SaaStr 839: Why Most SaaS Companies Will Fail at AI (And How to Avoid It) with Intercom's CPO

0:00Welcome to the official Sastra podcast where you can hear some of the best Sastra speakers. This is where the cloud meets. Up today on the Saster Podcast. There's loads of processes at Intercom around building software that I designed as a part of designing. I'm proud of them and love them. I've given talks about them. They don't exist anymore. You have to delete these things. They're not a part of the future. You need to ask yourself why you exist in a post-AI world. Most SaaS companies sold seats. These are seats that humans sit in. Those humans use some kind of GUI, some CRUD type tasks to achieve some outcome.

0:38None of these things make sense anymore. None of these things make sense. There's no seats in a post-AI world, or at least the way seats are orchestrated is very, very different. People don't use GUIs necessarily. A lot of the product is invisible and a lot of the quality of the product is invisible. And ultimately, the company with the best outcome, built on the best AI, built on the best RAG system, built using custom models, they're the ones that are going to win. And you got to do that really fast.

1:09Hey, Sasser, imagine having agents for every support task. One that triages tickets, another that catches duplicates, one that spots churn risk. That'd be pretty amazing, right? HappyFox just made it real with Autopilot. These pre-built AI agents deploy in about 60 seconds and run for as low as two cents per successful action. All of it sits inside the HappyFox Omnichannel AI First, support staff chatbot copilot and autopilot working as one check them out at happyfox.com slash saster hey everybody saster annual will be back may 2026 the world's largest sass and ai gathering for executives just this last may we hosted 10 000 attendees with 68 vp level and above attendees 36 ceos and founders and 25 were ai first professionals it's the very best of s to your attendees and decision makers that come to Saster Annual and AI Summit each and every year.

2:05But here's the reality, folks. The longer you wait, the higher ticket prices get. They're cheap now. They're cheap, so just get them. Early lock in your spot today. Use my code Jason100 for exclusive savings. Get your tickets at podcast.sasterannual.com or just use code Jason100 when you check out. See you there. Saster Annual and AI Summit 2026. It will rock. I got to talk to you, man. The transformation of our company, Chad Gbt just had a birthday three years ago. If it looks at Intercom three years ago versus Intercom today, they are entirely different companies. It's unrecognizable. And this is really important because a lot of SaaS companies, I think, are saying that they're AI companies, but they're not really AI companies.

2:52They're not true, deep AI companies. We've gone through this transformation. I'll show you what it's been like. I'll show you the lessons we've learned. It's been brutal. like brutal. It's really hard. It's really hard. And if you're a SaaS company who thinks you're an AI company and you've not gone through brutal transformation, you're not there yet. And so I'm going to show you what we did and hopefully it'll help you kind of think about things. So Intercom was founded in 2011. I've been there since the very start. It's been a long journey, a fun and an energizing one. It's been a bit of a roller coaster.

3:24We were a breakout SaaS company. We went from 1 million to 50 million in ARR in three years, which back then was really impressive. Now it's not so much. Things have changed. We really, really had these great years and they ran for quite some time. The company was going really well. We'd built a great product. We had really strong product market fit. We've got some customers who love the product, who would tell other people about the product. But things changed. We then had five quarters of declining revenue growth so revenue still growing but the rate at which was growing was declining and this was not a good period for the company we had gone down the ipo journey we i think it changed how we think changed how we work if anyone's done the ipo journey it changes the company we were in a kind of a bad spot and owen our co-founder and ceo came back into ceo role and then track gpt showed up Three months later, so Owen had just come back.

4:24We were in this kind of tricky spot, and ChatGPT showed up. We had a machine learning team at Intercom since kind of 2015, 2016. And a combination of the machine learning team and X telling us this is the moment. This is a one-way door for humanity. This is the real deal, the iPhone moment, the next technology cycle. We had very knowledgeable people telling us that. We also had personal experience doing it. A lot of us had started our careers in the early internet. We kind of knew what disruption felt like and looked like. A bunch of us had worked in different scenarios. I'd worked in the mobile team at Google when the iPhone came out, which is an interesting place to be.

5:01So you could see it. We were pattern matching and we could see that this was a really big deal. So we bet the entire company on AI. We bet the entire company. It was a risky thing to do. We ripped up our strategy. We ripped up our roadmap. We made these decisions in a very small number of days, like you're talking once, two weeks after ChatGPT. we decided we were in this tricky spot that we had nothing to lose or felt like we had nothing to lose. We had like hundreds of millions of ARR to lose. We felt like we had to do it. So we built Finn and launched it in March 2023. So we went very, very quickly from betting the company on AI to launching Finn.

5:40And Finn was an AI agent for customer service and people really liked it and found a lot of value in it and it worked. If you fast forward to today, Finn has over a million resolutions a week. So every single week, Finn is resolving over 1 million customer problems. We have over 6 ,000 Finn customers, and Finn's average resolution rate is over 65%. So the resolution rate over those entire 6 ,000 plus customers is 65%. So in other words, Finn resolves 65 % of the queries that it's asked about. Amazing. And Intercom is on this entirely different journey now. We have a very bright future, finish this groundbreaking product, transformational product.

6:21But the journey from Owen coming back and Chachi PT three years ago to today was really hard. It was really hard. And we learned a ton of things along the way. So I'm going to talk about a bunch of the biggest things for me. Building a culture of change and having no sacred cows. I'm going to talk about why does your business exist in a post-AI world? How do you build software? how we build software at Intercom is completely different to how we built software three years. And actually building software, I think, was our strongest thing. Some companies are marketing-led, sales-led, product-led.

6:57Different companies have different strengths. For sure, building great products that customers loved based on deep customer insight was our strength. We built processes around it, and that's all gone. We've completely changed how we build software. And if you're successful at these things, you end up with kind of two companies. You end up with intercom and you end up with fit and they look very different it's going to talk about what we're doing with that then lastly i'll finish up by talking to you about the biggest mistakes i think you'll make because it's certainly mistakes that we made at times and had to try and avoid other times okay so let's talk about culture this kind of sounds like ridiculously obvious to survive ai eat a culture don't resist ai and no one says that they're like gonna resist ai it's not a thing however look at the list of things that you have to change you have to change everything you have to re-found your company if you have a SaaS company and it predates AI and it's not an AI native company in other words you started the company designing an organization for the AI world you have to change everything your org your product your roadmap how you build the metrics how you measure success, how your sales team works, how you go to market, pricing, how you price, what does successful pricing look like, what does successful revenue growth look like, seats, outcomes, those two things together.

8:17You have to change everything. Everything. And it's really hard to do it. And culture is deep. I'm going to talk about it in a second. It's deep. It's really hard to get people to change things that they do. Humans hate change. humans we all hate change we're not fans of change so we're averse to it and it shows up in loads of different ways if you're a deep ai company you your product looks like this if you're truly a deep ai company of course you have an ai layer but you also have an app layer and a model layer so this ai layer this diagram is our rag system and this is a gross simplification of the rag system We believe we have one of the best RAG systems in the world.

9:01It's been three years in the making for a very talented team, and it's complicated. I do not understand the depths of that RAG system at all. It's complicated, but it's necessarily complicated. We also build an app on top of that. That is how do people access the RAG system, access your AI, and you need a model there. We have started building custom models, training custom models. I'll kind of show you how they work in a second. But if you're a true AI company, you're doing all three of these things. You're building an AI layer, and it's complicated. You're building a model layer. You're looking at how you train your own models, building custom models, looking at different parts of the kind of space that you solve for and building models, target bits and pieces of it.

9:43May or may not work. You're experimenting a lot, and even app layer on top of that. So to get to this place, you need to change the culture. And the problem was that no one's going to say no, but they'll say all these other things. They'll say, not now. They'll say, well, you know, we should start that next quarter. We've a lot going on. Our customers aren't ready for that. Our customers don't really, the AI thing, they're not sure about it. Last week, I updated the board. I gave them these new forecasts. If we do the AI thing, it's risky. Maybe we don't meet our number. They're going to give me pressure.

10:16We'll look bad. It feels risky. Do you know what we'll do? We'll dip our toes. We'll do a bit over here and a bit over here and a project or two. and we'll prove it out and see if it works. And then we'll change. Then we'll transform. So step one on this journey is you have to tell the company that it's happening and it's not a choice. You're changing the company, you're refounding the company, you're changing everything about the company and it's not a choice. And the reason is because AI is inevitable. To say like Like AI is the next technology cycle. Isn't even a new thing to say. That happened ages ago.

10:55We're like three years in now. AI is inevitable. Whatever you loved about SaaS and the last decade of mobile and social and the internet before that, it's gone. It's gone already. So you have to change. You've no choice. And you can do, you can kind of like, I feel like everyone, most SaaS companies at least, SaaS companies at scale are at this crossroads. They have two paths to go down. opportunity or whatever the opposite of that is death uh if you nail it if you truly change you'll become a category leader like finn is you will grow to like 100 million in air or very quickly um in finn's on that path very close to it you'll be on the front page of forbes if you care about that you'll have all the series you'll raise all the money that you want no problem But if you don't do it, and if you fail, and deciding not to do it, I think is failure, you'll slide into irrelevance, slowly but surely.

11:50New competitors will eat your lunch. You'll have down rounds. All your best people will quit because they'll go work at the true AI companies, and ultimately the company will die. And this is a really important, we're at this crossroads, a lot of companies, and it's really important that people pick the right. the AI revolution cannot be negotiated it's not something that you negotiate with people and you negotiate with staff you negotiate with your peers remember all those people that will not say no but they'll say to other things and here's the mistake people make when they go into this journey they say they make the easy fun choices, building AI features and fun things like that trying different models talking to customers about it that's all the fun stuff but they'll avoid all the hard messy ones like a third of your company you need to part ways with them they're not fit for this new world they don't want to build the same way you want to build they don't like fast crazy fast pace they don't like chaos it's going to feel chaotic um they're not meeting the moment um they're just going to you know slow things down get in the way that's the hard part that's the real hard part of this um and then you'll convince yourself you've done enough you'll do the fun stuff you won't do the hard stuff but you'll convince yourself that you've done enough and i see this all the time a lot of people it's just because it's hard and we don't like change so what we did at intercom was we went way too far because the only know the only way to know if you've gone far enough is to go too far the only way to know where there's a boundary is to cross the boundary so for example on paper i'm our chief product officer i've always run the product team at Intercom.

13:31Six months ago, I took over two thirds of marketing. So I'm now actually running marketing much more than I'm running product, at least for the last six months. So I'm running brand, product marketing, content marketing, advocacy. Why is a product leader running? Marketing is an entirely different 30 minute talk. The first thing I did is I blew the entire thing up. Everything's gone. Teams, roadmaps, marketing calendar, delete the marketing calendar. people loved Coda there was spreadsheets planning about plans like loads of stuff the only way I knew how to build a marketing org that was fit for purpose for this age is to build it from scratch blow it all up and the right things will grow back again so we have a calendar now again but it's way simpler than the calendar we had before and if I tried to iterate from the place we were it wouldn't be the same it just wouldn't be the same so you have to go too far and people hate this people don't like this and you have to kind of upset people and explain why you're doing it and bring them on the journey and for those of those people who just aren't up for that you have to part ways with them smile shake hands it's amicable it's all good they'll all go on to find different things to do that they're more happy with but you have to do that I'm always asking myself this question what would a brand new startup incorporated today do here I'm like literally last week we're doing some work design stuff and I'm asking myself what would a new company do would they have product marketers and would they have content marketers or is that actually kind of the same job a lot of the time maybe we should have a full stack marketer right so i'm constantly asking these questions what do product managers do what do product designers do all right would a new company have both of those roles too so product designer product manager product marketer content marketer all debt like chain together i don't think they do that so what would they do so ask this question all the time and i mentioned that culture runs really deep culture runs really deep so you have to be wary of the people you need to keep convincing because they're going to be a liability.

15:27You have to be wary of those people just to not everyone's up for this. It's a fight to transform from a SaaS company to an AI company to see the success we've had with Finn. By the way, it gets better. The hard part gets easier and then it gets exciting. Today with Finn, we're in a really good place, but to get to that place was really hard and was tricky. You have to delete all the parts that are process that are not part of the future. And there's a ton of these. I'm going to show you some in a second. There's loads of processes at Intercom around building software that I designed and I was a part of designing.

16:01Very proud of them and loved them. I've given talks about them. They don't exist anymore. You have to delete these things. They're not a part of the future. You need to ask yourself why you exist in a post-AI world. Most SaaS companies sold seats. These are seats that humans sit in. Those humans use some kind of GUI, some crud type tasks to achieve some outcome none of these things make sense anymore right none of these things make sense there's no seats in a post ai world or at least the way seats are orchestrated is very very different um people don't use guis necessarily a lot of the product is invisible and a lot of the quality of the product is invisible and ultimately the company with the best outcome built on the best ai built on the best rag system built using custom models they're the ones that are going to win.

16:49And you've got to do that really fast. You've got to build this world really fast. That's how we ripped it up. Took the pain because you have to do it fast. We could see new competitors entering the market. We were worried and we were paranoid about it. And we knew we had to do it really quickly. One thing you've got to do early is understand how ambitious you're going to be. So this is really kind of simple. Value goes up, but ambition and risk go up too. So you can decide that what you're going to do with your SaaS company is change it a bit. We're going to help humans do a single task in their job.

17:22That's pretty straightforward to do. Kind of low risk, but kind of low ambition. The value is like, you know, reasonably good, not bad. Or you might decide we're going to tackle a whole workflow. That's way harder. Now you've got all the steps in the workflow, but the ambition is greater. Maybe you can charge more for it. Maybe the company will be more successful in the long run. Or you can decide we're going to do the whole thing. We're going to be extremely ambitious. We're going to disrupt our entire business that we run today. We're going to do the work of the entire team. Today, we have all these people who pay for us with seats to use our SaaS product.

17:57And we're going to disrupt the entire thing, risk all our revenue, or at least risk the revenue goat by trying to change it into a new AI outcomes first world. So you're going to decide where you're going to be on the spectrum. And to do that, you really have to understand the job you're solving and deeply, deeply understand it. So for example, these are like two examples of kind of classic SaaS products. We've like a meetings and project management. We've an expense management product. These again are like GUIs that, you know, prod tasks. People are like reading things and writing back to systems.

18:32And it's very workflow focused, like a lot of SaaS products, stuff like this. But if you want to solve this problem in an AI-first way really well, you have to break down every single bit of these and understand them really deeply because what you're going to end up doing is pointing AI at all of these discrete things. Because getting AI and getting a RAG system to accurately do these things is easier when it's discrete. You can experiment much faster. You can solve bits of the problem independently and then kind of chain them together. So you need to understand each of these steps and you need to understand that you can reliably solve it.

19:10and the reason this matters is because all of these things compound if you're really really good at each of these steps really good like here we're really good very reliable you still have to multiply all these together and by the time you're finished the entire workflow especially workflow of an entire team you're down into like the 90 success rate and if you're not as good at you're down at the 80 success rate it's just way harder to build a product unless you really understand it. So that's what we did with Finn. This is again like an illustration of some of the system, some of our AI system with Finn, where you can kind of see we have these like custom models that pointed directly at steps along the journey, pointed directly at very specific things, customer service and customer experience team.

19:54And that's how we're able to get such incredible results with Finn. This is like a lot of experimentation. Days, weeks and months of experiments, constantly trying to find a new edge, constantly trying to find the like tiny little increases because the reliability compounds. Each single tiny incremental improvement in each of these steps adds up to something way better, adds up to the highest performing product, adds up to something people can trust, adds up to something that people will replace their humans in places with fit. But they'll only do that if they can trust it. They'll only do that if it's like credibly the highest performing.

20:27So you've got to attack these things very systematically. each step along the journey. And to do that, you really have to understand them. Another thing you ought to think about is marketing the product. There's a thing that we call the marketing overhang. And this is like very, very, very common. I think one of the hardest, I think marketing AI products right now is extremely difficult. Evaluating isn't difficult, but marketing is really, really hard. Marketing fin is really hard because a lot of companies claim that they have a thing too. And their thing works just as well as our thing. And they have demos.

20:59And so you can kind of see here, like they're building things and they work and it kind of works a bit and they can demo it. But a demo isn't a product. A demo isn't a product that works at scale. And so what happens is people market, go to market with this amazing product idea that they've real demos for. They do product launches of the demos. We have a rule at Intercom that we will not do a product launch that is not real. We're not going to do a product launch that isn't a real demonstration of the product, and we know that it's going to work at scale, which kind of inhibits us a bit and it slows down marketing sometimes.

21:35The reason is the product might not work. And when you build AI products, you really, really don't know if they're going to work. You need to have them in production, experimentation all the time, trying to find edges, trying to find improvements in performance. So you get this problem where you've promised the world, maybe you've already signed up customers, You've promised the world this incredible product and you've got one really good demo, but you've got a huge delta between what's actually possible in reality when it really goes live and the promise you delivered. This is a great example of this.

22:11Apple Intelligence. Amazing in June 2024. Apple are finally here. They finally entered the AI race. They're going to have this incredible product and maybe they're going to win. They've got your device. like they've got the OS layer, they've got loads of advantages here, customer data, all sorts of stuff. It's now next spring, 2026, where we'll see Apple intelligence finally. And then we'll see, right, this is the marketing overhang. If you're really careful that what you're building, demoing is real, and it really works in production at scale. Okay, the next thing is that how you build software is very different.

22:47So your market is different and how you build it's very different. Today's software is built with empirical evaluation. So running these evals, running, having a robust evaluation framework, and tweaking the architecture, using these experiments constantly to find these little edges in performance, comparing the old, comparing the new. We've run hundreds and hundreds and hundreds of experiments with Fin. And obviously not all of them work. I don't know if most fail, but certainly many of them fail. or we see an improvement in one part of the system, but back to that kind of line of the workflows, another part of the system isn't as, we've degraded performance there.

23:26It's going to have to think about this thing systematically. And then the majority of product improvements are invisible and unpredictable. So in the SaaS world, this is what building great SaaS products look like. Most of it you can see, you're building like really great user experiences, really great UI to help people do those jobs well. And the CRUD backend, and, you know, tiny, kind of invisible, not a big deal. There's no kind of technology problem here. You know you can do it. If you have, like, dedication, resources, you know you can build the thing you want to build. And actually the hardest parts are in the user experience, making it easy to use, easy to learn, all that kind of stuff.

24:01AI products don't look like this at all. It's actually the inverse. Building the UI is the easy part. That's the small part. The hard part is building great AI infrastructure, building products that actually truly work at scale. This is an entirely different world. And my background's in design, originally I was a designer, and the left-hand side here was like my happy place. You know, this is an awesome world. And I had a reckoning, a personal reckoning, weeks of anxiety. What happens in the new world? You know, the thing I'm good at is like the top part, and that's really small now. And things like infrastructure, I do not.

24:40I was like, I'm way out of my wheelhouse if we get into like infrastructural layers. But I had to learn it. I had to get better at it. I had to kind of face reality and say, okay, well, I've got to get better at this kind of stuff. We had these principles at Intercom that evolved over years, and they made us really good at building SaaS product. They made us really good at it. We had like kind of six core principles. We had like design principles. And what we would do is we would teach new people these principles. So a designer joins the company, an engineer joins the company. we'd say this is how we build software intercom.

25:11This is why we're good at it. Because we follow these principles. We have a little process for each principle. Like think big, start small, ship early. And there's like ways in which you would do that systematically. And we got really good at it. But these principles were designed for the SaaS era. They were designed to build a SaaS product. We had to abandon all this shit. It was, it pained me. It still pains me fucking at this. It pained me to do it. You know, Des, the product leaders, the engineering leaders, it was really, really hard. People did not like it. People like me didn't like it.

25:51I love those principles. You know, I helped make them. They're part of, I like looking at them on the wall. It doesn't matter. You know, looking at principles on the wall isn't going to build a great business. And people were disempowered. They're like, this is the way I know how to build software. I'm good at building software this way. I don't know how to build software the new way, and it sounds way too technical and way too hard for me. It's not how I think about things. So there's an old way and a new way to build software. The old way, you pick a job to be done, you listen to customers. If you're very good at it, you get really good at customer insight, getting true customer problems down into the depths.

26:24You design a solution, you build, and you ship. And in this world, execution is certain. The technology stock is pretty stable. you know design is one of the hardest parts of the process not the engineering side users could talk about their problems they were very easy to talk about like all the ways in which their workflows didn't didn't work and their company and their org design it was a stable world technology was stable but organizations were stable how organizations were structured how they think all stable there was never a question of like can this thing actually work is this even impossible.

27:00There's an entirely new way to build software. And this is what we've changed. Again, it's painful to go through this because you've got to delete the old way and it's risky. It feels very risky. There's an entirely new way to build software. You start by asking, what does AI even make possible? Like AI is a magical technology. I think people are like, we've kind of lost ourselves in it. There's so much hype, so much news. No one can keep up. There's all sorts of mad stuff happening. Is it like hyped, underhyped, overhyped? Is there a bubble? Is there not? like no one knows um but if you step back we have this like crazy magic technology like it's magical when you actually step back and think about ai and think about the fact i was only three years old it's kind of wondrous you know i think if you went back five years ago and looked at us today and the things we can do with chat gpt with claude like with all the coding tools i cannot code i never could code in my life but i can build things now like it's an entirely it's a magical thing and we lose that so there's a new way to build it you start with what what does it make possible um what might the product be and can we build it reliably i don't know let's find out let's prototype let's try stuff you then later build the kind of uh crud part the the ux the ui you build that later you ship and then you look at scale you look at like does it work in the real world at scale for thousands and thousands of companies so there's way more possibilities but but it's way more ambiguous.

28:22It's more chaotic. Execution's uncertain. You gotta throw stuff away sometimes, start again. And you learn the most when it's live. There's also like these new muscles to grow. These are blog posts. We have a research blog. Again, like our machine learning team has been around in Intercom for a long time. It's now way bigger than it was. It's now called the AI Group. We rebranded it. And the amazing team. And I've learned so much from these things, but this is like way out of my wheelhouse, way out of our product team's wheelhouse, I think way out of some of the engineers wheelhouse too. There's like empirical evaluation.

28:58It's way more scientific. There's like scientific rigor required. You need to like really understand causality and like how things change other things. There's a whole new world to learn. And design has changed as well. This is, you know, design, when we used to build software, design was kind of the hardest part. Design is often the bottleneck. Design is often the difference between a great product and a good product. Like it's because it was better designed. This is what we do. We would like sketch out the wireframe. We'd sketch some ideas. We'd pick one. We'd iterate that a bit. And then we'd ship.