Speaker unverified: belief

Transcript context

…nk of we've had to do and then optimize and then find the right model for and all that. But one of them is customer context and that obviously answers a lot of things. What are the other smart things? Abstraction. So I guarantee you've got no pages on your website that say “Stripe works really well for a dentistry office.” You probably don't have that in your docs. A very naive rag bot will be basically like, “Well, it doesn't say dentist and we're told not to hallucinate, so no, we don't do dentists. Sorry.” And the abstraction is in that case is well what is a dentist? It's a type of business. Does Stripe work for businesses? Can dentists be internet businesses? Well, we say we're great for internet business. So you're kind of working out what's the best risk-tolerant way to make grounded inferences without going over the cliff? That's one of, like, 27 different components of Fin. Then you've got obviously your rag and then you've got, you're like, “Hey, is an escalation appropriate at this time? Have they threatened something?” Every single type of problem, it kind of ends up you have to walk through it all to actually recreate customer service. I think a lot of times people will compare it with, “How do I reset my password? Ha! It found it!” And you're like, right, cool. That's like 0.4% of the scenarios you deal with when you're in customer service. There's a pattern. Did you ever see the movie Armageddon where—it's like Bruce Willis and Ben off or whatever—but the gist of it is they train a load of, I think it's like oil drillers to become astronauts and the comedy, the joke at the time that Ben off always says he got drunk and he recorded the voiceover for the DVD and he was like, “I always said, well, why didn't we just train the astronauts to drill oil? Surely that's an easier problem.” I think the thing that we're realizing with the AI movement is some version of what's going to happen sooner? Will AI people learn how to do CS or will CS people learn how to do AI? Thankfully, as I said, we kind of started off with CS and AI in our DNA. You say people, you mean companies in this case. Will OpenAI get better at customer service faster than Zendesk gets better at AI. Exactly, exactly. I think we were lucky in that we kind of had already backed both horses somewhere along the way. se. Will OpenAI get better at customer service faster than Zendesk gets better at AI. Exactly, exactly. I think we were lucky in that we kind of had already backed both horses somewhere along the way. One thing I find interesting about what you do is every company is thinking about AI right now. Every company had a board meeting in 2023 where the board is like, “Can we do a special deep dive on AI? It just feels like it's happening a lot and we need to be making sure we're on the leading edge of AI.” And then every company was like, oh, we're actually doing a lot in AI. For example, we've seen great automation wins in customer service and so it's like the joke about the bike shed versus the nuclear power plant, where everyone has opinions on how to build a bike shed. Similarly, everyone has opinions on how to do AI customer service, and so I'm curious how you sell given that I'm guessing a lot of your customers think, “Oh, we know how to do that. It's not that hard. We hooked it up to a model that we're actually very smart on this topic already.” How you sell in that environment where everyone has opinions. Yeah, I've never seen the build versus buy thing play out more often than we do today, especially with certain— a lot of customers are like, you know that meme on Reddit, I'm not good at girls or guys, whatever. There's a lot of that where it's like, “Oh, you would never possibly understand is B2C shopping company.” And you're like, really? I've never heard of such a thing. Sometimes honestly, we just be like, “Hey look, Godspeed, you go and start building this PS here's a torture test: When you think you've got something good, run these hundred questions to us, let us know.” Oftentimes that's where they're like, “Yeah, okay, we think we need to buy your product.” But I think everyone has this idea of, in a move to AI, what can we definitely do? And we can definitely answer questions like “how do I reset my password?” And again, this is the back to the whole that's such a small amount. What they can't do is actually have conversations and all that sort of stuff. We're like, “What is your opinion on the president and how they're performing?” And a lot of times you don't want anyone to answer that question on behalf of your company. But I think a lot of times people—they dip their toes. It's almost like they fired a tracer bullet. They're like, “Yep, this seems like we're making great progress”—and every AI product has this problem where you make epic progress in the two weeks and then you hit this wall, this plateau, and two years later you're telling people, “Oh, Apple intelligence is coming in ’26” or whatever. So in this case, a lot of people start the project, feel like they definitely don't need to buy Fin. We just help them understand the difference between a good bot and a bad bot, and then they come back and they buy Fin. So where is the Fin business these days? I'm curious both just how it's performing on revenue metrics and then are you selling it to existing Intercom customers? Are you selling it to new accounts? Just how does everything work? these days? I'm curious both just how it's performing on revenue metrics and then are you selling it to existing Intercom customers? Are you selling it to new accounts? Just how does everything work? We're about 6,000 customers and growing quickly. Fin does about a million resolutions a week. We're charging a dollar per resolution, so you can do… Yeah so 50 million revenue run rate Give or take. When we launched initially we sold just to our own customer base, and then as we kind of progressed we realized, hang on, moving help desk is a nightmare. You've probably done it once or twice. It's a whole ordeal, and Fin is brilliant. So we're like loads of people want this product but can't buy it. So we made the decision to launch what we internally call Fin standalone, or Fin for platforms. So now you can use Fin on top of Zendesk or HubSpot or Salesforce or any of those, as well. So we basically—Fin is available to everyone, and that's a relatively new muscle that we've been growing, but it's actually—that's kind of where we see a lot of the future growth And so is Fin like, you can connect your iPod to Windows for iTunes for Windows, but we hope that one day you buy a Mac and it's part of the whole digital hub strategy, or we're actually now all in on Fin the engine and whatever customer service platform you use is actually not a topic of huge interest for us? This is such a core question that we kick back and forth quite a lot. This is the offsite debate that just… Yeah, genuinely at least it's certainly one of them. The way we think about it first and foremost is the future is AI. So Fin just has to win kind of at all costs, including our help desk. Weirdly, our customers are like, they turn Fin on and are like, “Damn, this thing's good. Hey net 65% of our support volume, maybe we don't need X, Y, Z competitor and maybe we can go all in on your help desk too.” And we're like, okay, cool. That wasn't our game plan, but we're happy to help, if you know what I mean. I think the actual battleground we care most about, genuinely, has to be the AI agent. That's the one where we care about most, but it does produce a lot of demand for the actual help desk product too. Oh my god. Oh dear. Okay, So you've actually played darts. I've been around a dartboard. I'm curious what your AI stack looks like. Where concretely, what are the models or collection of models and prompts and everything that you are using in production? How do you handle model upgrades given that the behavior is changing so much under the hoods? How deep do you go in terms of developing the stack yourself? Then you can talk about the stack.…

Stored transcript either side of the excerpt. The highlighted words are the published quote; the surrounding text is unedited source, never generated.