SaaS is dead: This is how Intercom became a true AI first company and made $100M from ONE product.

Des Traynor (0:00)

Humans will tolerate any amount of chronic pain to avoid any type of self-afflicted acute pain. The way that translates into business is that people would rather be wrong slowly than be right and have to make a painful U-turn.

Wouter Teunissen (0:11)

That's Des Traynor. He co-founded Intercom back in 2011 and grew it to one of the biggest customer communication platforms in the world, with over 300 million in revenue. But in late 2022, growth was slowing down dangerously. This is when Intercom made one of the boldest pivots betting the entire company on an AI agent called FIM. Today, less than three years after launch, FIM is approaching 100 million in revenue and resolves over a million customer conversations every week.

Des Traynor (0:37)

For us, it was painful. Like, Owen challenged me to go and set up a team FIM, rob a load of engineers, isolate them from the rest of the work, make a whole FIM product, get a different sales team to sell FIM standalone. Every single one of those changes is painful and you're pissing somebody off. You're taking somebody away from your favourite manager and you're changing forecasts. But that's the kind of cost of entry.

Wouter Teunissen (0:54)

From this episode, you can expect to learn the exact steps to turn your company into an AI-first business, and why you need to stop building software like a factory and start building like a lab.

Des Traynor (1:04)

To run a company that can withstand changes, like there's basically three parts to true AI. There's certainty, where you make a prediction and you hope it's true. There's existential dread, where we all show up for work every day and ask, what does it all mean? And then...

Wouter Teunissen (1:18)

I think actually one of the things I want to start with is your PhD, but specifically what you did and why you decided to stop, because in the history of Intercom, when I was doing the research, you guys took a lot of unconventional turns. It's not a really linear path at all. So I want to kind of get into your thinking about how you made those decisions. Tell me about your PhD research and why you decided to stop.

Des Traynor (1:39)

So I was a good student in university and the.com boom hadn't really hit Ireland as strong. Obviously, I graduated in 2003 So the crash had happened, but the idea of going straight into a startup whatever wasn't on the table. Right. So what did the smartest kids do? They all went and did PhDs. At the time, there was a lot of funding or scholarships available. So I was kind of looking to do something, and it is a bit of like audience capture. It's like, if I wouldn't be a smart kid, I better do things smart kids do. So at the time, I was pursuing a PhD, and usually when you graduate, there's like all the professors and lecturers would have like, here's the research areas, and here's the funding we have, what do you want to research? And computer science education came up, and I was definitely bemused in my own class. I think, and my research friend is to be true university, the bottom, say 60% of the class can't really code. They can't really do the job. And at the time, I probably have better freshers thinking on this today, but at the time, I remember thinking, wow, how is it like that? We've been here four years, and like there's people who can't like, you know, and I don't mean they can't literally go, they could do Hello World, but they certainly couldn't solve a real problem in any real way, like write something to a file or whatever. And it was a substantial show. Maybe it wasn't quite 60, but it was like too many. And I remember thinking, jeez, well, what is medicine like?

Wouter Teunissen (3:08)

What is law like?

Des Traynor (3:09)

I mean, can you imagine producing like the majority of the people who graduated out of the medical degrees, can't actually do anything about it. It kind of just seemed so foreign to me. So I was just kind of fascinated by that. Like, why is it we are seemingly tolerant of this? And maybe it was just adopted at the university, but it really wasn't. So my research was into like, why is it we are so bad at educating or teaching people how to program?

At the time, the thinking was, oh, it's too much like MOTS and we should have MOTS as a mandatory requirement. And if you were to do it, it was a few other kind of path theories that I was exploring. But ultimately, where I zoned in was actually, it's our error, like the feedback mechanism, which is how you assess if somebody can program is what was broken. And I kind of zoned in very sort of specifically on, can we automate a way to get students real-time feedback basically as to what or not they're actually getting the concept? And this is obviously, needless to say, 22 years ago was before AI.