当时并不明显。于是我们试图围绕这个洞察来打造一个产品,就是 Google Plus。但同样,出发点是错的。在这过程中,我所做的所有研究通过一次会议演讲被公开了。Facebook 注意到了,联系了我,然后一步步发展,我离开了 Google 加入了 Facebook。对我个人来说,这是一件了不起的事。
当时的 Facebook 是一个非常了不起、令人兴奋的地方。他们做事的出发点是正确的——“好,让我们为用户打造一个出色的产品。”
Lenny: 所以这是在 Google Plus 还在开发期间,你基本上中途转去了那边。
Paul Adams: 是的,中途走的,说起来我现在都还觉得紧张。那个项目还没有发布,还在保密阶段,高度机密。当时 Google 做了很多前所未有的事情。我不知道后来是否还这样做过。比如,所有参与 Google Plus 的人被安排到另一栋楼里,那栋楼用不同的门禁卡,如果你不在 Google Plus 团队,你就进不去。当时还有各种反常规文化的做法。结果就是,Google Plus 在内部引发了很多对立情绪。所以当我在项目进行到一半时离开,带着脑子里的所有计划投奔了竞争对手——有些人把我看作叛徒,这可以理解。也有些人觉得我是开明的,觉得和我聊天很有意思。但对我来说,这是正确的选择。不过当时确实是一件很难的事。
Lenny: 我知道你带走了什么以及整个过程也受到了很多审查。
Paul Adams: 是的,当我离开时,Google 假设我是间谍之一。我被隔离了。我告诉他们我要离职,他们对我的笔记本电脑进行了取证分析,诸如此类的事情。所以当时挺紧张的。现在回头看,我能理解为什么会那样做。但根本原因在于,这个项目从一开始就是出于竞争恐惧在运转的,而我认为这不会带来好的结果。
拥抱失败
Lenny: 所以你刚才分享的这些故事中,一个贯穿的主题是——我不想说得那么刺耳——就是事情没有成功。我很好奇,作为一位产品领导者,你认为让人们经历这些有多重要?你是否觉得这甚至可能是一件好事?作为一名教练、导师,对于那些试图成为像你一样的人,你有没有觉得其中有什么特别有帮助的东西?
Paul Adams: 非常重要,非常重要。现在依然如此。我个人失败过非常多次。这两个故事——Google 那件事牵涉很深很广——只是其中两个故事。我失败过太多次了。我记得在 Facebook 的时候我非常开心。那时我认识 Eoghan 和 Des,Intercom 的联合创始人。他们试图说服我加入 Intercom。当时 Intercom 还是一家 10 个人的公司。Eoghan 当时对我说了一句话,从此一直留在我心里。他说:“在 Facebook,你可以设计产品。但在 Intercom,你可以设计公司。“这对我来说非常有吸引力,一个极好的 pitch。他说:“和我们一起设计你想在其中工作的公司。”
其中一部分就是打造一个拥抱失败的公司,一个允许尝试的公司。我非常信奉大胆下注,高风险高回报。增量式的东西没那么让我兴奋。不,我没说那个不对——当然那也有它的位置,尤其是随着公司规模变大。但大胆下注才让我兴奋。如果你大胆下注,你会在很多事情上犯错。所以我们在 Intercom 建立的很多原则都是关于如何构建软件的。
我们有一条原则叫做 Ship to Learn。后来我们改了,墙上现在写的是 Ship fast, ship early, ship often。你可以说 Ship to Learn。Ship fast, ship early, ship often。这个理念中就包含了失败的意思——事情不会总是对的,而且大多数时候会是错的。但如果你尽早发布、快速发布、快速学习,你就能快速调整、快速改进。这就是我们尽可能去拥抱和传授的文化。但说起来容易做起来难。
Lenny: 是的,尤其是当你身处其中的时候——“该死,一切都要崩了,我真的搞砸了。”
Paul Adams: 是的,而且这里有一个人们很难处理的质量上的权衡。我们对自己有很高的标准。Intercom 很大程度上源自设计出身的创始人背景,我们非常重视工艺。我们绝不想因为发布的东西而感到丢脸。所以这里存在一种真正的张力,一种真正的权衡——人们有这些高标准,这是我们鼓励的;同时我们又鼓励他们快速发布、学习、犯错。这是我们在不断权衡的一种持续的张力。
AI 的全情投入
Lenny: 说到大胆下注和全力投入,我知道 Intercom 发生了一个巨大的转变,就是转向 AI、拥抱 AI。那么,也许我们先从一个宽泛的问题开始——到目前为止,你对 AI 的思考以及 AI 将如何融入产品和产品战略,有哪些更宏观的洞察或意外发现?
Paul Adams: ChatGPT 是哪天发布的?去年 11 月 29 号吧,我记得。从那天起,我几乎每天醒来都在想 AI。我尽可能多地阅读,但仍然觉得自己远远落后。我觉得,当我跟人聊 AI 的时候,人们通常分成两个阵营。你要么全情投入,真正地全情投入——这是一颗正在朝你飞来的陨石,这比移动互联网作为一次技术变革更大,和互联网一样大,也许作为一种技术变革,它对社会塑造的影响甚至可能比互联网本身还大。所以我是全情投入的,我已经翻过了那座山,站在另一边了。这是一类人。
然后,我认为还有另一个阵营的人,他们的想法是:“我听过这种话了。这是炒作。去年是加密货币,是 Web3,那些东西都没兑现。还有元宇宙。” 所以确实存在很多怀疑,或者说犬儒主义。我不太理解为什么。那些东西确实没兑现。元宇宙还在慢慢回来。我在想那个定律——先是炒作高峰,然后是幻灭低谷,然后再从另一边走出来。
Lenny: 对,那条曲线。
Paul Adams: 对。我觉得很多人可能就处于那个阶段——之前炒作太多、噪音太大,现在噪音仍然不少,所以你会有点屏蔽它。有些人就落入了那个阵营。而我是完全站在另一个阵营里的。这将从根本性上改变社会,我看到新类型的产品不断出现都觉得不可思议,比如 ChatGPT Vision 刚刚发布,看到人们能用它做的事情,真的很震撼。而且我们还只是刚刚触及表面。所以,我们绝对是全情投入的。
Lenny: 太好了。我想展开聊聊这个。但我觉得还有一类人,他们的想法是:“是的,确实有大事在发生。我就是没时间去理解、去构建、去尝试。” 对于那些说”我想深入了解这个兔子洞,但我不知道从哪开始,因为我已经有太多工作了,这不是个附带的事”的人,你有什么发现或建议?
Paul Adams: 我给人们的建议,也是给我自己的建议——我自己也在其中——我每天醒来面对的是太多的邮件、Slack 消息、敲门的人、到我桌前的人,各种各样的事。所以这对我也同样是个挑战。你必须腾出时间,对我来说没有别的办法。这不意味着……这是关于优先级的问题。这不意味着你需要疯狂加班。我不相信疯狂加班。我都不知道自己工作多长时间,也许一周 50 个小时吧。我觉得超过那个限度你就会做出糟糕的决定之类的事情。你会疲劳。你还需要过你的生活。你必须把这件事安排进你的日常里。无论是专门留出时间来阅读。
阅读是关键。你必须阅读。你必须保持信息更新,必须去动手尝试、体验。如果你没有 ChatGPT……如果你没有——我记不清是不是需要 Pro 许可证什么的——但如果你还没有升级去获得 GPT Vision 那样的功能,可以用手机拍照、用移动端 App 的那些功能。上周五晚上我和妻子出去吃饭,我尽量不把工作带到和妻子的晚餐上。但我忍不住想试试。我拍了几张她食物的照片。你可以做各种疯狂的事情,比如告诉你这顿饭有多健康之类的。
Lenny: 哦,哇。
Paul Adams: 总之,你必须去尝试。你就是要去尝试。所以我对人们的建议就是,你必须去尝试,你必须腾出时间,否则它就会与你擦肩而过。这让我想起大约十年前的移动浪潮。我当时在 Google,在移动团队工作。所以我想,了解最新动态本来就是我工作的一部分。但在那个时候,像 Facebook 这样的公司全力投入了移动端,也许有点晚,但他们最终做出了勇敢的决定。我觉得如果人们不去好好探索 AI,它会把他们甩在后面。
Lenny: 这让我想到,我觉得在 Facebook,Zuck 这么做过,Airbnb 的 Brian 也做过——他说,“你给我看的任何新产品设计稿,必须是移动 App 或移动端的。从现在起不能再是桌面端的。”
Paul Adams: 对,是的。Facebook 也是一样。没错。
Lenny: 我猜,你觉得这就是作为领导者应对这件事的方式吗——直接说”你给我看的东西都必须有某种 AI 组件”?这听起来大概不是个好主意,但你有没有在想什么办法,或者做过什么来让大家相信这是他们应该花时间的地方?
Paul Adams: 对,这确实更难。因为——
Lenny: 你不想强行摊派。
Paul Adams: ——对,很多技术是看不见的。我们有一个机器学习团队,已经存在很长时间了,所以我们在这个领域已经工作了相当一段时间。但有趣的是,即使回到 18 个月前,如果 18 个月前我在你的播客上,你问我”你怎么看 AI?“我会说类似”那不是真的。机器学习是真的,我们来聊那个。“所以事情在变化,我对它的认知也在变化。但很多改进是在幕后的——大语言模型或者人们在后台基础设施上构建的各种东西。
战略性地思考 AI
所以我不知道”设计 AI 模型的 mock-up”会是什么样子,就像当年移动端的 mock-up 那样。但我确实认为人们需要开始真正地进行战略性思考。也许不是在 mock-up 阶段,而是开始对自己的产品进行真正的战略思考——它是否在这条冲击路径上,或者说是否即将被波及。并不是所有产品都会被波及。如果是的话,对于某些产品来说需要根本性的战略转变;对于另一些可能没那么大。但我认为这才是人们需要进入的那种思维状态。
Lenny: 能再深入说说吗?怎样才能真正深入思考你的产品是否在那颗陨石的路径上?
Paul Adams: 你确实可能被技术带偏。我会的。我刚提到,出去吃饭时拍食物照片。你可能被技术带偏,有些确实很酷。我不会从那里开始。我会从你的产品所做的事情开始。它背后的核心理念是什么?人们为什么使用它?它为他们解决了什么问题?诸如此类的。然后,再提出那个问题。回到根本——“我的产品是做什么的?人们为什么喜欢它?“然后再问:“AI 能做这件事吗?“对于很多产品来说,答案是”是的,它能。“对于有些产品,可能是”它能部分做到。“然后对于另一些,可能是”它做不到,至少目前还做不到。”
Paul Adams:
所以你需要把你产品的功能和 AI 的能力做一张对照图。AI 能做的事情很多。它能写作,我给你列一下——它能写作、能总结、能总结文本、能撰写文本、能回答问题、能查找事实、能扫描文本、能扫描图像。它能听你的声音并复述。它还能执行操作,这是下一个即将到来的大突破——它能执行操作,真正地去做事情。就像,“嘿,AI,不管这个 AI 叫什么名字——帮我把航班改到周二。“对吧?它能做这类事情。
所以它能做的事情很多。它能建立规则。因此我认为,任何包含工作流的产品——几乎所有 B2B SaaS 产品都包含工作流——任何涉及多媒体的产品,它们都在那条轨迹上。我不知道这个比喻是否恰当。但这股浪潮正在逼近,而它们正处在它的路径上。对于很多这类产品,你只需要看看 AI 能做什么。其中一部分将是替代——AI 会替代,直接完成工作。而在另一些场景下,它将是增强——它会增强人的能力,辅助人们工作,就像现在流行的 copilot 理念那样。但总之,我认为你必须把自己的产品和 AI 能做什么、将来能做什么做一张对照图,然后问自己:“好吧,我们打算怎么办?“
Intercom 全面转向 AI 的经历
Lenny: 在 Intercom 或者其他公司有没有这样的例子——“这是我们想解决的问题?哦,AI 实际上可以完全替我们做这件事。”
Paul Adams: 哦,有的。先说说 Intercom 的例子。那个日子,我想是 11 月 29 日,深深印在我们脑海里。我们的机器学习负责人 Fergal,那天他……我记得他那天发了条推文,大意是:“就是它了。就是这个时刻。这就是前后的分水岭。“我经常谈到——因为我有一个分析框架,叫”之前/之后”时刻。这就是一个”之前/之后”时刻。那之前是一个时代,那之后是另一个时代。一切都变了。所以我们几乎完全推翻了原来的战略,从第一性原理出发重新开始,问自己:“人们为什么使用 Intercom?“Intercom 是一款客户支持产品。紧接着,OpenAI 的创始人兼负责人萨姆·奥特曼说:“最先被颠覆的行业之一就是客户服务。“我们心想:“没错。”
所以我们真的这么做了。我们彻底改变了思考方式和工作方式,埋头全力打造了一款叫 Fin 的产品。其实我们先做了其他东西,Fin 是后来才推出的,现在回想起来是这样。但我们全力以赴了,带着一种几乎”赌上全公司”的心态。所以,我们做了。我认为其他公司比如 Google 和 Bard 也需要这样做,也许他们稍微慢了一点,但这个技术周期才刚开始,我觉得他们没问题。是的,我们做了。过程很艰难,但我们别无选择。
Lenny: 能简单介绍一下 Fin 是什么吗?给不太熟悉的听众说说。
Paul Adams: Fin 首先是一个 AI 聊天机器人。说到客户服务,人们会对企业提出问题,过去主要是邮件和电话,基本上以工单模式为主。你提交一个工单,收到很多”请勿回复”的邮件,等等。后来出现了对话式客户支持,就是基本的即时通讯,像 WhatsApp 或 iMessage 那样,我前面提到过。现在则进入了”机器人优先”的体验阶段。Fin 是一个 AI 聊天机器人,AI 优先,聊天机器人优先。客户支持团队的第一道防线是 Fin,而不是人。所以这从根本上改变了局面。Fin 取得的成果令人震惊。我们最大的挑战,其实是帮助客户支持团队思考组织变革的问题。
技术远远走在了前面。真正需要跟上的是让人们理解这对角色、对团队意味着什么——还有很多有意思的东西,比如新的岗位类型,比如对话设计师——我们有这样的职位,由你来设计 Fin 的对话流程——以及管理者。总之,这就是 Fin。Fin 后来做了扩展。现在 Fin 也进入了我们的 Intercom 收件箱。客服人员在那里回复客户的咨询,现在 Fin 也在里面了,协助支持代表工作——为他们建议回答,或者帮助他们重新措辞。所以它现在既能自主回答问题,也能增强人的能力。
全力投入 AI 后的影响
Lenny: 我觉得你们是为数不多完全转向 AI 的公司之一。我觉得这里面有很多关于团队结构可能如何变化、产品策略、优先级等方面的经验教训。所以我想再展开聊几个问题。首先,全力投入这个方向之后,你们看到了什么影响?
Paul Adams: 说实话,现在还太早,很难完整回答这个问题。而且取决于你用什么标准衡量成功。再说一次,AI 领域有大量的炒作和热议。如果你以关注度来衡量,那就是巨大的成功。我们的目标客户是客户支持领域——客户支持经理或负责人。他们非常好奇,会问:“它真的有效吗?“回到之前说的,现在炒作太多了,所以也伴随着一些质疑——“它真的有效吗?能和人一样好吗?“而客户支持领域的工作者通常非常有同理心,很在乎人。所以他们会问:“但它能和人一样好吗?它友善吗?它能理解人性吗?“于是,有大量的好奇、大量的兴趣,大量的人在尝试。
我们有一些客户取得了巨大的成功。他们可以用 Fin 回答高达 50%、60%、70% 的 incoming 问题。所以我们确实有客户看到了很大的成效。但还太早。那么,它是否已经在财务上改变了我们的业务?还没有。我觉得所有快速增长的创业公司……如果你把 AI 版 Intercom 看作——我想可以说是一个新的创业公司,尽管我们有 900 名员工——它的增长曲线,你要找的是那条指数增长曲线,而不是大型上市公司那种线性增长曲线。而指数曲线的话,需要时间。头一两年是那条曲线的底部。所以我觉得我们仍然处于摸索阶段,尝试搞清楚到底发生了什么,努力去与人们沟通、教育市场。但我们已经有足够的证据相信,这绝对是未来。
Lenny: 有没有什么例子——不管是这款产品还是其他 AI 的应用——让你觉得”哇,我从没想过它能做得这么好”?
Paul Adams: 我回到那个”之前/之后”的框架。第一版 ChatGPT 就是一个”之前/之后”时刻。我们一直在——就像我说的——深耕这个领域,我们有机器学习团队已经很久了。在 ChatGPT 出现之前,我们的机器学习系统需要手动设置。客户支持经理需要编排机器人,教它说什么,大量的编排工作,大量的教学。然后 ChatGPT 出现了——“哦,它可以自己做。“它有时候会出错。但人回答问题也会出错啊?在很多基础问题上,它几乎和人一样好。这让我非常震撼。然后,首先是”哦,它能回答问题。“但你接着发现,它能推理。
AI 的推理能力与代码生成
Paul Adams: 关于这到底是推理还是演绎,其实是有争论的。但它能自己把事情推导出来。我不是那种喜欢钻进极其哲学化问题里的人。我的态度是:“我们只需要动手建造。回去做产品就好了。“但它确实能把事情推导出来。这一点让我非常震撼。我们给 ChatGPT 喂了数据,其他公司也一样,我们也玩过其他大语言模型,比如 Entropik 等等,它能自己推导出结果。这太令人惊叹了。然后你看到它能做一些事情,比如写代码。我就想:“哇,它写代码真的很厉害。这意味着什么?“接着你开始思考,在 Intercom,我们的比例是一比五,一个产品经理大约配五名工程师组成一个团队。你看着这个东西在写代码,你就会想:“接下来会怎样?我们还需要那么多工程师吗?还是他们的角色会发生变化?他们会开始做不同类型的事情,比如审查代码而不是写代码?“
视觉能力与影像识别
所以这让我非常震撼。然后是视觉方面的能力,就像我之前提到的,我觉得视觉那个时刻比最初那个还要大。它能解析图像,能帮你理解这个世界。你拍一张自行车的照片说:“嘿,哪儿坏了?“它会告诉你哪里坏了,怎么修。你旅行的时候,拍一些东西的照片,上面是另一种语言,刻在一座十二世纪大教堂的石壁上。你想:“这写的是什么?“它会告诉你上面写了什么。就是那种能力,怎么能做到的。这些天我在爱尔兰跟人重复最多的是这个——如果你想成为一名放射科医生,也就是学习看 X 光片、告诉人们哪里出了问题等等,需要七年的训练才能掌握这项技能。七年才能当上放射科医生,然后你才刚刚入行。而 AI,看起来已经比人更擅长这件事了。它已经更擅长了,而且它能吸收有史以来所有的 X 光片。没有任何一个人能读完、思考并综合有史以来所有的 X 光片。所以它当然更强。然后你就会想:“好吧,接下来会怎样?“我想,整个工作会发生变化。放射科医生不会再分析 X 光片了——嗯,他们可能还是会去拍片,但分析肯定不会由他们来做。他们会看 AI 给出的结论,检查它是否正确,然后就到了医患沟通的环节——告诉患者结果,也许告诉他们治疗方案。所以这份工作从根本上变了。顺便说一下,这可能是件好事。在爱尔兰,医院排队长,人们等拍 X 光片的等待名单长得离谱。所以这对人们来说可能真的是一件好事。
语音克隆与数字复刻
最让我觉得疯狂的是这个:AI 能听你的声音然后复制它,它能说话而且听起来跟你一模一样,真的非常好,几乎无法区分。你会觉得:“那听起来就像 Paul。“我之前提到过 Metaverse,不知道你有没有看到 Zuck 和 Lex 的那次对话,你看了吗?
Lenny: 看了。
Paul Adams: 那是我第一次”哦”的一下。没看过的人说一下,他们是在 Metaverse 里见的,某种虚拟世界。
Lenny: 是一个黑色的房间。
Paul Adams: 黑色房间里。对。而且技术进步了很多,他们可以分析你的面部然后构建 3D 模型。非常好,真的非常接近了。所以你可以想象,那只会越来越好。从那项技术的发展轨迹来看,它会越来越好。所以语音那部分加上面部那部分,意味着这两样东西都几乎无法与真人区分。而 AI 将能够吸收人们说过的和做过的一切。当人们去世的时候,它将能够复制那个人。于是就有了某种”来世”——嘿,你的父母去世了,你仍然可以和他们说话。这可能成为最诡异的事情。也许对人们来说这并不是好事。我不知道。但这项技术就在眼前了。而且 AI 能回答你的问题,令人震撼。真的是令人震撼。
Lenny: 其实《黑镜》有一集就是这个设定,里面——
Paul Adams: 没错。
Lenny: 对,而且我觉得那一集结局不太好。
Paul Adams: 不好。
Lenny: 小心点。
Paul Adams: 当然,当然。还有那个语音翻译,也是另一个让我震撼的东西。我记不太清了,也许是《碟中谍》里,它能采集一段声音然后实时翻译。这项技术也已经到来了——如果我的母语是西班牙语、不会说英语,你和我之间仍然可以做这期播客。你的声音会实时被翻译成西班牙语给我听。再一次,令人震撼。
Lenny: 我们其实正在做播客节目的配音和翻译工作,全部通过 AI 完成——它识别你在说什么,翻译成西班牙语,还会改变你的嘴型来匹配。我们正准备推出几期这样的节目。这其实是非常依赖 AI 的。
Paul Adams: 很酷,真的很酷。
AI 对工程团队的影响
Lenny: 你提到你的工程团队可能会改变思路,因为 AI 可以让他们效率更高、工作方式不同。我很好奇你在团队中实际看到的变化,不管是使用 AI 类工具,还是构建 AI 产品本身。你觉得最大的不同是什么?我特别想从这样一个团队的角度了解——一个正在考虑整合 AI、开始拥抱 AI 的团队——你看到最大的变化是什么,以及应该发生什么变化?
Paul Adams: 归根结底,你需要非常优秀的机器学习工程师。这是一切的起点。如果你没有这个,你会发现很难做出真正出色的东西。OpenAI 提供的、Entropik 提供的、Claude 提供的,它们提供了令人惊叹的技术,但你必须在之上构建。如果你真的想要卓越的东西,你必须在它之上构建。我们把他们构建的东西针对客户支持场景做了适配。也许某一天我们需要去构建自己的、专为客服设计的 LLM。也许吧。我不知道这一切最终会走向何方。也许每个企业都会有自己的 LLM。老实说我真的不知道。也许这些公司会提供专门的 LLM。但不管怎样,这是第一件事。当然,这些人非常抢手。所以你需要投资建设这个职能,我认为。真正投资去建设这个职能。
我们一直在做的就是这个。我们的机器学习团队比以前大得多,比 Intercom 历史上任何时候都大。然后它就分叉了。一些项目非常依赖机器学习团队,需要他们深度参与。但另一些项目更偏前端,比如我之前提到的收件箱的东西——Fin 已经上线了,底层技术我们已经建好了。现在的问题是:如果有一个人类客服在收件箱里回答问题,那是一个自然的聊天对话界面,非常直接。但当里面多了一个 AI 助手呢?它们怎么交互?做什么?什么时候介入?你怎么在用户体验中呈现这一切,让它感觉自然?这是一个非常难的设计问题。
也就是说,回到团队层面——我们有一个产品团队,包括一个产品经理、一个产品设计师,也许三到四名、也许五名工程师,他们从机器学习团队那里获得支持。所以我们现在两种模式都有了。而且越来越多的工作可以用后一种模式完成——越来越多的团队可以基于我们过去十二个月左右一直在建设的基础技术来构建产品。这是第一点。我觉得第二件浮现在脑海里的事情是:不要把它想成是”外挂上去的”。我觉得有些人还停留在那种思维里。
不要把 AI 外挂到产品上
Paul Adams: 我还是回到移动互联网那个例子,两者之间有太多直接的相似之处了。就像我前面说的,在 Google 的时候,我在移动应用团队工作。我做过移动版 Gmail、移动版 Docs,我们是专门的移动团队。我们在伦敦,心里想着:“嘿,我们是伦敦的移动团队。“与此同时,在加州山景城,根本没人在意。他们的态度就是:“你们就二十个人,我们有两百个。没人在手机上用这些东西。“同样,充斥着大量的怀疑。“没人会在手机上写文档的。认真的吗?他们要在手机上写一整篇文档?你疯了吧?“所以,不要这样做。我们在努力避免这样做。不要把它外挂上去。不要想着”哦,我们找一帮 AI 专家……”我们确实有一些专家。但总体来说,我们希望所有人都能学习了解这个东西。
Lenny: 有意思。我很好奇,具体来说”不要外挂”是什么样的。你的意思是不要设一个独立的团队说”他们是 AI 团队,他们来给所有东西加上 AI”。你的经验和教训是,AI 应该融入每个产品团队。
Paul Adams: 这方面我们还处于早期阶段,还比较早。我们努力避免的是设立一个”AI 收件箱团队”,然后他们是唯一在收件箱里做 AI 功能的人。我觉得让所有人都去学习了解要好得多。顺便说一句,我是通才的坚定信仰者,非常非常坚定地相信……我想我的背景就是”样样通、样样松”,大概就是这么描述我自己的。我做过研究员、设计师、产品经理。所以我信奉通才,所以我信奉用这种方式组建团队。当然,专家在某些时候也很重要。机器学习毫无疑问就是一个深度专业领域。在 Intercom,我们在工程方面也普遍更青睐那些愿意学习新东西的人——不管是一门新的编程语言、一个新的框架,还是如何设计 AI 界面,等等——让更多人能够上手做这些事。
AI 时代如何保持学习
Lenny: 我感觉你们公司有点像是活在未来的状态,很多公司一旦意识到”天哪,我们真的需要在这上面发力”的时候,才会走到你们现在的阶段。或者他们已经在做了。我很好奇,你是否还遇到过其他坑,觉得大家应该尽量避免的?或者关于这个转型过程中还有什么你觉得可能对其他人有用的经验?
Paul Adams: 嗯,我到目前为止提到过的就是:不要外挂。保持学习更新。我前面提过了,读、读、读。我总觉得自己跟不上,这一切发展得太快了。
Lenny: 你都在读什么?你觉得在了解 AI 领域动态方面,什么最有意思、最有价值?
Paul Adams: 我很想告诉你我的学习方式非常有条理,有一个很棒的阅读清单,每周日早上会去读。但实际情况挺随机的。我花很多时间刷 Twitter——当然现在叫 X 了。我在 Twitter 上关注了一些人。其实我经常用 Twitter 的推荐信息流。我觉得因为我经常互动和浏览 AI 相关的内容,所以系统给我推的也越来越多。所以我会有意识地这么做,试图发现更多东西。我也会直接在 Twitter 上搜索,上面有很多很酷的内容。还有一些Newsletter,以及一些我关注的人。
Lenny: 有没有什么具体的 Newsletter 可以推荐的?
Paul Adams: 有的,Matt Rickard 是一个人,他经常谈论 AI。还有各个公司的博客也不错。OpenAI 的博客挺好的,他们会写论文,并且做摘要总结。
Lenny: 好的。如果你之后想到其他的,不管是 Twitter 上值得关注的账号还是 Newsletter,回头邮件发给我,我们加到节目备注里。
Paul Adams: 好,没问题。嗯,肯定还有的。我再找找。你刚才问怎么做——就是去试。试着留出半个小时,深入钻研半个小时,收藏几个东西,之后再回来看。就像所有人一样,你会很忙,会有很多干扰,你必须专门腾出时间来。
Lenny: 还有没有其他你觉得特别有用的工具或应用?听起来 ChatGPT 是你折腾这些东西的核心。还有没有什么你觉得特别有意思的?
Paul Adams: 我会试试其他的东西,比如 Bard。Bard 是 Google 的 AI 搜索引擎。Rewind 是另一家很吸引人的公司。应该是 rewind.ai。Rewind 基本上就是一个用 AI 增强你记忆力的工具。安装在你的本地电脑上,它会捕捉一切、记住一切。所有数据都在本地处理,所以没有隐私问题。你必须亲自试用这些东西,才能理解它到底好不好用、有没有用、边界在哪里、它是怎么运作的,等等。所以我信奉这种做法。
组织层面的挑战与信念
Lenny: 当你开始推进 AI、朝这个方向发力的时候,有没有遇到什么大的组织层面的挑战或障碍?或者个人利益、意见方面的冲突?我不知道。有没有什么让你绊了一跤、需要费力气去克服的事情?
Paul Adams: 有的。Intercom 内部对很多事情都有多元的不同意见。在 AI 这件事上,我已经全力投入了。我倾向于向前冲。大势所趋,我已经被说服了,早就过了那个犹豫的阶段。但同时,也没人真的知道未来会怎样。没人知道。所以很多时候我们内部讨论的时候,我们的联合创始人兼 CEO Eoghan 非常坚定地支持,联合创始人 Des 也是,还有我,以及高层管理团队中的很多人,我们都属于全力投入的阵营。这一点帮助很大。当然,如果公司的高层管理团队都全力投入,自然就会向下渗透。但同时,也有人说:“你为什么全力投入?“我的回答是:“基于经验的判断。一种直觉。”
商业战略和产品战略中,有部分东西就是很难。就像品味一样。人们常说产品品味,“谁有产品品味?“很多时候,这就是基于经验的判断力。我只能这么说。我真的不知道。就我个人而言,我不知道,我在 Google 亲历了移动互联网那波浪潮,因为我在那里做过移动端。我亲身经历了那个阶段。所以我能看到同样类型的事情正在发生,而且规模更大。我正是借助那段经验来做出全力投入的决定。
航线图可能全部作废
Paul Adams: 但对一些人来说这是个挑战,因为他们没有那样的经历,或者不认同这个判断。我们内部关于未来有很多争论。前面提到的 Fergal,给我和其他几位产品负责人以及 Des 做了一次——我不知道该叫什么,一次宣讲?还是一出戏?反正他的核心论点是,也许我们整个 AI 路线图都是错的。你熟悉 Horizon 框架吗?就是 Horizon 1、2、3 那个。
Lenny: 嗯,对。亚马逊用的那个。
Paul Adams: 对。Horizon 1 是短中期,未来 12 个月,12 到 18 个月。Horizon 2 就是”接下来会发生什么”,大概 18 到 36 个月之后。不同的人用不同的时间范围来划分。总之,我们现在处于 Horizon 1 的阶段,想着”好,明年我们做这个”。然后 Fergal 说:“对,但两年之后,如果这条路走下去,我们现在做的一切都会变得无关紧要、毫无用处。“你就想:“哦,好吧。“所以这类讨论是有的。而且模糊程度极高。所以很多挑战就在于在这种模糊中导航,帮助别人获得我拥有的那种确信,同时又不压制那些不同的声音和意见,因为这些意见往往也是有道理的。
Lenny: 那帮助人们建立确信的方式是什么?是不是就是给他们看例子,“你看这个。""哇,看看这个,太不可思议了。“还有我觉得应该也有帮助的是,你们所在的市场本身对 AI 来说就是一个非常清晰的机会,相比很多其他市场,这个推销起来应该更容易。
其他行业同样面临颠覆
Paul Adams: 是的,没错。确实如此。给别人看确实是最简单的方式。我觉得客户支持绝对是……就像我说的,客户支持是排第一的。所以你会想:“好吧,我们确实应该适应。“适应或者死亡,这是我们的信条——适应或死亡。我认为还有其他行业也在经历同样的旅程,只是没那么明显。比如报表软件,Tableau 或者任何报表产品,它们是怎么运作的?典型的读写型应用,构建仪表盘,筛选,查询,复杂查询,查询数据库,拿到数字,在界面上展示出来。其中倾注了大量心思来考虑如何向人们呈现数据,用什么样的图表类型最合适,最终帮助人们做出好的决策。
我觉得,这又是一个大胆猜测,谁知道呢。也许这一切都已经过时了。未来的报表产品可能就只是一个框,这个框直接连接数据库,你问”去年一月份我们最好的销售员是谁?“它就回答”一月份表现最好的代表是谁?Lenny。“未来的报表产品可能就是这个样子。项目管理工具也是另一个例子。我认为有很多产品,除了最显而易见的客户支持之外,同样成熟到可以让一个新入场者带着完全不同的范式来接管。
Lenny: 我喜欢这个点,它呼应了你最初说的关于思考 AI 如何融入的方法——想想你作为一家公司在解决什么问题。比如 Tableau,帮助人们可视化数据。那问题就变成了,AI 能不能直接替你做?如果可以的话,那你基本上就有了一整套战略——“好,我们怎么用 AI 来实现这件事?”
Paul Adams: 是的。不过报表这件事会不会真的那样发展,我也不知道。但如果你是一家 Tableau 类型的公司,你有大量的设计师在设计仪表盘、筛选器和查询式的工作流。他们该做什么?界面就是那个框。所以要在脑子里转过这个弯来很难——“我们必须……”如果你坚信必须做出改变,那真的很难。
团队成员如何建立 AI 认知
Lenny: 也许在这个话题上最后一个问题。对于正在学习和开始在这个领域工作的团队成员,除了你之前已经分享过的建议——大量阅读、关注 Twitter/X、订阅这些 Newsletter,然后动手尝试——还有没有什么你觉得对帮助他们上手有帮助的?
Paul Adams: 我也会去读那些说这一切都是垃圾的观点。这很容易……我很多次都犯过这个错。回到我犯过的错误这个话题,我很多次都犯过这个错,就是跳上了某个潮流的顺风车,结果完全搞错了。年纪越大……Web3 那个东西,我说”我甚至不知道 Web3 是什么。“加密货币,我从来没买过。也许我在这件事上是错的。但我不是一个爱追潮流的人。但可能我年轻的时候是。现在我尽量去读那些相反的意见,那些持怀疑态度或认为这事很糟糕的人。很多人认为这对人类是灾难,这项技术会把我们吞噬。所以我尽量平衡我的乐观。我是一个不切实际地乐观的思考者,所以我尽量用一些悲观来平衡。
Lenny: 这个建议真的很好。
Paul Adams: 是的。
Lenny: 在我们切换到另一个话题之前,在这个领域还有没有什么你觉得可能有用的东西想分享的?
不要害怕 AI
Paul Adams: 哦,对。另一件事是,不要害怕。我觉得人们对此有点恐惧。比如,如果我开始在我们办公室里到处说”我觉得今后每个团队只需要两个工程师”,那可能不太合适。而且实际上我觉得事情不会那样发展。我只是觉得过去这些年有很多很好的研究表明,人们最终并不会丢掉工作,工作会被重新分配。而且对于客户支持来说,比如,它本来就是一个高离职率的工作。所以人们说”每个人都会失业,机器人会接管一切”,也许确实会发生一部分。但更可能的情况是通过自然流失来实现——也就是有人辞职了,然后那个岗位不再补人。所以末日场景我觉得不会真的大规模上演。但确实,人们很容易对此感到恐惧,我认为你必须主动拥抱它。
Lenny: 说得好。好的,我想聊聊框架。你提出了很多有趣的框架。那我们快速过一遍你用过并且觉得有用的那些框架吧。你之前提到了”before, after”(之前/之后),我之前没听过这个。这个概念的大致思路是什么?
框架:Before, After
Paul Adams: Before, after 字面意思就是这么简单。我们目前正在做品牌重塑,这将是一个 before, after 时刻。我们正在重新设计定价。然后定价上线的那一天,就是一个 before, after,因为一切都不同了。所以我们需要重新出去跟人聊。我非常相信交流。你必须跟客户交谈,这是唯一的方式。你得不断地聊、聊、聊,学、学、学。不要停留在表面,要深入挖掘。所以很多这样的 before, after 时刻,一旦你跨过去了,进入了 after,你就得开始学习——“我们是对的吗?是错的吗?发生了什么?人们怎么想?”
Lenny: 你能多聊聊你提到的这个定价方面的学习/失误吗?你觉得你们做错了什么?当时发生了什么?
Paul Adams: 我们有一条原则叫”让价格与价值对齐”。顺便说一下,我认为定价极其困难。很多参与定价工作的设计师,我对他们说,这是我所知道的最难的设计问题之一。我认为新用户引导(onboarding)也是一个。把用户引导进产品,同样是。人们觉得”哦,你就设计几个步骤,挺简单的,用户会跟着步骤走。“但设计出优秀的新用户引导体验,同样是看似简单实则极其困难。
定价的教训(续)
Paul Adams: 所以,我认为定价看似简单,实则极其困难。但我们有一条原则,就是让价格与价值对齐。用户应该根据他们在产品中获得的价值来付费——说起来容易,做起来难得难以想象。价值是主观的。同样是 10 个单位的价值,有的人觉得大概值 5 美元,另一个人却说”我愿意为这 10 个单位的价值付 5000 美元。“所以最大的错误,其实是很多错误叠加在一起。而且在这个领域,我觉得我们过于保守了。我们最终搞出了太多的定价模型,在旧的竞争性错误之上不断叠加。后来需要做出一个勇敢的决定——“我们要从头开始。”
Lenny: 哇,光是你这些定价的教训和历程,感觉就能单独做一期节目了。基于你的经验,能不能给目前正在思考定价问题的人分享一条核心智慧?
Paul Adams: 我要说的第一件事就是:保持简单。保持简单。这个诱惑太大了……以我们为例,很多 SaaS 产品都有附加组件(add-on),就是那种”嘿,我们做了 X 功能,这个 10 块钱。“或者 10 万,取决于你卖的是什么产品。“我们做了 X,这是 X 的价格。嘿,我们刚刚又做了 Y。Y 非常棒,是一个新功能,能解锁所有这些新的能力。用户不应该免费得到它,因为这是个新东西,以前没有。所以我们要为 Y 单独收费。但这跟其他的定价方式不太兼容……好吧,那就做个附加组件。哦,不错,用户直接加购就行。“但后来呢,现在你有的用户买了附加组件,有的没买。然后你又想,“再加一个东西吧。“于是我们又加了层级(tiers),有产品、有层级、有附加组件、还有附加组件里的分层。天哪。用户连自己的账单都看不懂。所以我的建议是保持简单。拼尽全力抵制增加额外定价方式的诱惑。
Lenny: 太精彩了。我没想到会聊到这个话题,但很高兴我们触及了它。
Paul Adams: 我之前说到”终身伤疤”来着。这又是一条终身伤疤。
差异化与基本门槛
Lenny: 好,我们继续聊一些框架。另一个我发现并且非常喜欢的是你所谓的”差异化与基本门槛”。这是关于什么的?
Paul Adams: 这类似于 Kano 模型,如果你了解的话。但它非常简单。我们把 Kano 模型拿过来,试图做一个极其简化的版本。再说一次,我对这类东西有点过敏。我甚至讨厌自己提起 Kano 模型。我对人们过度理论化框架这件事过敏。就像”哦,你看过新的某某定律吗……”不管什么定律。我的反应是,“保持简单、实用、务实。然后我们都回去干活、做产品,让客户受益,因为那才是唯一重要的事情。“所以,差异化与基本门槛,非常简单。我认为用户采纳一个产品、购买一个产品、或切换到一个产品,有两个驱动力。一个是新解决方案的吸引力,基本上就是差异化。也就是,有什么不同且更好的地方?但关键的是,要在客户在乎的方面不同且更好。
回顾所有那些失败的项目,我从中得到的教训是——我们在 Google 的那些项目里,确实做到了不同和更好,但人们根本不在乎。Google 有各种各样的项目,比如 Google Wave 是一个极其创新的产品,但没人在意。所以,要在人们在乎的方面做到不同和更好。这就是吸引力,就像”哦,我想看看那个。看起来很酷。我想试试,看起来比我现在的要好。“但在另一面,还有一个准入要求,也就是基本门槛(table stakes)。要上牌桌,你得具备一定的基础功能。这些就是基本门槛功能。它们往往非常无聊,是非常基础的东西、枯燥的东西,很容易被忽略,也很容易不去构建。
Intercom 多年来犯的一个错误就是,我们更被差异化吸引,做了大量这方面的工作。我们的路线图经历了不同的迭代,有时候在一两年内就会发生变化——我们把精力全放在差异化上,结果发现大家都很喜欢,很想购买,但他们买不了,因为我们缺少他们需要的基本报表功能,或者缺少他们需要的基本权限功能。然后机器人是基于这些来构建的……在为什么需要更多差异化和为什么需要投入更多基本门槛之间做权衡。所以现在,Intercom 大概是 50/50 的资源分配,但历史上两边都曾出现过 70/30 的倾斜。
最后一点是,我觉得审视路线图或拟定的路线图时,问自己一个问题非常有力量:这些事情中哪些对客户来说更重要——注意,不是对我们,而是对客户。我们内部讨论很多的另一件事是,如果你是一家创业公司,进入一个成熟的品类——对我们来说就是客户支持,一个庞大的成熟品类,大量的基本门槛,经过多年、数十年积累起来的。ServiceNow、Service Cloud、Salesforce、Zendesk,几十年积累的基本门槛功能。所以要上牌桌,你需要大量的基本门槛,除非你有极其出色的差异化。所以在 Intercom 早期,人们会把我们和 Service Cloud 或 Zendesk 并排购买。他们就是并排购买。他们说”这个 Intercom 的东西……”我们是第一个现代化消息和现代化 UX。他们说”我们想把它给我们的客户用,跟那个大包大揽的基本门槛一起用。“因为 Intercom 当时完全没有那些基本功能。
然后经过多年,我们把基本门槛建设到了一定程度,好了,现在我们可以完全上牌桌了,用户可以切换——他们可以用 Intercom 替换 Zendesk。但我们花了多年才走到那一步。因此,如果你是创业公司,你需要在差异化上投入更多。然后随着时间推移,我觉得你开始逐渐平衡这二者。
Lenny: 我觉得这件事有意思的地方在于,首先它给了你一种审视路线图的方式。我们实际在做什么?是不是做了太多基本门槛?还是做了太多差异化?它让你意识到正在发生什么。其次,作为一种创业策略也很有意思——“我们是要花几年时间做基本门槛然后再发布?还是走 Intercom 的路线,先差异化,其他以后再说?“我想知道什么时候该走哪条路。
Paul Adams: 对。而且这可能取决于市场、不同的品类,以及各种因素。对。
Lenny: 好的。下一个框架是你所谓的”摆动钟摆”。这是关于什么的?
摆动钟摆
Paul Adams: 我刚才其实已经提到了一个例子。差异化与基本门槛就是摆动钟摆。摆动钟摆的意思是,你从日常工作中退后一步,观察到一个处于不理想状态的情况。比如,“哇,我们有天下所有的差异化,但用户无法采纳产品,因为我们从来没做过那些基本门槛。这不行。“或者,“哦,我们现在做了这么多基本门槛,却没有在差异化上投入。实际上我们对人们没有吸引力了,因为切换产品是件麻烦事,而我们也不够吸引人。好吧,这个不理想的状态。”
Paul Adams: 然后你去修正它,但诱惑在于你会过度修正。我们在很多领域反复做过这种事,比如:“好,差异化不够。“一年后,“哦等等,基本门槛全丢了。好吧,又偏到那边去了。“产品构建是一个方面,人员是另一个方面——搭建团队和人才。还有一个大的方面,大概是 Intercom 成立五年左右的时候,我们正处于高增长轨道上,定价问题出现之前那种很典型的优秀创业公司。我们环顾四周说:“我们谁都没干过这个。我觉得这不好。这是个不理想的状态。我们到底知不知道自己在做什么?我们就是一群随便凑在一起的人。我们到底知不知道自己在做什么?我们需要招一些专家。如果我们要往上走市场,就需要那些做过的市场人才。”
所以那就是不理想的状态,修正方式是招那些做过的人。然后我们招了一大堆做过的人,而他们做的事情是把前东家的文化和工作方式带进了 Intercom。所以我们完全过度修正了,很多情况下效果不好。大多数情况下效果都不好。因为我们不是想成为一家已经存在的更大的公司,我们想做的是我们自己。所以我觉得,招聘和搭建团队是另一个我们真的过度修正了的领域,后来才发现:“好吧,这里需要平衡。”
与招聘相关的,还有一个是通才与专才,类似的主题。做过的人,或者专门化的人。我们招了一批专才,结果发现他们不够适应变化。在 Intercom,我们有很多模糊性,而且我们主动拥抱这种模糊性。那些高度专门化的人在大公司可以如鱼得水,真的表现很好,是不可替代的员工。但在一个充满模糊性的、流动的创业文化中,他们可能真的会溺水,真的很挣扎。也许这个钟摆的中点,落在中间的是:“让我们招一个做过一点、也有一点专长的人,不多,但足以试着摸索出答案。“所以我们现在招很多这样的人。
Lenny: 首先,我很喜欢这些失败的案例,因为很多人不愿意分享这些。而这恰恰是大家想听的——“并非一切都很完美。这一路上犯了很多错误。“感觉这个框架就是太多次这样做之后总结出来的。这里的主要教训是不是一般来说要避免钟摆摆得太远?因为有时候值得这么做,比如在 AI 的情况下,“不,我们全力投入。“或者在移动端,全力投入是值得的。我想知道你怎么看?
Paul Adams: 之前和别人聊这个话题时,有时候对话的结论大概是:这是唯一的做法。你实际上没办法用另一种方式来做。所以也许真正的问题是,钟摆摆多高?也就是说,你必须要摆,但摆多远?而且你说得对,在 AI 上,我们确实摆得挺高的。也许我之前高估了——如果用混合框架的说法,AI 是否完全属于差异化阵营,我们也在构建很多基本门槛特性,在产品中构建深度。这是五五开,我想我之前提到过五五开,所以是五五开。所以我们并没有完全摆过去。钟摆是摆了,但我们也在做另一边的事情,保持平衡。所以我觉得你可能必须得摆。这让我想起——要知道边界在哪里,这是我本来想说的。
知道边界在哪里
Paul Adams: 这让我回想起早年的一些事情。我记得在 Google,隐私问题被放在极其重要的位置,重要到会阻碍决策、阻碍产品推进。围绕隐私的循环讨论,无数的循环讨论,结果什么都没有做出来、什么都没有发布。我在 Google 做了一个项目做了一整年,一年里什么都没发布,就是循环讨论,这在当时让我痛苦万分。所以当我去了 Facebook 之后,我发现他们对待隐私的方式不一样。再说一次,我不是在提倡那种做法一定好,它肯定也没帮到他们的品牌。但有一种观点认为,要知道边界在哪里,你就得越过它。越过它是痛苦的。但如果你不越过它,你永远不会知道。因为如果你以为自己在朝边界走,然后提前停下了,结果发现边界其实在很远的地方。
所以我觉得很多事情上,你其实没有选择。你得越过边界,感受痛苦,然后足够谦逊地意识到自己没做对,再重新来过或者采取任何修正措施。
Lenny: 对,先把钟摆从那个均衡点上甩出去。然后我们来修正这个钟摆,把它放回来。
Paul Adams: 对。
Lenny: 好的。下一个框架我之前简略读到过,而且我已经很喜欢它的基本思路了,我觉得你把它叫做产品-市场-故事契合。
Paul Adams: 是的。
Lenny: 这是什么?
产品-市场-故事契合
Paul Adams: 是的,product-market fit(产品-市场契合)很基础,大家都很理解,也非常重要。我对 product-market fit 的描述是:你必须为正确的市场打造正确的产品。顺便说一下,我觉得没有足够多的人去思考这个等式中市场的那一面。很多做产品的人不去想市场那一面。但对我来说很简单:市场就是人,他们的问题,以及这些问题对他们有多重要。要有一个好的市场,你需要大量的人拥有相同的问题,而且他们需要在很大程度上关心这个问题。回到 Google 做社交产品的经历,我们找到了很多有相同问题的人,但他们并不真的在乎。他们并不真的在乎。他们现有的东西就够了。所以,大量的人有相同的问题,且围绕这个问题的能量很强,而产品就是针对这个的解决方案。市场是”谁”,产品是”什么”。
在我的职业生涯中,又出现了这种情况——做出来的一批产品,是好产品,市场也不错,但它们失败了,我想不通为什么。最终,我回到了一个想法上……也许有人会说:“Paul,你说的就是营销。“但故事——故事错了,或者故事缺失了。有时候,一个好产品在一个好市场里,却被解释得一团乱。这种情况我见得很多。在 Google 又是,我经常看到这种情况——被以一种非常复杂的方式解释,过度理性化。结果是大家一脸茫然:“什么?你在说什么?“你抓不住他们的注意力。所以故事真的非常重要,同样重要。实际上,有时候你会看到产品不那么好,纸面上明显更差的……我想记起来当年 Spotify 的竞争对手叫什么来着,大家都在说……叫什么名字来着?
Lenny: Ordio?
Paul Adams: 对,Ordio。Ordio 就是那种——
Lenny: 我挺喜欢 Ordio 的。
Paul Adams: ——对,我对 Ordio 唯一听说过的就是:“产品太棒了。”
Lenny: 嗯。
Paul Adams: 但它失败了。为什么失败了?Spotify 和 Ordio 面对的是同一个市场,解决的是同一组问题。Ordio 当时可以说是更好的产品。我不确定这是不是真的,但可以说是更好的。我也觉得 Spotify 是一个了不起的产品。但他们的故事搞错了。所以再一次,我觉得所有做产品的人,不管你是设计师、产品经理、做研究的、做数据科学的,都需要一直思考故事这件事。去做营销的工作,去做产品营销的工作,去学习如何解释产品,和学习如何构建产品同等重要。
定位与”你为什么更好”
Lenny: 嗯。这让我想到定位,以及它有多重要。最近我们在播客上请了 April Dunford,聊了很多这个话题。
Paul Adams: 对,对,她非常厉害。说到底就是,“你为什么更好,你能解释清楚你为什么更好吗?”
Lenny: 这一点太重要了。我想聊的最后一个领域是 Jobs to be Done。我们请过 Jobs to be Done 框架的联合创造者上过播客,也请过 Shyam Krishnan 上过播客。他们对 Jobs to be Done 到底有多有效,看法非常不一致。我知道你们对 Jobs to be Done 很重视。所以,你对 Jobs to be Done 框架的总体看法是什么?对你们来说效果如何?你们怎么用的?哪些有用?哪些没用?随便聊什么都可以。
Jobs to be Done 框架
Paul Adams: 好。我完全坦诚地说吧,冒着得罪人的风险——我们很多年前和 Bob West 合作过。Bob 是个好人。我们遵循的是他那一路的 Jobs to be Done 模式,而不是 ODI——ODI 是另一个流派的思路,我觉得是这样。总之,我试着用简单的方式来说。我们发现 Jobs to be Done 很好,非常有用。但是,是以一种非常简单的方式……还是回到简单框架这个思路,在一种简单的方式下使用。另外,有太多人花了大量精力去辩论某个版本中那些细微的差异和特殊之处。谁在乎呢?没人在乎。好吧,我不在乎。他们显然在乎。但你的客户不在乎。你试图为之构建产品的人不在乎,没人在乎。那是一场很有意思的智识辩论。但对我来说——也许这太极端了——它在我们的日常工作中根本没有位置。我们只是在努力构建一个伟大的产品。
所以对我们来说,Jobs to be Done 是一种非常好的方式,帮助我们聚焦于客户的问题,专注于不被干扰,基于扎实的、有研究支撑的洞察,来告诉我们人们试图完成的事情是什么。人们试图完成的事情是什么?还是那个词——能量。他们在这件事上有没有很多能量?能量这个概念,现在想想,可能就是跟 Bob 聊的时候学来的。我觉得确实是。就是这样一个想法:你需要找到那些对这个问题充满能量的人。而大多数时候你必须去采访他们,才能感受到他们的能量。一个人是冷漠还是很投入,非常容易看出来。
我们在这一点上收获很大。而且,我们偶然间发明了 job stories 这种写法。我记不太清具体是怎么回事了。但我写出了那种写 job story 的方式。其实,我们当时没叫它 job story,是别人这么叫的。我们当时就是……我甚至都记不清了。它就是一个触发器。总之,我们甚至没给它起名字,是别人命名的,我觉得。我就觉得,“我们只是在努力构建一个伟大的产品。” 所以我们在那方面一直做得很好,非常简单。另外,我们现在还在大量使用的另一个工具是四力模型(four forces),这是 Jobs to be Done 框架下的一个子框架。四力就是——当人们试图切换产品时,有不同的力量在起作用。其中一部分是差异化、基本门槛(table stakes)这些东西,比如新解决方案的吸引力,以及你可能不采用它的原因。习惯。人们会有焦虑感。
再讲一个有趣的故事,来说明四力模型有多好用。之前我说过 Eoghan 和 Des 试图说服我离开 Facebook——我当时非常喜欢 Facebook——加入他们。他们为我列出了四力模型,分析我加入的力量。然后,在私下喝啤酒的时候,他们跟我聊天,把我的焦虑一点点喂给我。基本上就是用四力模型来”运作”我。我当时就觉得,“这太天才了。这简直是天才之举。也许有点……但确实天才。” 所以,四力模型在帮助理解人们为什么做出决策方面,效果非常好。
Lenny: 我很喜欢你很多建议最终都归结为——保持简单,砍掉一切不必要的东西。我在 Jobs to be Done 上也有完全一样的感受。我觉得它作为框架对播客、对 Newsletter 都很有用,但我认为有无数种流程和优化方式让人分心,而且往往只是拖慢一切。
Paul Adams: 对,对。有时候讨论这些确实有趣,确实令人着迷——除非你是学术界的人。但如果你是在一家公司工作,试图为人们构建一个软件产品,以某种微小但有意义的方式改善他们的生活,这些都不重要。就用那个能帮你做到这件事的东西。目标就是这个。用那个能帮你做到这件事的工具,就这样。
闪电问答环节
Lenny: 说到这里,我们进入了非常令人兴奋的闪电问答环节。准备好了吗?
Paul Adams: 准备好了。
Lenny: 你最常推荐给别人的两三本书是什么?
Paul Adams: 我总是向所有人推荐的两本书,我办公室里有——一本是 It’s Not How Good You Are, It’s How Good You Want to Be,作者是 Paul Arden,他很久以前在广告行业工作。这本书非常棒。它让人们意识到,如果你用正确的方式思考,你会感受到无限的潜力——每个人都有。我推荐给所有人的第二本书,也是我买了送给别人的,是 Ray Dalio 的《原则》。我是 Ray Dalio 的超级粉丝。我觉得他不可思议。我坚信原则的力量。我们 Intercom 很多人都是……我总是推荐这两本书。而它们完全不同。Paul Arden 的那本,你二十分钟就能读完。《原则》则有那么厚。
Lenny: 你最近最喜欢的电影或电视节目是什么?
Paul Adams: 最近的是《熊家餐馆》(The Bear),我看晚了。我喜欢这个剧的原因是,我觉得它在某种程度上歌颂了那种苦干。我觉得这很重要。我年轻时在咖啡店打过很多工,靠这个供自己读完大学什么的。苦干是生活的一部分,苦干是完成事情、有时是成就伟大事物的必要条件。我喜欢它这一点。我真的很喜欢这一点。
Lenny: 你最喜欢问候选人的一道面试题是什么?
Paul Adams: 好,我给你一个稍微不同的回答。我并没有固定的几道面试题。我不喜欢那种依赖记忆的问题,比如”告诉我你上次做某件事的经历”。我也不喜欢答案千篇一律的问题。不过,我最近得到了一道很棒的问题,是以前在这里工作过的 Alyssa 给我的。我当时要做背景调查电话。就是你在面试一个人,你想给他发 offer,他有推荐人,当然,他给的推荐人都是他合作过的最优秀的人和他最喜欢的管理者。所以这道问题是:“在第一次绩效评估中,我会给这个人什么反馈?“这道问题太妙了,因为对方无法回避。一定有一个答案。而且它会揭示非常多的信息。
Lenny: 这是你在背景调查电话上问的问题?
Paul Adams: 对,背景调查电话上问的。
Lenny: 这问题太好了。我喜欢。
Paul Adams: 对,非常妙的问题。
Lenny: 好,真是个宝贝。谢谢你分享。你最近发现的、非常喜欢的某个产品是什么?
Paul Adams: 这可能有点取巧,但我会回到那些 AI 产品上。我觉得 ChatGPT Vision 令人震撼。我最近在玩 Rewind。我上手有点晚了。Des、Kiran,还有 Intercom 的一群创始人,都喜欢 Rewind,都在用它,都很喜欢。这东西太厉害了。所以我上手晚了点。但它就是一种增强记忆。令人震撼。所以 Rewind 挺好玩的。
Lenny: 他们刚出了一个可以录制你实际日常的音频小设备。
Paul Adams: 嗯,那个我不太确定。
Lenny: 嗯,被批评了不少。
Paul Adams: 是啊。
Lenny: 我也不太确定。说不好。我不知道那东西是不是真的。他们发布的时候看起来不像个真实产品,但我觉得应该是真的。
Paul Adams: 它试探性地触及了 AI 什么是可以接受的、什么是不可以接受的边界。不过这个想法本身确实很酷。
人生信条
Lenny: 有没有一个你经常回来、喜欢分享给别人、对自己也很有帮助的人生座右铭?
Paul Adams: 有。我显示器上贴了一张便利贴,上面写着”只做最重要的事”。就贴在显示器上,一张便利贴。有时候它掉下来,我就再写一张。只做最重要的事。效果很神奇。我到公司上班,有人给我发邮件,我就”天哪”。然后我看到”只做最重要的事”。第二个相关的信条是:别为你无法控制的事情操心。所以我有两条。只做最重要的事,别为你无法控制的事情操心。这就让一切降温了。同样是人生经验教训。我以前发过很多蠢邮件,比如”Red Energy,天哪,他们在想什么?“你在都柏林一觉醒来看到旧金山来的邮件,你就”天哪,想砸键盘”。如果你的显示器上写着这两句话,你就不会那样做了。你就深呼吸一下,去倒杯咖啡,回来再说。这事真的有那么重要吗?
Lenny: 好极了。第二条,我最早是从《高效能人士的七个习惯》里学到的。你读过那本书吗?
Paul Adams: 噢,读过。
Lenny: 就是关注你能控制的事情的圆圈,然后是你能影响的圆圈,再然后是你完全无法控制的事情。我自己觉得非常有帮助。我很喜欢你用便利贴的方式。我感觉我得把大家分享的这些人生座右铭全做成便利贴。
Paul Adams: 是的,显示器上贴便利贴是一个真正的生活技巧,我几年前发现的。说起来有点蠢。贴在显示器上的便利贴,它会挡住视线。
Lenny: 等等,你真的贴在显示器上挡住屏幕?
Paul Adams: 对,对。
Lenny: 哇。
Paul Adams: 在左下角,就盖住底部。因为如果不贴在那里,我就不会去看。我逼自己去看它。
Lenny: 真是的。我还从没听说过有人把便利贴贴在显示器那么宝贵的地方。
Paul Adams: 嗯。
Lenny: 确实管用。
父母的教诲
Lenny: 你爸妈教给你的最有价值的一课是什么?
Paul Adams: 最大的一个,同样非常简单朴素——就是对人友善。我觉得友善的作用远比人们真正意识到的要大得多。我这辈子学到的一件事,同样是通过教训学到的,就是你完全不知道别人生活中正在经历什么。你不知道。一个人可能正在经历各种压力极大的私人问题,而他在工作中做了那件你不喜欢的事,正是因为这些。所以我会试着想:“友善一点。你不知道发生了什么。也许以后你才会了解。别做出自己会后悔的事。“我觉得在人生中,友善的作用远比大多数人认可的要大得多,因为它太容易被当成一句空洞的老生常谈了。
Lenny: 我百分之千认同。有人说过我太友善了,我得变得不那么友善一点。但我始终丢不掉这一点。所以我完全认同。我父母也教过我类似的道理。
Paul Adams: 是的。有时候这很难。比如在加入 Intercom 之前,我从没开除过任何人。我真的非常不喜欢做这件事。后来,我在各种不同的情况下做过很多次,并意识到这对双方来说最终都会好起来。最友善的做法其实是做那个更难的决定。它实际上是更友善的做法。在这个例子中,对方会感到如释重负。这是更友善的做法。所以这件事其实挺复杂的。
Lenny: 说得真好。
爱尔兰美食推荐
Lenny: 最后一个问题。你是爱尔兰人,住在爱尔兰。有没有一道爱尔兰美食,你觉得大家如果去爱尔兰一定要尝尝的?
Paul Adams: 我能取个巧说健力士啤酒吗?那算食物吗?
Lenny: 当然算。
Paul Adams: 爱尔兰的健力士。人们经常谈论这个,而且是真的。爱尔兰的健力士好得多得多,原因有很多。它本质上是一种新鲜产品,而且是在本地酿造的。他们看待它的方式就像牛奶一样。牛奶会变质,健力士也会变质。健力士如果超过几天,就开始走味了。所以爱尔兰的健力士棒极了,因为就在本地生产。另外我觉得爱尔兰做得很好的是鱼。顺便说一下,爱尔兰多年来一直没有最好的美食声誉。我觉得爱尔兰菜在美国尤其不行。但这里的鱼太棒了。你能吃到非常棒的鱼。爱尔兰显然是个岛国,所以鱼类资源很丰富。
Lenny: 关于健力士,有什么办法能在爱尔兰以外喝到好的吗?还是说只能亲自去?
Paul Adams: 有的,确实有办法。你只需要靠近一家酿酒厂。健力士在尼日利亚也有酿造。尼日利亚有一个巨大的健力士市场。
Lenny: 这我还真不知道。
Paul Adams: 我觉得他们用的配方可能不一样,但确实在那里酿造。美国的酿酒厂应该在东海岸,纽约和加拿大东部之间的某个地方。所以纽约的健力士其实可以相当不错。旧金山的健力士往往就很差。我记得跟一个在健力士工作的朋友聊过这个。健力士的船要穿过巴拿马运河再北上到旧金山。所以运到那里已经是十二周陈了。
Lenny: 没想到我们竟然会学到健力士的运输路线——
Paul Adams: 至少这是我听说的。健力士有太多传说了,你真的很难分辨哪些是真的。但这些就是别人告诉我的故事。
Lenny: 太精彩了。Paul,你太棒了。非常感谢你来参加节目。最后两个问题。大家想联系你的话,在网上哪里可以找到你?听众可以怎么帮到你?
Paul Adams: 我有一个统一的账号名,到处都用。基本上是 P-A-D-D-A-Y。就是 Paddy 多了一个 A。所以是 P-A-D-D-A-Y。到处都用这个。paddy@gmail、@Paddy,所有平台都是这个账号。这就是你找到我的方式。我很欢迎大家联系我,真的,我是真心想学习。我很想听到那些觉得我的 AI 观点是胡扯、认为更像是 crypto Web3 的人的声音。或者有不同观点、能挑战我看法的人。这就是我喜欢学习和进步的方式。如果有人有这些观点,我很想听听。我很想和他们交流。
Lenny: 小心你许的愿。YouTube 评论区可是个火辣的地方。我们走着瞧吧。太棒了,Paul。再次非常感谢你来。
Paul Adams: 嗯,谢谢你 Lenny。我真的很感激。
Lenny: 大家再见。非常感谢收听。如果你觉得这期节目有价值,可以在 Apple Podcasts、Spotify 或你喜欢的播客应用上订阅节目。也请考虑给我们评分或留言评论,这真的能帮助其他听众发现这个播客。你可以在 lennyspodcast.com 找到所有往期节目或了解更多关于节目的信息。下期再见。
术语表
| 原文 | 中文 |
|---|---|
| add-on | 附加组件(add-on) |
| Alyssa | Alyssa(Intercom 前员工) |
| April Dunford | April Dunford(定位领域专家) |
| Bard | Bard(Google 的 AI 搜索引擎) |
| Bob West | Bob West(Jobs to be Done 领域的实践者) |
| ChatGPT Vision | ChatGPT Vision |
| copilot | copilot(副驾驶式辅助工具) |
| Des | Des(Intercom 联合创始人) |
| Entropik | Entropik(文中提及的 LLM 公司,疑为 Anthropic 的转录误差) |
| Eoghan | Eoghan(Intercom 联合创始人兼 CEO) |
| Fergal | Fergal(Intercom 前机器学习负责人) |
| Fin | Fin(Intercom 的 AI 聊天机器人产品) |
| first principles | 第一性原理 |
| four forces | 四力模型(Jobs to be Done 框架下用于分析用户产品切换行为的子框架) |
| Gartner Hype Cycle(文中提到的”炒作高峰、幻灭低谷”曲线) | 加特纳技术成熟度曲线 |
| Google Wave | Google Wave(Google 推出的实时协作平台,已停服) |
| Guinness | 健力士(爱尔兰黑啤酒品牌) |
| Horizon framework | Horizon 框架(亚马逊等公司用于规划创新投资的三层战略框架) |
| job stories | job stories(Intercom 团队发明的以任务为导向的用户故事写法) |
| Jobs to be Done | Jobs to be Done(以用户任务为导向的产品创新框架) |
| Kano model | Kano 模型(产品特性与用户满意度关系模型) |
| Kiran | Kiran(Intercom 联合创始人) |
| Lenny | Lenny(播客主持人) |
| Lex | Lex(Lex Fridman,播客主持人) |
| Matt Rickard | Matt Rickard(AI 领域博主/Newsletter 作者) |
| Metaverse | Metaverse(元宇宙) |
| mock-up | mock-up(模型/设计稿) |
| ODI | ODI(Outcome-Driven Innovation,成果驱动创新,Jobs to be Done 的一个分支流派) |
| onboarding | 新用户引导(onboarding) |
| Ordio | Ordio(Spotify 的早期竞争对手) |
| Paul Arden | Paul Arden |
| product market story fit | 产品-市场-故事契合 |
| product-market fit | product-market fit(产品-市场契合) |
| Ray Dalio | Ray Dalio |
| Red Energy | Red Energy(文中提到的某产品/公司名称) |
| Rewind | Rewind(AI 记忆增强工具,rewind.ai) |
| Seven Habits of Highly Effective People | 《高效能人士的七个习惯》 |
| Shyam Krishnan | Shyam Krishnan |
| table stakes | 基本门槛(table stakes) |
| Tableau | Tableau(数据可视化/报表软件) |
| The Bear | 《熊家餐馆》(FX/Disney+ 剧集) |
| tiers | 层级(产品定价层级) |
| Web3 | Web3 |
| Zuck | Zuck(Mark Zuckerberg 的昵称) |
| 《碟中谍》/ Mission Impossible | 《碟中谍》(电影系列) |
| 《黑镜》/ Black Mirror | 《黑镜》(Netflix 剧集) |
| 萨姆·奥特曼 | Sam Altman |
此文档由 AI 分片翻译(translate_long_document)
What AI means for your product strategy | Paul Adams (CPO of Intercom)
The Cannes Speech Story
Paul Adams: This is a meteor coming towards you. This is going to radically transform society. And I think if people don’t explore AI properly, it will leave them behind. I’d start with the thing your product does. “What’s the core premise behind it? Why do people use it? What problem does it solve for them?” That kind of thing. So, go back to basics. And then ask, “Can AI do that?” And for a lot, the answer is going to be, “Yes, it can.” For some it might be, “It can partially do it.” And then, maybe for others, “It can’t do that, at least not yet.” And then, for some of it’ll be replacement, AI would replace, it’ll just do it. And, in other places, it’ll be augmentation. It’ll augment. It’ll help people. But yeah, I think that you’ve got to match your product, and what AI can do, and what it will be able to do, and then ask yourself, “Okay, what are we going to do?”
Lenny: Today my guest is Paul Adams. Paul is chief product officer at Intercom, a role that he’s held for over 10 years. Prior to this role, he was global head of brand design at Facebook, a user researcher at Google, a product designer at Dyson, and his first job was an automotive interior designer. In our conversation, Paul shares some amazing stories of failure, including the story of him giving a huge presentation where he froze on stage and had to walk off. And what he learned from these experiences of failure. We then get deep into how to think about AI as a part of your product strategy, including a ton of great examples from Intercom’s experience going all in on AI. Paul also shares some of his favorite frameworks, and product lessons, and so much more.
And now, Hex’s AI tools can generate queries and code, create visualizations, and even kickstart a whole analysis for you all from natural language prompts. It’s like having an analytics copilot built right into where you’re already doing your work. Then, when you’re ready to share, you can use Hex’s drag and drop app builder to configure beautiful reports or dashboards that anyone can use. Join the hundreds of data teams like Notion, AllTrails, Loom, Mixpanel, and Algolia using Hex every day to make their work more impactful. Sign up today at hex.tech/lenny to get a 60-day free trial of the Hex team plan. That’s hex.tech/lenny. Paul, thank you so much for being here and welcome to the podcast.
Years at Google and Facebook
Paul Adams: Thanks, Lenny. Nice to be here.
Fully Embracing Failure
Lenny: It’s nice to have you here. I’ve heard so many good things about you from so many different people, so I’m really happy that we’re finally doing this. Also, you have an Irish accent, which is always a boost for ratings in my experience, so thank you for bringing that with you here.
Paul Adams: Yeah, that’s nice to hear.
Going All In on AI
Lenny: I wanted to start with a couple stories. So the first is your story of giving a keynote at Cannes. Can you share what happened there?
Paul Adams: Yeah, some things that happened in work are very memorable at the time and they don’t really scar you. This goes in the book that have scarred for life. Yeah, it’s good. Long story short, I was at Facebook just over a decade ago. Loved it at the time. I think it was a great place to be at the time. And, basically San Francisco, I did a lot of talks for Facebook internally and externally. Facebook had a keynote slot, always had a keynote slot at Cannes, the world’s biggest advertising festival. And, the year prior, Zuck had been interviewed. He was the speaker, he’d been interviewed. He’d gotten a hard time on privacy. It didn’t go well as well as they’d hoped.
So, the next year they asked me to do it. Maybe it was the Irish accent that made the offer come my way. And, yeah, I got out and spun a stage, the world’s biggest advertising stage. And, I’d say, I was three, four minutes into the talk, a very similar talk when I’d given lots of times. And, I just froze. I couldn’t remember what I was supposed to say. It was the first ever time in my life I’d rehearsed the talk word for word. Usually, I have talking points, and things get mixed around, and it’s informal. This was media trained, “Do not say the wrong thing.” Kind of talk. And I just could not remember what to say. I had some version of a panic attack, walked off-stage, I was still mic’d up, cursed. Everyone started laughing. I was like, “Geez, are they laughing at me? Oh my God, this is…”
But, I managed to turn it around, I walked back out. I’d been disarmed internally in my head. And, the most of it went well. And I was famous that night. Out in Cannes afterwards on whatever the sea front, it’s just like rose everywhere. And yeah, I was famous and infamous for my performance.
Thinking Strategically About AI
Lenny: I feel like you lived the worst nightmare that everybody has when they’re thinking about giving a talk. And, I think what’s interesting is you survived. And, I think that’s a really interesting lesson is you could freeze in front of thousands of people, walk off-stage, and then it works out okay.
Paul Adams: Yeah. And it all happened organically, I guess, or very naturally. But yeah, ever since then, every time I walk out onto a conference talk stage, still today, I have this tiny doubt in the back of my head. It’s never happened since. But yeah, I think you have to go with it with these things, when life throws you these, whatever, curveballs you have got to adapt and it’s not that big a deal. None of these things are that big a deal, at the end of the day. You move on and live and learn. So yeah, but I still hope it doesn’t happen again.
Intercom’s Full AI Pivot
Lenny: I also hate public speaking and I always fear this is exactly what’s going to happen to me. And so, I think this is nice to hear, that even when the worst possible thing basically happens, things can survive.
Paul Adams: You can turn it around. Yeah.
Impact After Going All In
Lenny: A second area I wanted to hear from is your time at Google. And, there’s a couple products you worked on at Google. Both of them were not what you’d call big successes. And then, there’s a transition to Facebook, which was also messy. Can you just share a couple stories from that time?
AI Reasoning and Code Generation
Paul Adams: Yeah. Similar to the walking on stage thing, you live and learn. And, I was at Google for four years now and I was at Facebook for two and a half years or so. And, in both of those companies, this is at the height of… The social tech wave was at its peak. Google were very afraid of the existential threat posed by Facebook. Facebook were very confident they could pull off some new social advertising unit that would be an AdWords or something like that, that would destroy Google’s revenue, eat them from the inside out. And so, being there at the time was fascinating and moving to the new companies. At Google, I worked on a lot of failed social projects, like you mentioned. Google Buzz, Google Ventilator, Google Plus. I think, a lot of the motivation for those projects came from a place of fear. It didn’t come from a place of, “Let’s make a great product for people. Let’s really understand the things people struggle with when communicating with family and friends. Let’s really, really try and create something wonderful.” It came from a place of fear.
And so, during those times, I learned I think how not to lead in places. And by the way, I should say, at the time in Google, there was other things happening that were amazing, like Google were building Google Maps, an incredible product. One of my favorite products. I think one of the best products ever made. They were building Android. I was in the mobile team and the mobile apps team at the time, the Android came out. So, they can make an incredibly good product. So, I just happened to be in the social side, which wasn’t as good. And, yeah, Google Buzz is a privacy disaster, and Google Plus is similar.
And so, halfway through I’d published research about groups and I’d done a ton of research. An interesting side note there is, at the time, I was working in the UX team as a researcher, I was been asked to do a lot of tactical research, like usability study type stuff, like can people use these products? And, I ended up doing a lot of formative research as well in the same session. So, I’d say to the team, “Hey, I’ll do the research. I’ll answer your questions. But also, I’m going to do this other thing, and I’m going to take 20 minutes doing that.” And so, what we used to do is, what I used to do with people was map out their social network, all the people in it, their family, their friends, how they communicate. We’d map on all the channels, we’d talk about what worked well, what didn’t. And, we did this with dozens and dozens of people over the course of maybe 18 months. And the same pattern emerged every single time, which was, people need way better ways to communicate with small groups of family and friends.
And I look back now and go like, “WhatsApp.” Or it may be iMessage if everyone’s on Apple. But, really obvious in hindsight. But at the time, not obvious. And so, we tried to build a product around that called Google Plus. But, again, it came from the wrong place. And so, halfway through, the research that I’ve done, all this research had been made public through a conference talk. And, Facebook noticed, got in touch, one thing led to another, and I left and joined Facebook, which was an amazing thing for me, personally.
Facebook was an amazing place at the time and exciting. And they were trying to do things for the other reasons, the good reasons. “Okay, let’s build an amazing product for people.”
Lenny: And this was during Google Plus being built, you basically shifted.
Vision and Image Recognition
Paul Adams: Yeah, midway, I’m stressed to even tell you about it. The project hadn’t been launched, it was still under wraps. It was highly confidential. Google had done a lot of things at the time that were the first for them. I don’t know if they’ve done them since. But things like, everyone worked in Google Plus was sent to a different building. That building had a different key card. If you didn’t work in Google Plus you could not get in. All sorts of counter-cultural things at the time. And, as a result, there was a lot of antagonism internally for Google Plus. And so, when I left in the middle of the project, leaving with all of the plans in my head to the enemy, some people saw me as a traitor, understandably. Other people thought I was enlightened, too fancy you talked to. But it was the right thing for me to do. But at the time, it was a hard thing to do.
Voice Cloning and Digital Replicas
Lenny: I know there’s also a lot of scrutiny in what you took with you and the process.
Paul Adams: Yeah, when I left, Google assumed that I was one of the spies. I was quarantined. I told them I was leaving. They forensically analyzed my laptop, all sorts of stuff like that. So, it was pretty intense. Looking back, I can understand why that happened. But the root cause for me is that the project has been run from a place of competitive fear, which I don’t think leads to good things.
AI’s Impact on Engineering Teams
Lenny: So one of the themes through the stories you just shared is, let’s say, failure is… I don’t want to make it that harsh, but just things not working out. And, I’m curious as a product leader, how important you think that is for people to go through, if you think that’s something that is almost a good thing? And, I guess just is there anything there that you find helpful as a coach, as a mentor, as two people that are trying to become basically you?
Don’t Bolt AI Onto Products
Paul Adams: Very, very. It still is. It still is. I’ve personally failed so many times. There are two stories and the Google one is long deep tentacles. They’re two stories. I failed a ton of times. I remember, when I was at Facebook I was very happy. And, I knew Eoghan and Des, the co-founders of Intercom. And, they were trying to persuade me to join Intercom. We were like, it was a 10-person company at the time. But, Eoghan said something to me at that time which has stuck with me ever since. He said, “At Facebook, you can design the product. But at Intercom, you can design the company.” And, that was extremely appealing to me, a great pitch. He’s like, “Just design the company with us that you want to work in.”
And so, part of that was a company that embraces failure, that says it’s okay to try things. I’m a big believer in big bets, high risk, high reward. I don’t get as excited about incremental things. No, I haven’t said that. There’s of course a place for that too, especially as companies get bigger. But, I get excited about big bets. And if you make big bets, you’re going to get a lot of it wrong. So a lot of the principles that we built here at Intercom are in building software.
We have a principle called Ship to Learn. And, we’ve actually changed it since. It’s over on the wall here. Ship fast, ship early, ship often is what it says now. You say Ship to Learn. Ship fast, ship early, ship often. So, in that idea is the idea of failure. It’s not going to go right. And, it’s going to go wrong more often than not. But if you ship early, and fast, and learn fast, you can change fast, and you can improve fast. And, that’s the culture that we, as much as possible, try to embrace and teach people. But it’s much easier said than done.
How to Keep Learning in the AI Era
Lenny: Yeah. Especially when you’re in the moment like, “God dammit. Everything’s going to fall apart. I really messed this one up.”
Paul Adams: Yeah. And there’s a trade-off with quality that people really struggle with. We’ve high standards of ourselves. A lot of Intercom comes from a design founder background. We value the craft a lot. We never want to be embarrassed by what we ship. So there’s a real tension there, a real trade-off, where people have these high standards, which we encourage. We encourage them to ship fast, and learn, and make mistakes. It’s a constant tension that we’re navigating.
Organizational Challenges and Convictions
Lenny: Speaking of taking big bets and going all in, I know there’s been a huge shift at Intercom to move towards AI and embrace AI. And so, maybe just to start broadly, I’m curious just what are some of your broader insights or surprises so far in how you’ve thought about AI and how you think AI will integrate into product and product strategy?
Your Roadmap Might Be Obsolete
Paul Adams: What day that ChatGPT launch? November 29th, I think, last year. Ever since that day, I literally wake up every day thinking about AI pretty much. And, I read as much as possible and still feel like I’m way behind in it. I think, for me, when I talk to you about AI, people typically fall into one of two camps. You’re either all in, really truly all in. This is a meteor coming towards you. This is bigger than mobile as a technology shift, as big as the internet. Maybe it’s bigger than the internet itself as a technology shift, the way it’ll shape society. So I’m all in. I’ve gone over the hill or whatever. I’m over the other side. And so, there’s people in that camp.
And then, I think there’s people in another camp, which is, “I’ve heard this before. It’s hype. Last year was crypto. It was Web3. None of those things worked out. There was the metaverse.” So, there’s definitely I think a lot of skepticism or maybe cynicism around it. And I don’t understand why. The other things didn’t really pan out. The metaverse is coming back. And, I’m trying to remember, there’s the law where you have the hype, and then the trough of disillusionment, and then you come out the other side.
Lenny: Yeah, that little curve.
Other Industries Face Disruption Too
Paul Adams: Yeah. And I think that’s where a lot of people might be, where there was so much hype, it was so noisy, and still is a little bit so noisy that you tune it out a little bit. And, I think, some people have fallen into that camp. I’m all in in the other camp. This is going to radically transform society and it blows my mind even seeing new types of things that come out, like ChatGPT Vision just came out recently, and just seeing the things that people can do with it. And we’re just scratching the surface still. So, we’re all in, for sure.
Building AI Literacy in Your Team
Lenny: Awesome. I want to unpack that. But, I think there’s also this camp of people that like, “Yes, something big is happening. I just don’t have the time to understand, to build, to play around.” What have you found and/or what advice would you share to people that are just like, “I want to go deeper down this rabbit hole. I just don’t know where to start, because I have so much work to do already and this isn’t a side thing.”
Don’t Fear AI
Paul Adams: The advice I have for people, and the advice I have for myself, I’m in that too, I wake up every day to too many emails, and Slack chats, and people knocking on my door, and my desk, and all things. So, this is a challenge for me too. You just have to take the time. There’s just no other way for me. And that to me doesn’t mean… It’s about priorities. It doesn’t mean that you need to work crazy hours. I don’t believe in working crazy hours. I don’t know what hours I work. I don’t know, 50 hours a week maybe. I think, beyond that, you start to make bad decisions and things like that. You get tired. And you need to live the rest of your life. You got to put it into your day. Whether that’s setting aside dedicated time to read.
Reading is the thing. You got to read. You got to stay up to date, and you got to play with things, and try things. If you don’t have ChatGPT… If you don’t have a… I can’t remember if it’s a pro licenser, whatever, but if you haven’t upgraded to get access to things like GPT for Vision, where you can take photos and you have the mobile app. And I was going out for dinner last Friday night with my wife. I try not to take work to dinner with my wife. But, I wanted to try it. And, I took some photos of her food. And, you can do all sorts of crazy stuff, like tell you how healthy the meal is or whatever.
Lenny: Oh, wow.
Framework: Before and After
Paul Adams: Anyway. You got to try it. You just got to try it. So, my advice people is, you’ve got to try it. You’ve got to set aside the time, or it’ll pass you by. It does remind me the mobile wave about a decade ago. Again, I was at Google at the time, I was working on the mobile team. So I guess, it was my job to stay on top of things. But, at that time, some companies like Facebook went all in on it, maybe a bit late, but they eventually made the brave decision. I think if people don’t explore AI properly, it will leave them behind.
Lenny: It reminds me, I think, at Facebook, Zuck, and also Airbnb, Brian did this, is he said, “Any mocks you show me for new product designs have to be in a mobile app or on a mobile web. They can no longer be desktop for now.”
Pricing Lessons (Continued)
Paul Adams: Right. Yeah. Same with Facebook. Yeah, that’s right.
Differentiation vs. Table Stakes
Lenny: I guess, do you think that that’s the way to approach this is as a leader, just, “Everything you bring me needs to have some AI component.” That sounds probably not like a good idea, but is there something that you’re thinking about, or have done of just convincing people this is where you want to spend your time?
The Swinging Pendulum
Paul Adams: Yeah, it’s harder, for sure. It’s harder, because-
Lenny: You don’t want to force it.
Knowing Where the Boundaries Are
Paul Adams: … Yeah, a lot of the tech is invisible. We have a machine learning team we’ve had on here for a long time, so we’ve been working in this space for quite some time. But, it’s funny, even if you go back 18 months, I think if I was on your podcast 18 months ago and you said to me like, “Hey, what do you think about AI?” I would’ve said something like, “It’s not real. Machine learning’s real, let’s talk about that.” So, things change, and my perception of it’s changed. But a lot of the improvements are behind the scenes. They’re with large language models or different types of things people are building in the background of infrastructure.
So I don’t know what it looks like to design mobile mock-ups that are AI mock-ups. But I do think that people need to start really thinking strategically. Maybe it’s just not a mock-up stage, but start to think really strategically about their product and whether it’s in the line of the media, or it’s coming or not. It’s not everything is. And if so, for some I think they require a foundational strategic change. Others, it might be less so. But, I think that’s actually the head space that I think people need to be in.
Lenny: Can you impact that further? What does that look like to really think deeply about whether your product is in the way of the meteor?
The Product-Market-Story Fit
Paul Adams: You can get sidetracked by the technology, for sure. And I do. I just mentioned, hey, going out for dinner and taking a photo of my food. You can get sidetracked by the tech and some of it’s really cool. I wouldn’t start there. I’d start with the thing your product does. What’s the core premise behind it? Why do people use it? What problem does it solve for them? That kind of thing. And then, ask the question. So go back to basics. “Okay, what is my product for? And why do people love it?’ And then ask, “Can AI do that?” And for a lot the answer’s going to be, “Yes, it can.” For some, it might be, “It can partially do it.” And then, maybe for others, “It can’t do that, at least not yet.”
So you’re going to need to map what your product does against what AI can do. And AI can do a lot. It can write. I’ll give you a list. It can write, it can summarize, it can summarize text, it can write text, it can answer queries, it can find facts, it can scan text, it can scan images. It can listen to your voice and repeat it. It can take actions. That’s the next big thing coming. It can take actions, actually do things. It could like, I mean, “Hey AI. Whatever the AI is called. “Change my flight to Tuesday.” Right? It can do things like that.
And so, it can do a lot of things. It can build rules. So, I think any product that has any workflow in it, which is almost all B2B SaaS products, any product that has multimedia in it, they’re in the media line or whatever. I don’t don’t know if this metaphor is working. But, the media is coming and they’re in its path. And so, for a lot of these products that you just need to look at what AI can do. And then, for some of it’ll be replacement. AI would replace, it’ll just do it. And, in other places it’ll be augmentation. It’ll augment. It’ll help people as the copilot ideas that are going around. But yeah, I think that you’ve got to map your product, and what AI can do, and what it will be able to do, and then ask yourself, “Okay, what are we going to do?”
Positioning and Why You’re Better
Lenny: Is there an example of that at Intercom or a different company of, “Here’s a problem we’re trying to solve? Oh, AI can actually do this fully for us.”
The Jobs to Be Done Framework
Paul Adams: Oh, yeah. I’ll give you Intercom first. Again, this date, I think it was November 29th, etched in our head. We have Fergal who was our head of machine learning. And, Fergal just turns around that day and he’s like… Okay, I think he tweeted something actually. He had a tweet that day that was like, “This is it. This is the time. This is the moment. This is the before after.” I actually often talk about people… because this is a framework I have, before, after moments. This is a before after moment. That was before. And that is after. And everything has changed. So, we literally ripped up our strategy almost entirely, and started again, from first principles and said, “Okay, why do people use Intercom?” Intercom is a customer support product. And then, very soon after that, Sam Altman, who’s the founder and head of OpenAI, said, “Hey, one of the first industries that’s going to be disrupted is customer service.” We’re like, “Yep.”
So we did. We totally changed how we think, how we work, and we just went heads down and built a product called Fin. We built other things first actually. Fin came later, now that I think about it. But we went all in on it. It was a little bit of a bet the farm mindset. So we’ve done it. I think other companies like Google and Bard have to do it, and maybe they’re a little bit slow, but it’s so early in this tech cycle that, I think, they’re fine. So yeah, we did. It was hard, but we had to do it.
Lightning Round Q&A
Lenny: Can you share briefly what Finn is just for folks that aren’t familiar?
Paul Adams: Fin, first and foremost, is an AI chatbot. So, if you think about customer service, people have questions for a business, and historically, that was mostly email, and phone, and mostly ticketing based. You’d file a ticket, a lot of do not reply email, and so on. And then, came along conversational customer support, which is just basic messaging, like WhatsApp or iMessage, like I mentioned earlier. Now, there’s bot first experiences and Fin is an AI chatbot, AI first, chatbot first. So the first line of defense for a customer support team is Finn, not a person. And so, it fundamentally changes. The results we’ve seen with Fin are mind blowing. Our biggest challenge is actually trying to help customer support teams think about organizational change.
The tech is way ahead. It’s actually people wrapping their heads around what this means for the role, the teams, loads of cool stuff, like new types of jobs for people, like conversation designers, a job we have where you design the conversations that Fin does or managers. So anyway, that’s what Fin is. Fin has expanded. So, Fin is now also in our Intercom inbox. They’ve placed a people answer queries, customers support queries, and now Fin’s in there too, helping the support reps. Suggesting answers for them to use, or helping them rephrase things. So, it’s now augmenting people as well as answering questions by itself.
Core Life Philosophies
Lenny: I think you’re one of the few companies that has pivoted fully into AI. And, I think there’s a lot of lessons here about how team structures might change, product strategy, priorities, things like that. So I’m curious just to unpack a couple more things here. First of all, what impact have you seen after going all in and going in this direction?
Paul Adams: It’s very early, honestly, to be able to answer that properly. And it depends what you measure as success. So, again, there’s a lot of hype and buzz with AI. So, if you’re measuring it by interest, it’s a huge success. Our target customer is customer support. Our customer support manager leader. And so, they’re very curious. They’re like, “Does it actually work?” Again, back to the earlier thing of there’s so much hype, there’s a bit of skepticism around it. “Does it actually work? Is it as good as a person?” And in customer support, people who tend to work in that role are typically very high empathy, care a lot about people. And so, they’re like, “But is it as good as a person? Is it nice, friendly? Does it understand humanity?” And so, a lot of curiosity, and a lot of interests, and a lot of people trying it.
We have some customers who are hugely successful with it. They can answer up to 50, 60, 70% of their inbound questions with Fin. So we’ve some customers who see huge success. But it’s early. And so, has it transformed our business financially? Not yet. I think, all fast-growing startups… If you think of AI Intercom as, I guess, a new startup, even though we’re 900 people, the growth curve, you’re looking for this exponential curve, as opposed to big public company linear growth curve. With the exponential one, it takes a while. The first year or two years is the bottom of that. And so, I think we’re still in the trying to figure out exactly what’s going on, trying to talk to educate people. But, we have enough evidence to believe it’s the future for sure.
Lessons From My Parents
Lenny: Are there any examples of either this product or other instances of AI just blowing your mind where you’re just like, “Wow, I never imagined it would be this good”?
Paul Adams: I go back to that before after thing. So, the first version of ChatGPT was a before, after, where we we’ve been working, like I said, in this space, we’ve had a machine learning team for a long time. The way our machine learning thing worked before ChatGPT was that there was not a manual setup. A customer support manager would have to orchestrate the bot, and teach it what to say, and just a lot of orchestration, a lot of teaching it. And then, ChatGPT showed up and it’s like, “Oh, it can do it by itself.” It gets it wrong sometimes. So, do people get the question wrong too? It’s as good as a person nearly for a lot of these basic things. So that blew my mind. And then, that was, “Oh, it can answer questions.” But then, you’re like, it can reason.
There’s actually a debate about whether is this reasoning or deduction. But, it can work things out. And, I’m not one for going down into these really philosophical things. I’m like, “We just need to build. Let’s go back, build the product.” Or whatever. But it can work things out. And that blew my mind. And, we fed ChatGPT and other companies too, we played with other LLMs, like Entropik and so on, it can work things out. And that was mind-blowing. Then you can see it doing things, like writing code. And I was like, “Wow, it’s really good at writing code. What does that mean?” And then, you start thinking, here at Intercom we have a one to five ratio. So a PM has about five engineers on a team. And you’re looking at this thing writing code and you’re like, “What happens next? Do we need as many engineers or will their role change? And they’ll start doing different types of things like reviewing code instead of writing code?”
So that blew my mind. And then, the visual stuff, like I mentioned earlier, I think the visual thing was bigger than the original one. It can parse imagery, and it can help you see the world. You take a photo of your bike and say, “Hey, what’s wrong?” And It’ll tell you what’s wrong, how to fix it. You can be traveling, take photos of stuff. It’s in a different language. It’s etched in stone on a 12th century cathedral. You’re like, “What does that say?” And it’ll tell you what it says. It’s just like how to do that. This is what I’m actually repeating most to people these days, here in Ireland, if you want to be a radiologist, so study X-rays and tell people what’s wrong, and so on, and forth, it’s seven years training to learn that skill. So, seven years to be a radiologist, and then you’re just into the job. AI, it seems it’s already better at it. So, it’s already better at it, and it can ingest every X-ray ever made. No human can ever read, and think about, and synthesize every X-ray ever made.
So, of course it’s better. And then, you’re like, “Okay, what happens now?” I guess, the whole job changes. Radiologists will not take x-ray. Well, I guess they might take them. But, they won’t analyze them, for sure. They’ll look at what AI says, check that it’s right, and then it’s bedside manner time. Tell the patient, maybe tell them what course. So the job just fundamentally changes. And by the way, that could be amazing. Here in Ireland, we have long queues for hospitals, epic waiting lists for people getting X-rays. So, this is a really good thing possibly for people. Here’s the craziest one I have. AI can listen to your voice and copy it, so it can say things and it sounds exactly like you and it’s really, really good. Almost in distinguishable. You’re like, “That sounds like Paul.” And so, I mentioned the Metaverse earlier. I don’t know if you saw Zuck talks to Lex [inaudible 00:32:35]. See that?
Irish Food Recommendations
Lenny: Yep.
Paul Adams: So that was my first, “Oh.” For people who haven’t seen it, they met in the Metaverse, I think, or some virtual world.
Lenny: It was a black room.
Paul Adams: In a black room. Yeah. And, the tech has come on so they can analyze your face and build a 3D model. It’s really good, really, really close. So, you can imagine, that’s going to get better. Based on the trajectory of that technology, it’s going to get better. And so, the voice thing and the face thing means both of those things are almost indistinguishable from a real person. And, AI will be able to ingest all the things people say and do. And, when people die, it’ll be able to replicate that person. And so, there’s an afterlife, hey, your parent dies and you can still talk to them. And, that could be the weirdest thing. Maybe it’s not good for people. I don’t know. But, that tech is just around the corner. And the AI can answer your questions, mind-blowing. It’s mind-blowing.
Lenny: There’s actually a Black Mirror episode with that same premise, where-
Paul Adams: That’s right.
Lenny: … Yeah. And I don’t think it ended well.
Paul Adams: No.
Lenny: Be careful.
Paul Adams: For sure. For sure. Yeah, I think, the [inaudible 00:33:48] and the voice translation thing is another one. I can’t remember. Maybe it’s in Mission Impossible, where it can take a voice, translate it, and translate it in real-time. And this tech is, again, just here, where if I was a native Spanish speaker and couldn’t speak English, you and I could still have this podcast. Your voice would be translated in Spanish in real-time for me. It’s, again, mind-blowing.
Lenny: We’re actually working on dubbing/translating podcast episodes, which is all done through AI, where it figures out what you’re saying, makes it Spanish, and then also changes your lips to match. And, we’re trying to launch a couple of those. And that’s actually very AI-based. Yeah.
Paul Adams: That’s cool. That’s really cool.
Lenny: You mentioned that your ENG team might change your thinking, because AI can make them much more efficient and work differently. I’m curious what you’ve seen actually change on your team, either using AI-ish tools, or just building AI products. What do you think is most different? And I’m curious from the perspective of a team that’s trying to think about integrating AI and starting to lean into AI, what have you seen most change and should change?
Paul Adams: Ultimately, you need really great machine learning engineers. That’s where it starts. And if you don’t have that, then you’re going to find it hard to build truly, really, truly great things. So, what OpenAI provide, and what Entropik provide, and Claude, they provide an amazing technology, but you got to build on top of it. If you really want something brilliant, you got to build on top of it. So, we adapted what they build for customer support. Maybe someday we need to go build our own LLM that’s just for customer support. Maybe. I don’t know where that will all go. And maybe everyone will have their own LLM for every single business. I don’t really know, to be honest. Maybe these companies will provide specialized LLMs. But anyway, that’s the first thing. And, of course, these people are in high demand. So, you need to invest in building out that function, I think. Really invest in building out the function.
So that’s what we’ve been doing. Our ML team’s way bigger than it was and way bigger than it ever has been at Intercom. And then, it forks. So, some projects are very heavy on that ML team and it needs them. But other projects are more front end, like the inbox stuff I mentioned earlier, where we have Fin and Fin is working, we’ve built the underlying technology. Now it’s a question of if you have a human support person answering questions in the inbox, that’s a natural chat conversational interface, pretty straightforward. What happens when there’s now an AI assistant in there? How do they talk? And what do they do? And when do they interject? And how do you represent that in the user experience that feels natural? So that’s a really hard design problem.
So, saying back into like, okay, we’ve a product team that’s a product manager, a product designer, maybe three, four, maybe five engineers, and they’re getting help from the machine learning team. So, we now have both setups. And increasingly, we can do more with the latter, more teams who can build on the foundational technology that we’ve been building over the last 12 months or so. So that’s one thing. I think a second thing that comes to mind is not to think about it as bolted on. I think some people are still in that camp.
Again, I’ll go back to the mobile thing. There’s just so many direct parallels with it. Like I said earlier, at Google, I worked in the mobile apps team. I worked on mobile Gmail, mobile docs, and it was the mobile team. And we were in London. We’re like, “Hey, we’re the mobile team in London.” And meanwhile, over in Mountainview in California, no one cared. It’s was like, “You’re 20 people. We’re 200. No one uses this stuff on a phone.” And again, a lot of skepticism. “No one’s going to write docs on the phone. Seriously? They’re going to write a full document on a phone, are you crazy?” So, don’t do that. We’re trying not to do that. Don’t bolt it on. Don’t be like, “Oh, we’ll have a bunch of AI people…” And we do have some specialists. But generally speaking, we’re trying to have everyone learn about it.
Lenny: Interesting. So, I’m curious just specifically what that looks like, don’t bolt it on. The idea there is don’t just have a site team that’s like, “They’re the AI team. They’re going to add AI to all this stuff.” You’re finding and lesson is integrated into every product team.
Paul Adams: And we’re still early there. We’re still early. So, what we’re trying not to do is have the AI inbox team, and they’re the only people who work on AI features in the inbox. I think it’s much better to have everyone learn about it. By the way, I’m a big believer in generalists, a big, big believer in… I guess, my background is jack of all trades master of none. That’s probably how I describe myself. I’ve worked as a researcher, designer, PM. And so, I believe in generalists, and so I believe in setting teams up that way. And, yes, specialists matters at times. Machine learning for sure is a deep specialism. And in Intercom, we generally, in engineering too, much prefer people who learn new things, whether it’s a new coding language, or framework, or how to design AI interfaces, or whatever, get more people being able to do it.
Lenny: I feel like, again, your company is a little bit of living in the future, where a lot of companies are going to get to once they realize, “Oh shit. We really need to get big here.” Or they’re already working on it. I’m curious if there’s other maybe pitfalls you ran into that you think people should try to avoid and something you could share there, or just any other lessons about making this transition that you think might be useful to other people.
Paul Adams: Yeah, what I’ve mentioned so far, don’t bolt it on. Stay up-to-date. I mentioned earlier, read, read. I feel like I’m behind all the time. It’s moving so fast.
Lenny: What are you reading? What do you find is most interesting and informative for reading about what’s happening in AI?
Paul Adams: I’d love to tell you that it’s incredibly structured. I have a great reading list that I got to read every Sunday morning. It’s pretty random. I’m on Twitter, which is now called X, of course, a lot. I follow some people on Twitter. I actually use the recommended feed in Twitter a lot. I think, because I interact and look at a lot of AI, I get to see a lot more. So I do that and I do it deliberately to try and generate more stuff. I’ll search Twitter as well. There’s loads of cool stuff there. There’s some newsletters as well and some people I follow.
Lenny: Any newsletters you could call out that you think are most interesting?
Paul Adams: Yeah, Matt Rickard is one guy who talks a lot about AI. The blogs of companies too. OpenAI have a pretty good blog, and they write papers, and summarize them.
Lenny: Cool. If there’s any other ones you think of, either people on Twitter to follow or newsletters, email me after, and then we’ll add them to the show notes.
Paul Adams: Yeah, perfect. Yeah, yeah, there definitely is. I’ll dig them out. Your question earlier, how do you do it? You just try. Try book out half an hour and just go deep for half an hour, and then bookmark a few things, come back to them. Like everyone, you could be so busy, so many distractions, you just got to have to set aside time.
Lenny: Are there any other tools or apps that you find really helpful? Sounds like ChatGPT is at the center of how you play around with it. Is there anything else that you find really interesting?
Paul Adams: I’ll try other things like Bard. For example, Bard is Google’s AI search engine. Rewind is another fascinating company. I think it’s rewind.ai. Rewind is basically augmented AI for your memory. So, install it on your local machine, and it captures everything, and remembers everything. It’s all local, so there’s no privacy issues. And, you got to try these things to understand whether it’s any good, or useful, or where’s the boundaries, and how does it work, and so on. So, I’m a believer in that type of thing.
Lenny:
The best products today use Command K for in-app search and navigation. HelpBar makes that readily available within your app without engineering or new code. Give users a faster and more delightful self-serve experience that reduces friction and increases in-app engagement. Upgrade your user experience with this modern component and supercharge your product-LED motion. Sign up for HelpBar today. It’s free and easy to set up in minutes. Check it out at helpbar.ai/lenny, that’s helpbar.ai/lenny. When you started rolling out AI and leaning into this direction, did you run into any big challenges or hurdles organizationally, or personal interests, or opinions? I don’t know. Is there anything you ran into that was a big stumbling block and something you had to get over?
Paul Adams: Yeah, Intercom is full of diverse opinions about things. And, I think with AI, I’m all in. I’m leaning forward. The media is coming. I’m sold. I’m way past that point. Also, no one knows. No one knows. And so, a lot of the time, when we talk internally, the strong buy-in from Eoghan, our co-founder and CEO, Des co-founder, like me, like a lot of the senior leadership team we’re in all in camp. And so, that helps a lot. Of course, if you’re senior leadership team in the company are all in, of course, then it trickles down. But equally, some of the hurdles have been like, “Why are you all in?” And I’m like, “An educated guess. A hunch.”
The part of business strategy and product strategy that, it’s just hard. It’s like taste. People talk about product taste, “Who has product taste?” And a lot of it is, it’s judgment based on experience. That’s all I can say. I don’t know. For me, personally, I don’t know, I lived through the mobile thing pretty closely, having worked at Google on mobile. I lived through that phase. So, I can see the same type of thing happening now with bigger. So I’m using that experience to go all in.
But it’s a challenge for some people, because they don’t have that context, or they disagree with it. We have a lot of debate here about the future. Fergal, I mentioned earlier, gave myself and a few other product leaders and Des he gave us a… I don’t know, is it a pitch or what? A play? I don’t know, about how maybe all of our roadmap with AI is wrong. I don’t know if you are familiar with the Horizons framework of Horizon 1, 2, and 3.
Lenny: Mm-hmm. Yeah. Amazon.
Paul Adams: Yeah. So, Horizon 1 is the medium short to medium term, next 12 months, 12 to 18 months. Horizon 2 being like, “Hey, what’s happening?” Whatever, 18 to 36 months out. Or, I think, people use different timeframes, different Horizons. Anyway. We’re in Horizon 1 land. We’re like, “Yeah, and the next year we’re going to do this.” And he’s like, “Yeah, but two years from now, if this path plays out, everything we’re doing now is going to be irrelevant and useless.” And you’re like, “Oh, okay.” And so, those discussions happen. And, the level of ambiguity is off the charts. So, a lot of the challenges have been navigating that ambiguity and helping people get the conviction I have without drying out voices of alternative voices and opinions, which are often valid too.
Lenny: What does help people get that conviction? Is it just showing them examples of, “Here’s something.” “Wow, look at this thing. This is unreal.” And, I think, partly what helps, I imagine, is the market you’re in seems like such a clear opportunity for AI, feels like an easier pitch than maybe a lot of other markets.
Paul Adams: Yeah, that’s true. For sure. That’s true. Yeah, showing people is definitely the easiest way. I think customer support is definitely… Like I said, [inaudible 00:46:20], number one, customer support. So you’re like, “Okay, I guess we should adapt.” Adapt or die is our mantra. Adapt or die. I think that there are other industries where they’re on the same journey, it’s just not as obvious. So for example, reporting software, Tableau or any reporting product, how do they work? Well, they’re the typical read, write app, build dashboards, filtering, querying, hardcore querying, query database, get some numbers, show it in a UI. A lot of thought and care goes into how you present that data to people. The different types of charts that are appropriate help people make good decisions ultimately.
I think, again, this is hand wave, who knows. Maybe that’s all done dead now. And, the reporting product of the future is just a box, and the box just goes to the database, and the box is just, “Who was our best salesman last year January? Okay. Who was our top performing representative in January? Lenny.” The report product to the future might look like that. And so, project management tools is another one. There’s a bunch of products that I think are just outside the most obvious customer support one. And yet, equally ripe for a newcomer to come with a completely different paradigm and potentially take over.
Lenny: I like that this connects back to your very first point about trying to think about where AI integrates is. Think about what problem are you solving as a company. For example, Tableau, helping people visualize data. And then, the question is, can AI just do this for you? And in that case, oh, and maybe you can. And that gives you basically a whole strategy of like, “Okay, how do we actually do that with AI?”
Paul Adams: Yeah. And, I don’t know if the reporting thing will play out that way. But, if you’re a Tableau type company, you’ve tons of designers who design dashboards, and filters, and querying type workflow. What do they do? The UI is the box. So, it’s really hard to get into your head like, “We must…” If you have conviction that we must change really hard.
Lenny: Maybe one last question here. For team members learning and starting to work within this realm, is there anything you find helpful to get them ramped up, other than the advice you’ve already shared, which is just read a lot of stuff, watch Twitter/X, subscribe to these newsletters, and then just try it?
Paul Adams: I also try and read things that say it’s all a load of crap. So, it’s very easy… I’ve been guilty of this many times. Back to the mistakes you’ve made. I’ve been guilty of this many times, where I’ve jumped on a bandwagon and it was all wrong. And the older I get… The Web3 thing, I’m like, “I don’t even know what Web3 is.” Crypto, I never bought crypto. Maybe I’m wrong about that. But, I’m not a bandwagon jumper. But, maybe might’ve been when I was earlier. And I try these days to read the alternative opinion. People who are skeptical or think it’s bad. A lot of people think this is terrible for humanity. This technology is going to eat us alive. So, I try and balance my optimism. I’m a delusively optimistic thinker, so I try and balance that with a negativity, I guess.
Lenny: That’s really good advice.
Paul Adams: Yeah.
Lenny: Is there anything else in this realm that you think might be useful to share before we shift to a different topic?
Paul Adams: Oh, yeah. The other thing is, don’t be afraid. I think people are a bit afraid of it. And, for example, if I started walking around our office here saying, “Hey, I think we need two engineers per team going forward.” That’s probably not really a good idea to do that. And I think in reality that’s not going to be how it plays out. I just feel like there’s loads of great studies over the years about how people don’t end up losing jobs, the jobs get moved around. And also, for customer support, for example, it’s a high attrition job. So, people saying, “Hey, everyone’s going to lose their job. A bot’s going to take over.” It’s like, maybe some of that will happen. But probably to attrition, as in someone quit and just didn’t get back-filled. So, the doomsday scenarios that I don’t think would play out as much. But, for sure, it’s easy to be afraid of it. And, I think you have to lean into it.
Lenny: I love that. Okay, I want to chat about frameworks. You have a lot of interesting frameworks you’ve put out there. So, maybe we do a rapid fire through a number of frameworks that you’ve worked with and find useful. And, you actually mentioned this before and after, which I hadn’t heard about. What’s the general idea to that concept?
Paul Adams: Before, after is literally that simple, I think. We’ve a rebrand at the moment happening, and that’ll be a before, after moment. We’re redesigning our pricing. And then, the day that pricing goes live, that would be a before, after, because nothing’s the same. And so, we need to go back out and talk to people again. I’m a big believer in talking. You got to talk to customers, it’s the only way. You’ve got to talk, talk, talk, learn, learn, learn. Don’t take with the safe face value, go deeper. And so, a lot of these before, after moments, once you’ve passed, yeah, into the after you got to start learning, “Were we right? Were we wrong? What happened? What do people think?”
Lenny: Can you talk more about this pricing learning/mistake you shared? What do you think you did wrong? What happened there?
Paul Adams: We had a principle called align price to value. By the way, I think, pricing is incredibly difficult. A lot of the design team who work in pricing here, I say to them, it’s one of the hardest design problems I know. I think onboarding is another one. Onboarding people into a product is also. People are like, “Oh hey, you just design a few steps and it’s pretty easy. People will follow the steps.” Again, deceptively difficult to design great onboarding.
So, I think pricing is deceptively difficult. But we had a principle around allowing price to value. People should pay based on the amount of value they get in the product, easy to say and incredibly hard to do. Value is subjective. The price, for some person they get 10 units of value. I think that’s about 5,000 for those 10 units of value.” So, the biggest mistake was a lot of mistakes compounded. And, this is an area where I think we were risk averse. We’ve ended up with too many pricing models. We’ve built on top of old competitive mistakes. And, it took a brave decision to say, “We’re going to start again.”
Lenny: Wow, this feels like it could be a solo episode, just talking through your pricing lessons and journey. Maybe just is there a nugget of wisdom you could share for someone that’s trying to think about pricing right now based on your experience?
Paul Adams: Number one thing I would say is keep it simple. Keep it simple. It’s so tempting to… With us, for example, a lot of SaaS products have add-ons, where you’re like, “Hey, we built X and that’s 10 bucks.” Or 100,000, depends on what product you’re selling. “We built X and that’s the price of X. Hey, we’ve just built Y. Y is awesome and it’s a new thing you can do, and it unlocks all these new capabilities. People shouldn’t get that for free, because it’s a new thing that didn’t have. So let’s charge more for Y, but that doesn’t really work with the other… Okay, let’s look at an add-on. Oh yeah, cool. People just add on.” But then, later, now you’ve got people who have the add-on, and people who don’t, and then you’re like, “Add another thing.” And so, we’ve added tiers, with products, tears, add-ons, tearing in the add-on. Oh my god. People can’t understand their bill. So, my advice is keep it simple. Fight so hard to resist the temptation to add extra ways in which you price.
Lenny: Amazing. I didn’t think about going into this topic, but I’m glad that we touched on it.
Paul Adams: Think I was talking about scars for life earlier. That’s another scar for life.
Lenny: All right. Let’s keep talking about some frameworks. Another that I found that I loved is something that you call differentiation versus table stakes. What’s that about?
Paul Adams: It’s like the Kano model, if you’re familiar with that. But, it’s very simple. I guess, we took the Kano model and just tried to make this really crazy simple version of it. Again, I’m a little bit allergic to things like this. I even hate myself for bringing up the Kano model. I’m allergic to people over intellectualizing frameworks. And like, “Oh, well if you’ve seen the new different law…” Of whatever. I’m like, “Keep things simple, practical, and pragmatic. And then, let’s all, again, go back to work and start building the product, so that customers can benefit, because that’s actually all that matters.” And so, difference versus table stakes, very simple. I think people who adopt a product, or buy a product, or switch to a product, there’s two driving forces. One is the attraction of the new solution, and that’s basically differentiation. So what’s different and better? But critically, what’s different and better in ways that customers care about?
Again, back to all the failed projects, my lesson for a lot of these was, we were different and better in these Google projects in ways people didn’t care about. All sorts of Google projects, like Google Wave was an amazingly innovative product that no one really cared about. So, be different and better in ways people care about. So that’s the attraction that’s like, “Oh, I want to check out that. That looks cool. I want to check that out. That looks better than what I have today.” But, on the other side, there’s a entry requirement or table stakes. To play the game, you got to have a certain amount of things. And so, they’re table stake features. They’re often very boring. They’re real basic stuff, boring stuff, and easy to ignore, and easy to not build.
And again, a mistake with Intercom maybe over the years is that we were much more attracted to the differentiation and built a lot of that. So we went through different iterations of our roadmap, sometimes changing over the course of a year or two, where we were all the differentiation to realize that everyone loved it and really wanted to buy, but they couldn’t, because we didn’t have the basic report that they needed or we didn’t have the basic permission feature that they needed. And then, the robot is built based on those… Trading off why do we need more differentiation or trading off why do we need to invest more table stakes? And so, these days, the basic Intercom today is we’re 50/50 probably in terms of resources, but it has swung 70/30 in both directions at times.
The last piece about it is, I think it’s really powerful to look at a roadmap or look at a proposed roadmap and ask yourself, which of these do things matters more to us, not to us actually to our customers right now? The other thing that we’ve talked a lot about here internally is if you’re a startup and you’re entering any established category, customer support for us, big established category, massive, a lot of table stakes, built up over years, decades. ServiceNow, Service Cloud, Salesforce, Zendesk, decades of table stake feature building. So to play the game, you need a lot of the table stakes, unless you have incredible differentiation. So from the early years of Intercom, people just buy us alongside Service Cloud or Zendesk. They just buy us alongside. They’re like, “This Intercom thing…” We were like first modern messaging and modern UX. They were like, “We want that for our customers, alongside the big giant bag of table stakes.” Because Intercom doesn’t have any of those.
Then over the years, we’ve built the table stakes to a point where, okay, now we can fully play the game and people can switch, so they can swap Zendesk for Intercom. But it took us years to get there. And then hence, if you’re a startup, you need to invest a lot more in differentiation. And then, over the years, I think you start to balance the books a bit.
Lenny: I think what’s interesting about this is one, it just gives you a way to think about looking at your roadmap. How much are we actually doing? And are we doing too much table stakes? Are we doing too much differentiation? So it gives you a awareness of what’s happening. And I think, it’s an interesting strategy as a startup like, “Do we spend years doing table stakes and then launch? Or is it go the way Intercom went, like differentiate first we’ll build everything else later?” Wonder when it makes sense to go one or the other.
Paul Adams: Yeah. And it probably depends on the market, different categories, and all sorts of things. Yeah.
Lenny: Yeah. Awesome. Okay. The next framework is something that you call swinging the pendulum. What is that about?
Paul Adams: I actually mentioned an example a bit earlier. Differentiation in table stakes was swinging the pendulum. So, swinging the pendulum means, you take a step back from everyday work life, and you make the observation that something’s in an undesirable state. So, maybe it’s, “Whoa, we’ve all the differentiation in the world, but people can’t adopt the product, because we’ve never built any of these table stakes. It’s undesirable.” Or, “Oh, we’ve now built all these table stakes and we’ve not been investing in differentiation. And actually, we’re not that attractive to people, because switching product is a pain. And we’re not just attractive to people. Okay, so this undesirable state.”
And then, so you go and fix it, but the temptation is that you over-correct. And we’ve done this so many times in so many domains, everything from, “Okay, we don’t have enough differentiation.” A year later, “Oh, wait a minute, we’re missing all the table stakes. Okay, we’re over there.” So, product building is one, people is another one. Building out teams and people. Another big one was, I don’t know, maybe five years into Intercom, we were on this high growth trajectory, really good classic startup before our pricing problems. And, we looked around and said, “None of us have done this before. I don’t think that’s good. Undesirable state. Do we even know what we’re doing? We’re just a bunch of random people. Do we know what we’re doing? We need to hire some experts. We need to hire some experts. If we’re going to go up market, we need market people who’ve done it before.”
So, that was undesirable state, fix it by hiring people who’ve done it before. And then, we hired loads of people who’ve done it before, and what they did was brought the culture and ways of working of their prior company to Intercom. And so, we totally over-corrected, didn’t work out in a lot of cases. In most cases, it didn’t work out. Because, we weren’t trying to be a bigger company, that already exists. We’re trying to be us. So, I think, hiring and building teams is another where we really over-corrected to find out, “Okay, it’s a balance here.”
Related to hiring, one is generalists and specialists, similar theme. People who’ve done it before, or people who are specialized. And, we hired a bunch of specialists only to realize that they’re not adaptable. And, in Intercom, we have a lot of ambiguity, and we lean into the ambiguity, and people who are highly specialized can thrive in big companies, really thrive. They’re invaluable employees. But in a fluid startup-y culture with a lot of ambiguity, they can really drown, really struggle. Maybe the middle of this pendulum, landing in the middle is, “Let’s hire someone who has done a bit of it and have a bit of specialism, not much, but enough to try and figure it out.” So, we hire a lot of those people today.
Lenny: First of all, I love all these stories of things that don’t work out, because a lot of people don’t like sharing these. And, this is what people want to hear, like, “Here’s not everything was perfect. Here’s a lot of mistakes that are made along the way.” And, it feels like this framework is a result of just doing this too many times. Is the main lesson here generally avoid swinging the pendulum too far? Because sometimes, it’s worth it, like in this case of AI, is like, “No, we’re going all in.” Or in mobile, it was worth going all in. I guess, yeah, what do you think of when I say that?
Paul Adams: In talking to people about this before, sometimes the conclusion of the conversation is something like, it’s the only way to do it. You actually can’t do it a different way.” And so, maybe the question is really, how high does the pendulum go? Versus, you got to swing it, and then it’s like, how far do you swing it? And for sure, you’re right. With AI, we are swinging it pretty high. Maybe I overestimated earlier, if AI is in the differentiation camp to mix the frameworks, we’re still building a lot of table stakes features too, building depth into the product. And that’s 50/50, I think I mentioned 50/50 earlier, so that’s 50/50. So, we’re not totally swinging it. It’s swung, but we’re also doing the other thing and balancing things out. So, I think you probably have to swing it. It reminds me to know where the boundary is, is what I was going to say.
It reminds me back to the olden days stories. I remember, at Google, privacy was really top of mind, to the point that it would block decisions, block product progress, just privacy circular conversations, so many circular conversations, and nothing ever got built or shipped. I worked on a project for a year at Google and we shipped nothing in the year, just circular conversations, which killed me at the time. So, when I went to Facebook, I realized they have a different approach to privacy. And again, I’m not advocating it’s necessarily good, it certainly didn’t help their brand. But, there was an idea that to know where the boundary is, you got to across it. And crossing it is painful. But, if you don’t cross it, you’ll never know. So if you think you’re going up to the boundary and you stop before it, turns out it’s actually miles over there.
So I think with a lot of this stuff, you don’t really have a choice. You got to cross the boundary, feel the pain, be humble enough to realize you didn’t get it right, and go again or whatever the corrective course is.
Lenny: Yeah, get that pendulum off the even pivot thing that it’s on. And then, let’s fix that pendulum. Let’s put it back.
Paul Adams: Yeah.
Lenny: Okay. Another framework that I read about briefly, and I love the general idea of it already, which is something that I think you call product market story fit.
Paul Adams: Yeah.
Lenny: What is that?
Paul Adams: So yeah, with product market fit, pretty basic, well understood, very important. The way I describe product market fit is, you’ve got to build the right product for the right market. I think, by the way, as an aside, not enough people think about the market side of that equation. A lot of product people don’t think about the market side. But for me, it’s very simple. The market is the people, the problems they have, and how important the problems are to them. To have a good market, you need a lot of people with the same problem, and they need to care a lot about it. Going back to the Google social stuff, we found a lot of people with the same problem, but they didn’t really care. They didn’t really care. What they had was fine. So a lot of people with the same problem and a lot of energy around the problem and the product is the solution to that. The market’s the who, the product’s the what.
And, I don’t know, in my career again, so a bunch of products that were built, there were good products in good markets, and they failed and I couldn’t work it out. And eventually, I came back to this idea that… And maybe someone might say, “Paul, it’s marketing. You’re talking about marketing.” But story, the story’s wrong or the story’s missing. And so, sometimes, it would be a great product in a great market explained in a convoluted way. I see that a lot. I used to see that a lot at Google again, just explained in a very complicated way over intellectualized. And, as a result, people are like, “What? What are you talking about?” You don’t get their attention. And so, the story is really important, as important. And actually, sometimes you’ll see not great products, certainly worse on paper… I’m trying to remember the Spotify competitor back in the day, people were like… What was the name of it?
Lenny: Ordio?
Paul Adams: Yeah, Ordio. Ordio was one of these where-
Lenny: I like Ordio a lot.
Paul Adams: … Yeah, all I’ve ever heard about Ordio was, “Amazing product.”
Lenny: Mm-hmm.
Paul Adams: It’s failed. And why did it fail? Spotify and Ordio had the same market. They were solving the same set of problems. Ordio was arguably the better product at the time. I don’t know if that’s true, but arguably the better. I also think Spotify’s an incredible product. But, they got the story wrong. And so, again, I think, all product people, whether you’re a designer, product manager, people in research, data science, need to think about the story all the time. Work of marketing, work of product marketing, and learn about how to explain the product, as much as how to build the product.
Lenny: Mm-hmm. Makes me think about positioning and how important that is. And, we had April Dunford on the podcast very recently talking a lot about that.
Paul Adams: Yeah. Yeah, she’s excellent. Yeah, it is really, “Why are you better and can you explain why you’re better?”
Lenny: That’s such an important point. A final area I wanted to touch on is jobs to be done. So we had the co-creator of Jobs to be Done on the podcast. We had Shyam Krishnan on the podcast. They very much disagree about how effective Jobs to be Done is. I know you guys are big on Jobs to be Done. So, what are your just general thoughts on the Jobs to be Done framework? How effective was it for you all? How do you use it? What do you find work? Doesn’t work? Whatever comes up.
Paul Adams: Yeah. I’ll be totally honest, at the risk of finding people do this, we worked with Bob West years ago. I think Bob’s a great guy. And we followed that model of Jobs to be Done more than the ODI, I think, is the other skill of thought. Anyway. I’ll try say this in a simple way. We found Jobs to be Done really good. Very, very useful. But, in a very simple way… Again, back to this idea of simple frameworks, in a simple way, separately, there’s so many people who spend so much of their energy debating the nuances and peculiarities of one version. Who cares? No one cares. Oh well, I don’t care. They care obviously. But your customers don’t care. People you’re trying to build a product for don’t care,. No one cares. That’s a cool intellectual debate. But, for me, maybe this is too extreme. It doesn’t really have any place in the work we do. We’re just trying to build a great product.
And so, for us with Jobs to be Done, it was a really good way of us centering on the customer problem, focusing on not getting distracted, basing it in good solid research informed insight, that told us the thing people are trying to do. What is the thing people are trying to do? Again, energy. Do they have a lot of energy around it? Maybe the energy thing might’ve come from talking to Bob actually, now that I think about it. I think it did actually. I think, the idea of this idea that you need people who have a lot of energy around the problem. And you have to interview them for that most of the time to feel the energy they have. It’s very easy to see if someone’s apathetic versus into it.
So, we’ve had it pretty good. And, we invented this job stories thing by accident. I can’t remember exactly what happened. But, I wrote out this way of writing a job story basically. Well, we didn’t call it job stories, someone else called it that. We just, at the time, were like… I can’t even remember. It was a trigger. And, anyway, we didn’t even give it the thing a name, someone else named it, I think. And, I’m just like, “We’re just trying to build a great product.” So, we’ve had it really good in that way, really simple. And then, the other one that we use a lot still here is the four forces, which is this framework of Jobs to be Done. The four forces being… There’s different forces when people try and switch product. And some of it’s the differentiation, table stake stuff, like the attraction of the new solution, the reasons that you might not adopt it. Habits. People have anxieties.
Here’s another funny story to tell you how much… The four forces is really good. Here’s a funny story, I was saying earlier that Eoghan and Des were trying to convince me to leave Facebook, which I loved at the time, join and to come. They wrote out the four forces for me to join. And then, secretly, over a few beers, talked to me and fed me my anxieties. And basically worked me on the four forces. And I was like, “That is genius. That is ingenious. Maybe it’s a bit… But it’s ingenious.” And so, the four forces is incredibly good at helping understand why people make decisions.
Lenny: I love that a lot of your advice just continues to come back to, keep it simple, cut away anything that isn’t necessary. And, I find the same exact thing with Jobs to be Done. I find it really useful as a framework for the podcast, the newsletter, but I think there’s this endless set of processes and ways of optimizing that gets people distracted. And, often just slows everything down.
Paul Adams: Yeah, yeah. And it’s interesting and fun to talk about sometimes, really fascinating, unless you’re an academic. But if you’re working in a company that you’re trying to build a software product for people to improve their lives in some small meaningful way, it doesn’t matter. Just use the thing that helps you do that. That’s the goal. And use the thing that helps you do that. And that’s it.
Lenny: With that, we’ve reached our very exciting lightning round. Are you ready?
Paul Adams: I’m ready, yeah.
Lenny: What are two or three books that you’ve recommended most to other people?
Paul Adams: Yeah, the two books I recommend to everyone always, I have copies in my office here, It’s Not How Good You Are, It’s How Good You Want to Be. It’s a book by Paul Arden who worked in advertising a long time ago. It’s an excellent book. It shows people that you feel an unlimited potential if you think about it the right way, everyone does. The second book I recommend to everyone and buy for people and give to them is Principles by Ray Dalio. I’m a big fan of Ray Dalio. I think he’s incredible. I’m a big believer in principles. A lot of us at Intercom are… I always get those two books. And they’re totally different. The Paul Arden book, you can read it in 20 minutes. Principles is that thick.
Lenny: What is a favorite recent movie or TV show that you really enjoyed?
Paul Adams: Most recent is The Bear, which I came to late. The reason I love the show is because I think it somewhat celebrates the grind. And I think that’s important. I worked in coffee shops a lot when I was younger, when I put myself through college and stuff. And, the grind is part of life, and the grind is a necessity to get things done, and make great things happen sometimes. And I like that about it. I really like that about it.
Lenny: What is a favorite interview question you’d like to ask candidates?
Paul Adams: Yeah, I’ll give you a slightly different answer. I don’t really have certain few questions for candidates. And I don’t like answer question diversity. I don’t like questions that rely on memory. Like, “Tell me about the last time you did X.” Here’s an amazing question I got given recently by Alyssa who used to work here. I had to do referral calls. So, you’re interviewing someone, you want to give them the job and they’ve got referees, and of course, the referees they have are the best people that they’ve ever worked with and their favorite managers. So this question is, “What feedback will I be giving this person in their first performance review?” It’s an amazing question, because the person can’t dodge it. There’s an answer. And, it’s incredibly enlightening.
Lenny: And that’s a question you ask on reference calls?
Paul Adams: Yeah, on reference calls.
Lenny: That is such a good question. I love it.
Paul Adams: Yeah, it’s a amazing question. Yeah.
Lenny: All right, what a gem. Thank you for sharing that. What is a favorite product you’ve recently discovered that you really love?
Paul Adams: This is maybe cheating, but I go back to a lot of the AI products. I think ChatGPT Vision is mind-blowing. I’ve been playing with Rewind lately. I was a bit late to it. Des, and Kiran, and a bunch of people here, founders of Intercom, love Rewind, use it and love it. Thing’s amazing. So I’m a bit late to that. But, it’s just augmented memory. It’s mind-blowing. So, Rewind’s been fun.
Lenny: And they just came out with a little audio thing that can record your actual day.
Paul Adams: Yeah, I’m not so sure about that.
Lenny: Yeah, got some flack.
Paul Adams: Yeah.
Lenny: I’m not so sure. I don’t know. I don’t know if it’s real. It looked like not a real product when they launched in, but I think it’s real.
Paul Adams: And it tippy-toes into what’s okay and not okay with AI. And, yeah. Yeah, it’s a cool theory though, for sure.
Lenny: What is a favorite life motto that you often come back to share with people, find helpful for yourself?
Paul Adams: Yeah, I have a post-it on my monitor that says, “Only work on what matters most.” It’s on my monitor, a post-it. And it sometimes falls off, and I have to write it again. Only work on what matters most. And, it’s amazing. I go into work, someone emails me, and I’m like, “Oh, God.” I’m like, “Only work on what matters most.” The second one related is, stop worrying about things you can’t control. And so, I have two of those. And so, only working what matters most. Stop worrying about things you can’t control. It just reduces the temperature. Again, life lessons learned. I sent a lot of dumb emails in my past, like, “Red Energy, oh my God, what are they thinking?” You wake up in Dublin to a San Francisco email. And you’re like, “Oh god. Keyboard.” And, if your monitor says these two things, you just don’t do that. You just take a breath, get a coffee, come back. Does it really matter?
Lenny: Beautiful. The second one, I think, I learned first from Seven Habits of Highly Effective People. Have you read that?
Paul Adams: Oh, yeah.
Lenny: Just think about the focus, the circle of things you can control, and then there’s the circle of things you can influence, and then there’s the things you have no control over. And, I find that really helpful myself. I love that you have it as a post-its. I feel like, I need to make post-its of all these lessons people share as their little mottos.
Paul Adams: Yeah, the post-it on the monitor is a real life hack, I found a few years ago. Because it’s dumb in a way. The posts on the monitor, it’s in the way.
Lenny: Wait, you actually put it on the monitor in the way of your screen?
Paul Adams: Yeah, yeah.
Lenny: Oh, wow.
Paul Adams: It’s in the bottom left, just covering the bottom. Because otherwise, if it wasn’t there, I wouldn’t look at. I make myself look at it.
Lenny: Yeah. Wow. I haven’t heard of people putting it over precious real estate on their monitor.
Paul Adams: Yeah.
Lenny: That works. What’s the most valuable lesson your mom or your dad taught you?
Paul Adams: The biggest one, again, so reductive and simple is to be nice to people. I think, being nice goes way further than people really realize. One thing that I’ve learned, again, the hard way through life is you have no idea what’s going on in people’s lives. You’ve no idea. People could have all sorts of really stressful, all sorts of personal stuff going on, and the reason they did the thing at work that you didn’t like is because of that. And so, I try and think, “Be nice. You don’t know what’s going on. You might learn later. Don’t act in a way you would regret.” I think, being nice in life goes far further than most people give a credit for, because it’s too much of a, I don’t know, fluffy truism or whatever.
Lenny: I 1000% resonate with that. I’ve been told I’m too nice and I had to become a little less nice. But, I still can’t lose that. So I fully buy into that. My parents taught me a similar lesson.
Paul Adams: Yeah. And sometimes it’s hard. I’d never fired anyone before I joined Intercom, for example. I really did not like doing it. And, since then, I’ve done it quite a few times in a bunch of different circumstances, and realized it always works out for both sides. And the nicest thing to do is to do the harder thing. It’s actually the nicer thing to do. People are relieved in this example. It’s a nicer thing to do. So, it can be a complicated one.
Lenny: I love it. Final question. You’re Irish, you’re based in Ireland. What is an Irish food you think people should definitely try out if they ever visit Ireland?
Paul Adams: Can I cheat and say Guinness? Is that food?
Lenny: Absolutely.
Paul Adams: The Guinness in Ireland. People talk about this and it’s true. The Guinness in Ireland is much, much better for a whole bunch of reasons. It’s basically a fresh product and it’s brewed here. It’s the way they think about, it’s like milk. Milk goes off, Guinness goes off. Guinness is older than a few days old, tends to start deteriorating. So, Guinness Ireland is amazing, because it’s made here. The other thing I think that Ireland does really well is fish. Ireland has not had, by the way, the greatest reputation for culinary excellence over the years. I think Irish food in the States in particular is not good. But, the fish here is incredible. You can get incredible fish. And Ireland’s obviously an island, so there’s a lot of fish.
Lenny: On the Guinness front, is there any way to get the good stuff not in Ireland? Or is that just you got to go?
Paul Adams: No, there is actually. You just need to be near a brewery. So Guinness is brewed in Nigeria. There’s a huge Guinness market in Nigeria.
Lenny: I did not know that.
Paul Adams: I think they actually use a different recipe, but it’s brewed there. I think the brewery in the U.S. is somewhere in the east coast between New York and Eastern Canada. So, it’s somewhere there. So, often, the Guinness in New York can be actually pretty good. The Guinness in San Francisco tends to be really bad. I remember talking to someone about this that works in Guinness. One of my friends, does a lot of work in Guinness. I think the boat carried the Guinness goes down through the Panama Canal back up to San Francisco. So, it’s 12-weeks-old or something.
Lenny: Wow. Did not think we would be learning about the travel path of Guinness from-
Paul Adams: At least this is what I’ve heard. The Guinness has so many myths, you just don’t really know what’s true. But, these are the stories I’ve been told.
Lenny: … Amazing. Paul, you are awesome. Thank you so much for being here. Two final questions. Where can folks find you online if they want to reach out? And how can listeners be useful to you?
Paul Adams: I have a handle, it’s everywhere. Basically, P-A-D-D-A-Y. It’s Paddy with an extra A. So, P-A-D-D-A-Y. That’s everywhere. So, paddy@gmail, @Paddy. It’s my handle everywhere. So, that’s where you can find me. I’d love people to reach out to me, right, genuinely learn. I’d love to hear from people who think my AI talk is nonsense and it’s more a crypto Web3. Or, I’d love to hear people who have alternative opinions and challenge mine. That’s how I like to learn and get better. So, if people have those opinions, I’d love to hear them. I’d love to talk to them.
Lenny: Be careful what you wish for. The YouTube comments are always a spicy place. We’ll see what we see. Awesome, Paul. Thank you again so much for being here.
Paul Adams: Yeah, thanks Lenny. I really appreciate it.
Lenny: Bye, everyone. Thank you so much for listening. If you found this valuable, you can subscribe to the show on Apple Podcasts, Spotify, or your favorite podcast app. Also, please consider giving us a rating or leaving a review as that really helps other listeners find the podcast. You can find all past episodes or learn more about the show at lennyspodcast.com. See you in the next episode.
Glossary
| English | 中文 |
|---|---|
| add-on | 附加组件(add-on) |
| Alyssa | Alyssa(Intercom 前员工) |
| April Dunford | April Dunford(定位领域专家) |
| Bard | Bard(Google 的 AI 搜索引擎) |
| Bob West | Bob West(Jobs to be Done 领域的实践者) |
| ChatGPT Vision | ChatGPT Vision |
| copilot | copilot(副驾驶式辅助工具) |
| Des | Des(Intercom 联合创始人) |
| Entropik | Entropik(文中提及的 LLM 公司,疑为 Anthropic 的转录误差) |
| Eoghan | Eoghan(Intercom 联合创始人兼 CEO) |
| Fergal | Fergal(Intercom 前机器学习负责人) |
| Fin | Fin(Intercom 的 AI 聊天机器人产品) |
| first principles | 第一性原理 |
| four forces | 四力模型(Jobs to be Done 框架下用于分析用户产品切换行为的子框架) |
| Gartner Hype Cycle(文中提到的”炒作高峰、幻灭低谷”曲线) | 加特纳技术成熟度曲线 |
| Google Wave | Google Wave(Google 推出的实时协作平台,已停服) |
| Guinness | 健力士(爱尔兰黑啤酒品牌) |
| Horizon framework | Horizon 框架(亚马逊等公司用于规划创新投资的三层战略框架) |
| job stories | job stories(Intercom 团队发明的以任务为导向的用户故事写法) |
| Jobs to be Done | Jobs to be Done(以用户任务为导向的产品创新框架) |
| Kano model | Kano 模型(产品特性与用户满意度关系模型) |
| Kiran | Kiran(Intercom 联合创始人) |
| Lenny | Lenny(播客主持人) |
| Lex | Lex(Lex Fridman,播客主持人) |
| Matt Rickard | Matt Rickard(AI 领域博主/Newsletter 作者) |
| Metaverse | Metaverse(元宇宙) |
| mock-up | mock-up(模型/设计稿) |
| ODI | ODI(Outcome-Driven Innovation,成果驱动创新,Jobs to be Done 的一个分支流派) |
| onboarding | 新用户引导(onboarding) |
| Ordio | Ordio(Spotify 的早期竞争对手) |
| Paul Arden | Paul Arden |
| product market story fit | 产品-市场-故事契合 |
| product-market fit | product-market fit(产品-市场契合) |
| Ray Dalio | Ray Dalio |
| Red Energy | Red Energy(文中提到的某产品/公司名称) |
| Rewind | Rewind(AI 记忆增强工具,rewind.ai) |
| Seven Habits of Highly Effective People | 《高效能人士的七个习惯》 |
| Shyam Krishnan | Shyam Krishnan |
| table stakes | 基本门槛(table stakes) |
| Tableau | Tableau(数据可视化/报表软件) |
| The Bear | 《熊家餐馆》(FX/Disney+ 剧集) |
| tiers | 层级(产品定价层级) |
| Web3 | Web3 |
| Zuck | Zuck(Mark Zuckerberg 的昵称) |
| 《碟中谍》/ Mission Impossible | 《碟中谍》(电影系列) |
| 《黑镜》/ Black Mirror | 《黑镜》(Netflix 剧集) |
| 萨姆·奥特曼 | Sam Altman |
Reformatted by reformat_english.py