Intercom’s Generative AI Playbook To Build Hypergrowth Products with Des Traynor, Intercom

This episode is brought to you by Intercom. Connect with your customers at exactly the right moment using powerful messaging and automation. Scale your customer service without additional investment while still providing a fast and personal experience. Apply to get a 95% discount@intercom.com traction that's I N T E R C O M.com T R A C T.

I o n Very soon you should be pivoting towards a AI empowered or infused roadmap. Because if you're not, someone else is. And like it won't be close. We happen to have been building product at a once in a generation tectonic shift and our options are cling on to the past and die or cling on to the future and thrive. I need some traction. You need some tract. Get some traction.

Hey what's up innovators, entrepreneurs, visionaries and disruptors. This is your Traction podcast host, Lloyd Lobo. We're a community of over 100,000 people just like yourself, on a mission to help you get the methods, the money and the madness to explode your business growth. Featuring stories and tactical advice straight from those who've done it before, like Shopify, Twilio, Asana, and many more, Generative AI is all the rage these days. I can't open my LinkedIn without scrolling through dozens and dozens of posts and carousels on the topic. Thousands of generative AI tools just swarming the market and VC funding for generative AI startups have exploded with 2 1/2 billion in funding last year. But regardless of the hype, the generative AI has the power to transform the way we do business and reshape the entire customer journey. And we're not just talking about looking at old data and improving it a little bit. We're talking about rebuilding the future today. And I'm excited to interview one of the pioneers in the space, one of the smartest product people I know, Desk Trainer, Co Founder, Chief Strategy Officer and Board Member at Intercom. Guests at Intercom covers many areas leading R and D, product engineering and design. He's also a renowned speaker, blogger and a visionary on product strategy. If you don't already do, check out his blog, Phenomenal Insights and he frequently represents Intercom and speaking at international conferences. He's been to Traction a couple times already. Plus he's an active angel investor in more than 50 startups with breakout successes like Stripe, Notion and Miro. So Des is going to demystify the most popular buzzword of 2023 for us. Welcome back to Traction Desk. How you doing?

Doing very well, thank you that's quite an intro. I hope I can live up to it.

You're more than living up to it. You know, it's funny, in 2013 I was working on a startup called Automatically, which was a chatbot built on top of Zendesk to offer like basically look at historical data and respond like real humans. And, and being the arrogant engineers that we were, we couldn't get it to work because Zendesk customers were so small they didn't have historical data. And so it was spitting out gibberish. And we tried for a while and people were like, make this stop. So we made it editor approved, but they're like, we're editing the whole sentence. And so we gave it up. If I knew then what I know today, I would have just done a decision tree, but then came along intercom and I'm like, hey, you guys solved the problem we set out to solve and now you've taken it further. So excited. Excited to see what you guys are working on. But what's the hottest buzzword of 2023? I'm using it every day. Most people are using it every day. Everyone, including VCs seem to have gone from blockchain and crypto experts to gen AI experts. What's your take?

I do want to offer one comforting word, which is it is true to like 2 years ago everyone was talking about webtree or blockchain or crypto and now they're all talking about AI. But also that is the kind of job of a VC to jump into new areas and very quickly familiarize yourself with what the hell's going on, understand the trends. That's like literally the gig. So we should not too surprised when a new technology emerges and they become override experts because like, I'm not going to say that they actually are full experts, but did they back their entire opinion up with like tens of millions of dollars, granted, over people's money to find winners? So it doesn't surprise me and I think it's okay that they aim to gather a lot of expertise very quickly. So to the higher level question, what's going on? The way I just think about it is for the longest time AI was this mythical thing that we weren't sure we were ever going to see really come to fruition. I think there was early warnings out of OpenAI that they were going to do something pretty cool, maybe as far back as like two years ago. I think once they launched the playground where you could actually just interact with text and it could do some pretty smart stuff that was pretty impressive. Then Dall e was obviously pretty impressive from an entirely different angle. And then obviously November 30th, I think it was ChatGPT dropped, basically. Want to say like the whole world from a technological perspective changed overnight. Certainly that was, you know, the closest thing we would have an intercom to like a code red type scenario of like, hey, some of the fundamental assumptions about our business, about what our customers want to do, have been irreparably and massively changed based on what is now technically possible. So we have to get to work on that. When I take a step back and look at the actual trend or like, what's been happening, I don't think it's hype, I don't think it's bluster. I think some people, specifically professional Twitter commentators who, like, do soft boy think fluencer treads, they're tempted to just try and like pile on. We've got AI is going to write like this and it's going to do that. Oftentimes I feel like I want to correct them and say that's not yet technically possible. But my corrections are largely academic because five weeks later it will be possible. I do still think what is like literally live and available today is groundbreakingly different. It's bigger than mobile, it's bigger on the cloud, it's bigger than Web two, obviously bigger than Web three. I think what is live today checks all those boxes, so it actually doesn't need any further hype on top. However, not only is it a breakthrough period, but the rate of progress at the moment is so astounding. The new capabilities, even like with the, say, most recent, the plugins that OpenAI released, even that alone is just like another orthogonal dimension of power that is now possible. We were all happy and impressed with like just chatgpt, but it just keeps going in different directions. And to give my sort of take on it, I think this will be the single biggest disruptive layer to the startup and generating a software ecosystem that we have seen in my lifetime. Perhaps it's as big as the Internet, but it's definitely somewhere bigger than mobile because it affects not just startups, but affects the real world too. You don't have to be a professional Internet person to be dramatically affected by what has happened. And I think every product category, whatever your software does, is going to shift in a huge way either because AI can do the tasks that the software exists to do. So for example, like find the meaning in this data that used to be like 27 mouse clicks and five tableau dropdowns and allowed SQL. They just ask a pretty dumb question like what's going on in this data set? Or what are the abnormalities? And it answers them. But even if it's not a direct application of AI, I also think interface design and like how people engage with software will change. I think there's evidence of this we're seeing like say Dharmesh Shah from HubSpot, he's building a tool hacking around a tool called Chat Spot, which is basically a plain text interface to HubSpot and a few other tools. What's obvious there is one of the things that the large language models have made possible is the idea that if you simply know the thing you want to know, so if you understand the thing you're trying to do, you can literally just explain it in English and it will get done. So to give you a very simple example, and I'll hand back to you, a lot of us have Google Analytics on our website. Most of us don't really have a clue how to use Google Analytics properly. So if you said to me, hey Des, buy the intercom's biggest referred on ends in co.uk and sent us the most traffic in March, like I know that that's technically possible in ga, I also know it could take me approximately two to three hours to actually get that done. But in a world where Google Analytics has embraced the LLM style plaintext interface, I would just type the thing I'm trying to do and it would just go and do it for me. That isn't just an optimization for the desert of the world. That means all of a sudden overnight everyone's a Google Analytics user, everyone's an expert. So it totally changes the addressable market. And the sort of the fundamental nature of product categories is now so much bigger. It comes back to this idea of if you can think it and you can explain what it is you're trying to do, it can be done. That's a step change for like productivity and honestly for society. So if that's not enough, I can answer more, but I'll hand back to you and see where you want to go.

Historically, people have built large businesses by selling crappy software just on the virtue of it's must need it. And then it spawns this whole cottage industry of consultants who teach you how to use it. You guys had a very early thesis on Chat UI as an interface and you must be laughing as you see that taking over the world. Because if anyone could just talk to it and get the answer they want, then truly it increases the addressable market for every piece of software. So then software that have historically been built on, hey, you know what, let's be sticky by virtue of just being painful will cease to exist.

Absolutely. And then there's a second order effect, which is if you can describe what your actual job is, you can just ask somebody to do it for you. So like, one order of this is just if you can explain it, like how to, let's say even AdWord optimization tool, you can imagine one version of this being like, Dear AdWord Optimization Tool, you'd say, find me my most performant ads and put all the spend into them and run 100 variations of them with subtly different words. And then please re optimize or something like that. Again, this is something that would otherwise be a lot of mouse clicks, a lot of dropdowns, and a lot of like copywriting tweaks to do. It can now be kind of done. And then if you just take one step back and be like, what if you say, hey, GPT, I want you to log in every day, run the following checks, see what's working, what's not. Turn off the ads that aren't working. Turn on the ones that are re optimized, produce multiple variations. Look at our product launch feature list, here's the URL. If we've launched anything new. Run ads for that as well. Turn them off if they don't work. If you can articulate your job, then all of a sudden you start to ask yourself, what does the interface for that product look like? Because you've literally said, hey, here's everything I want you to do, go do it. Then you could add on like an email me a performance report every Monday. Well, like, when you get there, you start to realize the very nature of what software is and what it does and how we humans interact with it entirely changes, you know, and like, there are so many jobs in the world that exist to use software to look at performance metrics or to see who's working what shift. And now if we can describe the job, it can be done. The repercussions of that are still to be seen, but I can't help but think we're still not fully gripping what's possible.

And you know, the familiarity of the interface makes people more likely to use it, right. Almost everybody I know with a mobile phone, their first mode of communication is a chat interface. Whether it's WhatsApp, whether it's Telegram, whether it's Discord. And then for business, it goes in an Intercom or Slack, they are already familiar with it. And so then for me, eventually I spend like 80% of my relationships, my conversations on chat, not phone, not email, 80% on chat. And now that I got to log into this clunky piece of software and click and so on, Everything at a chat interface, it just changes the game. It gets more people to adopt your software. And a long time ago we had this chat on onboarding. Onboarding being the leading indicator of engagement and retention. I mean what better way to onboard people than have an interface like that?

Totally. One other thing that we're again will have repercussions is so like generally speaking we're saying like people can type faster than they can write with a pen and paper, right? But they can talk faster than they can type. These are all these different modalities. And generally speaking the gap becomes if you say if measure the gap between the thought and the expression of the thought. So going back to our Google analyst example, I wonder what traffic is working in the or what ads are working in the uk. You can have that come out in a second and it can be like 2 error feedback loop before you get your answer. With a text driven UI it's going to be a lot shorter. But with a voice driven UI is going to be a lot shorter again. OpenAI have also fired out like Whisper, they're like real time audio transcription. And then on the other side you've companies like either Synthesia or Synantic or Papercup whatever who are doing lots of really impressive stuff in the generative voice technology space. So you can literally imagine that movie with Scarjill where she's like a digital assistant to him. We're very close to that and we're not far off things like hey Workday, please book a two week vacation for me on August 1st. Or like in some like even more optimistic world there'll be some sort of zapier plug into workday that's triggered by your Google Calendar or something like that. Well like very quickly you end up basically just talking to computers and hearing things back. Or if you're a library or an office, maybe you're typing where I just think we're going to end up multimodal input and output. So when I'm driving I'll be talking to software and getting audio back and it all feels great. Like hey Intercom, one of my most pressing issues in customer support today. And Intercom will reply back in Intercom's tone of voice to answer that question. And then if I step into I don't know, a library or an office. I can carry on the conversation by taking my phone and just typing away. I just think that's going to be where software is going to go. Like this multimodal mixed method communication between people and software and people and people. But that's different. Again, I just think this is what's happening. And if your roadmap is clinging desperately to the idea that this disruption won't affect you, I don't really see much help for you.

Customers don't want software, they want outcomes. I don't sign up to the gym to just go and push some weights. I go there to get a six pack. I don't sign up for a marketing automation tool. I go there to get more leads. And we're in an age where an interface and tectonic shift will help you get the outcome without clicking and pushing knobs and buttons. So that is the most exciting thing.

That is very correct. That's the right way to approach the roadmap, which is customers come to you with a desired outcome or a job to be done, a certain state they want to get to. And the right way to build the software from this point forward I think will be to say what is their desired endpoint? And they might say, look, I want to be running an optimized ads campaign or I want to be kept aware of all my projects and if they're running on time, wherever. And then the only question you have to ask is what further information or decision points do we definitely need the human for here? That becomes like, where will the AI not yet work reliably? Right? So as in it probably can't read every single JIRA ticket and story and make the right accurate inferences and reprioritize the roadmap. It's probably not there yet because there's still stuff in humans heads that JIRA didn't capture or whatever. Like as in there might be knowledge that hasn't been transcribed, or it might be decision logic that the AI can't faithfully replicate. I think the right way to start is let's assume if the user could, they would just express the thing they're trying to get done, press return and walk away. They don't care about your gradients, your fancy buttons, your beautiful dropdowns, your typography. They don't give a single shit. They just want to be able to say here's the thing, I wanted to get done, return and walk away. And the only reason you should come back to them for more information is because you're either Looking for further context or decision logic that you haven't already caught somewhere else. That is the future sort of software design method. I think I could be wrong to be clear, but that's how I see it going.

Everything you know about user experience or you learn in school, you almost have to throw that away and reimagine this, right? Like no button, no optimizing for conversion rate optimization. It's the simplest thing which has the most explosive growth. It's a chat interface. Hey, I need to speak with this type of customer in this geography and I want this type of profile and then it spits out a CSV. It may not have the fanciest interface, neither is intercom chat or chatgpt, but it works, it gets the job done. And I think ultimately that's what it is. Bring it to its simplest form, but give me an outcome. I don't care about putting lipstick on a pig, I want the outcome. That is great. What was the single most thing that tipped this? What caused this shift? Because generative AI is while it's buzzing right now, OpenAI has been working on it for years. We had access, I'm sure you had access a couple years ago, but AI as an industry, people have been talking about it for 10 plus years. So what changed that caused this huge shift?

Spectively, I think it was OpenAI training the best large language model on the entirety of the Internet effectively up until wherever it was November 2021. I think that was what really enabled this all knowing thing. I do want to give a small shout out to whoever the designer, specifically the UX designer who designed ChatGPT because I think there was something initially compelling or addictive about that experience. Right. And it's like subtle things, it's like it's the typed computer output vibe that was backwards. It's a little bit of a subtle sort of tone of voice but it made it engaging in a way that we hadn't seen before. And whilst clearly being a competitor to Google, they didn't take massive inspiration from Google. The last thing I think is we're much more used to computers and specifically from like a search or like task point of view as being stateless, like single shot queries. Like if you do a Google search and then you do a follow on search, it's not very obvious to the end user how Google has like connected the two of them. Whereas if you say something like buy me the best places to eat in downtown Dubai, ChatGPT will spit them out and you can say oh sorry, I meant only Vegetarian ones. It adds that to the previous conversation and it comes back with a better answer. And then you can say that are open right now. And then you can say that like that costs less than 100. But you can daisy chain your thinking. Whereas if you think about how you do that in a Google world, you're going to end up at Yelp or OpenTable and you'll be progressively applying filters, but in a very clunky UI way. So there's something addictive about that. And I don't see, you know, searching for restaurants might seem like an easy one that Google or OpenTable could do, but another version of this and be like, write me an email, write me a song, write me a story. Or like summarize a project. No, please be longer. Please explicitly refer to people. You can just keep adding this thing in. I think that has become a more intimate form of computing where you actually feel like you're having a conversation with the product as opposed to just issuing one single shot commands.

What are some other use cases or examples of generative AI being used in different industries outside of the startups and the software we talked about?

We are so early into this, so the answers I would give will be nowhere near complete and we'll be out of date in like six months. I want to precede this by saying like GPT Turbo, which is what ChatGPT is built on. And then GPT 4 Turbo I think is like three and a half months old. Four is like I want to say six weeks old, five weeks old at this point. So we have yet to see commercial availability of the best of what's possible. But I have seen T shirt companies, I've seen children's storybooks where you say, write me a great story about my child called blah. Here's a photo of her. And it can go and use mid journey style, Dall E style generation to actually build illustrations. And it'll write a story. And whoever the child you fed to it was is the main protagonist in the story. It can do that. I would then you've seen other like everything is software in some sense. Well, like, you know, I'd say obviously we're using it to do customer sports, right? We're using it for payments, to detect fraud or whatever. You're going to consistently see anything where like large data analysis or anything where like repetitive judgment allocation is applied. You just got to see those things just get entirely like rewritten from the ground up. The other stuff we have seen, like we've seen AI to generate movies. We've seen AI to generate music. We've seen AI to generate obviously visuals and imagery. I think you're going to continue to see all of that. The next stuff, I suspect the next order we'll see is like songs and books and movies have a start on an endpoint that's for no good reason other than the fact that the directors top down or whatever. I think the AI doesn't get tired. So you can have like a world that goes forever or a movie that goes forever. They also generally tend to have like just one camera angle. In a world where you're actually generating whole worlds, you can tell a story from multiple different angles and produce multiple different scenes describing the same scenario. I think books can be effectively endless. They can choose your own adventure style. Books will have a whole different meaning when there's AI involved. There'll be new types of art forms added to what we currently have across everything from like art all the way through to music, sound, movie theater, you name it.

Fantastic. You know, I've been playing with Mid Journey quite a bit for tattoos and designs and all kinds of things. T shirt design. So definitely epic. And we haven't seen the end of it. But I am really bullish on the whole chat interface and you talk to something and say, give me just this. And it specifically spits that out. Now let's shift to intercom a little bit. What are you guys doing and what does it mean for customers and their interactions with brands? Like, how can it be used to create personalize experiences and drive this loyalty which for the longest time customer service has? You know, although with all the tools and everything else, still it leaves a lot more to be desired. Right. Because it's still in the hands of humans to cause the interaction.

Yeah. So most support teams are under extreme pressure. They're under extreme pressure because oftentimes businesses devalue support. It's either the call center, the data priority work, what's the minimum we can get away with as opposed to what's the most we can do? Generally speaking, any given customer support team, if you ask them like, hey, how are you staffed relative to your caseload? They'll say they're under pressure, under extreme pressure. And then obviously things like layoffs affect that. Things like the grand digitization caused by Covid have like increased the load, etc. So what can change on how can we make this a better experience for all? Well, ChatGPT launched. The first thing we did was we started the work getaway early on. You might recall there was like a hallucination problem where if the intelligence didn't know, it would just make shit up. And that's obviously risky from a customer support point of view. So what we figured we could do would be agent assist. So our first release in the space was in late January, maybe six weeks after we first experienced ChatGPT. And what that was was in the inbox. You could do things like automatically summarize conversation, automatically span upon a piece of text. So if you knew the answer was like, no, you're not getting a refund, you could say it up and you could say please expand and fill in the gaps here and a few other basic things that you could automatically publish an article or generate an article. What we were really targeting there was how can we take some of the undifferentiated heavy lifting off of the support team to decrease what they would call their average handling time, which is how much effort you have to put in to close the conversation. That release was wildly popular. Thousands of businesses using it. Right after that went out, we started to work on, we believe to be like, how can we produce an end user facing bot that recognizes and answers to most common queries? Now we have like a lot of prior art in this space. We had built resolution bot before, we built numerous chatbots. But the difference with ChatGPT was its judgment was pretty much impeccable and its knowledge and its ability to articulate and formulate answers was really, really good. The biggest effort we had to put in was how can we constrain it so that it stays on topic, it doesn't deviate and start offering like political opinions or anything like that. Because brands don't want that. Brands want a boss that can just literally be on top of grit. And then separately, how do we make it not make up stuff when it doesn't know the answer? Because the danger of having an end user facing bots, that's bullshitting. Is that the brands won't even know what the question was and the user won't even know that the answer they got was wrong. So it could cause widespread confusion. So we honestly spent the next while working out what are the guardrails we can put on a bot that would face end users to keep it on topic, to keep it only speaking when it was high confidence, keep it trustworthy and ultimately like have a just a high accuracy, really fast bot. That culminated in the launch of Fin, which is intercom.com fin. It is chatgpt for customer support. It stays on topic, it talks to you about your own business. It doesn't deviate. We have it in beta at the moment and it's answering like thousands of questions a day and it's doing really, really, really well. We're really blown away. So it does like, I've seen it push back on users. Someone says, hey, I tried to do blah blah. And Fin will reply and say, tell me about the exact steps you tried to do. And I'm like, where is this coming from? And then the person like, right, well, I tried to do 1, 2, 3. I'm like, ah, you made a mistake on step two. And I'm never going, where is it? The intelligence is staggering at times. The way Fin works is it learns from your knowledge base. You can give it any amount of URL as you can say, please read the following 10 sites and fin will go and consume all that and add that to its knowledge collection. It'll go and read your entire conversational history of everything your support team have ever said before. And it starts to formulate kind of a really good understanding, ultimately a very authoritative understanding of what the hell your business is about. And then it answers questions on your behalf. And the resolution rates we've seen are frantically staggering. I can't understate that. The highest cases we've seen are like high 70% resolutions. It's insane. Like just devastatingly smart. Now how does that actually translate to good customer support? I think two things I'd say. One is a lot of queries to customer support teams are, I would call it transactional type queries. Hey, how do I reset my API key? Or hey, I'm just curious about where's my order? And these are not what I would call brand building opportunities. Or put another way, there are questions where the asker, the customer, they just want an immediate answer. That's all they want. And if you start coming in, hi, my name's got me. Please wait a while while I go and type up this response, blah, blah, Even like a 20 second delay, is it really worth it? Versus like an instant response. So that's one thing. Which means happier customers because they're getting instant answers to their questions. And then the second piece is without reduced workload because of the repetitive questions being taken away on the high emotion, high urgency, high stress, high complex issues. Now the support team actually has the time to get into it and handle it with a lot more diligence and a lot more care and perhaps empathy, depending on the issue that they just wouldn't have been afforded on a time when, if they close your 60 tickets a day and like Ooh, this one's complicated. They didn't want to do this one.

Once you.

Freedom of the shackles of the repetitive work. I think you'll see support teams being a real force for change, a force for good, and a force for ultimately happier customers. And I think that touches on a thing that we see a lot of which is like Jevons Paradox. The idea being that as something gets more affordable, you tend to do more of it. I think what we'll see is because you don't have to spend so much money on repetitive support work, you'll realize the value of what's left is actually a lot higher and you'll be tempted to do a lot more of it. So I expect we'll see a rise in proactive support, a rise in like, perhaps like more personal support, be that like video chats or whatever. But I think you're going to see a shift towards people saying all of the undifferentiated transactional support volume, or most of it is gone. And bear in mind, Fin is going to keep learning it. So, like, one vision we have for Fin is the idea that, like, your support team should answer questions for the first time and the last time. So Fin hasn't seen it before. So it goes through the humans. The humans answers now finish Finn. Make sure it never happens again. That type of experience is what was what we're dead set on building. And then what's left is, I think the brand building opportunities, the stuff where it's like, hey, I've already complicated issue. I ordered a bike off, you arrived, it's damaged, I don't know how to return and I'm very annoyed. It's for Western Birthday. That's where they're like, hey, let me jump on a call with you right now. I'm going to trash this out and we're going to get you set for your son's birthday. That's the opportunity where you can actually delight and surprise customers. But that's only possible in a world where they don't have 65 people asking, how do they reset an API key? That's how I sort of see it.

All playing together, deploying it at scale. What are we learning here? What are the key learnings?

I think there's a few things thinking about building on a large language model off the shelf, be it like OpenAI or Anthropic or Cohere or any of the crew. There's a few considerations that are new. One of them is obviously around cost. Right? These things are not as cheap to run as like sending an SMS by Twilio or something like that. And what that means is there are features that you have to first ask yourself, can we afford to do it? Will it make sense? So intercom powers half a billion conversations a month. I think Griffith figure could be all this, probably more. It would seem like a cool feature to be able to always maintain a running summary of every single conversation, but that would probably bankrupt the company if we did that. It's a rare muscle for a product manager to have to be say, hey, things we could build, but we can't afford to in a sense. Right? Because yes, it's technically possible, but the value isn't sufficient that we could pass the cost onto the user, nor is it small enough that we could just eat it ourselves. So cost is a huge thing you have to think about. The second one that I suspect will fade away over time is just latency. Every one of these calls you are going to a third party and like hanging and waiting for a return. Obviously that's still quicker than going to a customer support representative and waiting for them to type the answer. The bot's still faster, but it makes you think about like, what are the right experiences and what are the right order in which to do things. The third one is just around prompting and customizability, I think will become more of an opportunity for differentiating association. So what does that mean? Well, right now, oftentimes these, all these bots speak with the same basic tone of voice and you might want that to be different. Like you could imagine, take like an online bank, take like an online funeral service, take like an online cannabis dispensary and take like a young girl's fashion brand or some of that. They probably want four different tones of voice. The room for playfulness, creativity, interesting. Language is very different in each of those scenarios. And right now, obviously you can prime it by saying, please speak like a whatever. But like, I think you'll see people invest in differentiated generative experiences to proxy a tone of voice. Bluntness, conciseness, elaboration, whatever. These are things you have to think about because a danger might be like you homogenize thousands of customers by giving them all the exact same thing. That's not necessarily what you want to do now in Intercom's case will follow and end up following the same tone of voice that you've used in your help center articles or in your historical customer support conversations. So there is a kind of natural mimicry there to your business already. These are the things that we're thinking about and then if you step one degree up from a high level product strategy perspective, like people often talk about moats in software, I generally think that that tends to be overstated unless it's like a network effect or a community or a brand. Like generally speaking, most of the stuff people build can be built by other people. Like we can all right click and view stores. No matter how beautiful your UI is, I can work out the CSS you use to generate it, I can generate my own version of it. Similarly, most SaaS apps are ultimately data in data out. I can work out what API calls you're making, see what you're doing with the data, and probably just build a version of it. So long story short, things are pretty easy to copy in our world. So I've never really believed in product as a moat, full stop. However, it's doubly true that if you're outsourcing huge chunks of your functionality to a third party that hasn't natively available, that it's going to be a tenuous moat at best. So you have to really ask yourself where do you differentiate? And I think it'll become harder. One positive effect for society might be that software becomes slightly cheaper because of this, because it becomes ultimately easier to build. Not just because GPT can go and write code for you, which it can, but also because functionality like you'll just have more startups able to do more things competing with each other. And generally speaking that tends to form a spectrum where you have like the cheap version, the middle version, the expensive version, and the expensive version has a premium brand, top tier service or whatever. The cheap version might be like lighter, but ultimately it results in like the democratization of software where most people can use things. So maybe you're not rich enough to afford Asana, but maybe you can afford like the budget version of a SANA that doesn't have all the nice typography or maybe not the most beautiful UI separate. Don't forget our comments about UI going away. So that's another realm of differentiation that could be irrelevant. But just generally I expect we'll see more software and I think that more software will find different points on the price curve, possibly like geo specific stuff or use case or vertical specific stuff. But I just think a lot of things that it wouldn't have been quote unquote worth building will they'll be worth building because it's easier.

Now you advise invest in a number of startups. How many angel investments you've done? Did I have 50 plus, maybe it's 100 plus.

I think it's like 90 plus or something like that. I don't know if it's 100, but it's somewhere around there. Yeah.

And being like one of the pioneers in the space, pioneers of chat ui, you're probably getting hit up by a lot of founders. What advice advice are you sharing with them as they brace for this?

I don't know if the moat thing will play out the way everyone wants to do it. I think what's going to happen is UI is going to get reduced to text input and functionality might get outsourced to OpenAI. You just can't sell a text area, you know what I mean? So I don't know where the moats will be built. I think brand will be one, I think community will be one, I think network effects. I think perhaps platform and integration like interoperability will be a mode. With that said, what's my advice to portfolio of companies I'm invested in? The assumption of a startup is you can move insanely fast. You literally wouldn't start one if you moved at the same speed as the incumbents. You're into Disrupt and that does all sorts of other stuff. We can talk about in that regard, but let's just take it as an axiom that startups have to move as fast as they can and at that speed is a lot faster than the incumbents. What I would say to any startup is basically what is your angle attack on the market you're going after? And they'll say, no, we're better UI better up or we connect better tools or it's a subtle use case in H order no one ever saw before or whatever. I'd say, okay, well then what's changed in the landscape so far is I do think UIs will fade back to simpler inputs text boxes. I do think you can now generate some risk, expand collapse, change tone of text in a very easy way. I think you can answer complex questions given a prompt answer a question. I think OpenAI is very good at that. You can generate visuals, you can generate artwork. These are all things you can do. So if your product is. We are brilliant at sending newsletter campaigns to audiences. What I would say is it's probably a lot easier to pick the right audience in a post GPT world. It's probably a lot easier to generate better artwork in a post GPT world. I'm personally a bit bearish on the text generation stuff, but you could probably generate variations of call to actions in a post GPT world. So Everything about your product space has probably changed and you need to work out where are the most incisive. If we define incisive as being most likely to convince somebody to switch to your product, what's the biggest pain points that you can take away? Because you can build faster against OpenAI or anthropic or whatever than the incumbents can. So, like, if you're going after mailchimp and they say, oh, the biggest pain in the ass of mailchimp is maybe it's audience selection, maybe it's visuals. Whatever you say, right, well, that's used in new tech. Take away the biggest pain point and let's go after their customers and say, we have this tech now. It is true that the mailchimp or the campaign monitor or whoever in this space will get there, but they might get there in two years. And in these two years, that's your opportunity to build your brand as, like, the future of the industry and ultimately to use these two years to create many other angles of attack, such that when they arrive a day late and a dollar short with this feature that you launched, you have like 24 or more of them. So that's really it. Use speed as your strength in your opponent's weakness, to use the new technologies to exploit the weaknesses of the incumbent product area. That's the actual plan. I can't be much more descriptive than that without getting into a specific example, like I did, you know, but like, generally that that's the angle of attack. A lot of startup founders I spoke to are like, they're very okay, but we were halfway through this product sprint and I kind of get that, which is like, oh, we were building out this beautiful new reporting ui. Does that still matter? Like, honestly, the chances are it probably does, especially if you have customers. I understand the logic or the reasoning behind let's finish what we were doing first, but I think very soon you should be pivoting towards a AI empowered or infused roadmap, because if you're not, someone else is. And, like, it won't be close. When I look at, like, say, customer support platforms that don't have a generative AI chatbot like Fin, I'm not sitting here scratching my head going, oh, I wonder which will people pick? They're going to pick the one that resolves most of your common queries immediately. So I do think if you are leading a smaller startup and you're not getting on this and you think you can kind of narrativize your radar by saying, you know what, we don't need AI. We're just a really cute note taking up whatever. I think you should first ask yourself, are you just lying to yourself to preserve what might be a difficult conversation? It's okay to go back to your team or your investors, your employees or whatever and say, hey, I'm sorry, but we happen to have been building product at a once in a generation tectonic shift and our options are cling on to the past and die or cling on to the future and thrive. I'm choosing the future. And yeah, that means we have to abandon a bit of work we were doing. Those things always seem painful in the moment, but it's a no brainer when you're the outsider. In this case when you're me talking to the portfolio company, it's a no.

Brainer for me to say, love that analogy. Cling on to the past and die or cling on to the future and thrive. That is a great way to prove it. And a lot of people, if you see, and this is probably controversial, right, with the AI expert at Google quitting and these letters coming out and saying it's dangerous, it's going to spawn a bad era for the universe. What is your take? Because there's a lot of controversy also around this.

I don't think there's been a single new technology that hasn't produced a shitload of controversy. You can literally go back to Plato had an issue with the concept of writing because he was like, how will people have pure thoughts if they're always expressing them? And what is the nature of knowledge if you can write it down? Like the radio, the Internet, the newspaper phones, I mean I remember it was like, you know, if you were like Foursquare, the location checking app, like that used to get a lot of bad press because it was like, oh, it's tracking you as you move around the world. So I think one, there has never been like a significant new tech that hasn't got a lot of like safety as trolling, for lack of a better word. But two, I do think this is different. I don't think it's different from a point of view of if you're working on a project management startup, I don't think you're accidentally going to start a nuclear war. I don't think that's likely. And this isn't anything to do with any of the commercial players. I think the reality is the GPT paper by Google is public and large language models are a thing and they have leaked and you can get them and you can Train your own. So really anyone is now capable of building their own AI. I think things like digital SIOP type stuff are a lot more possible now. So you could probably spawn hundreds of bots to do things in a much more conceivable way than you ever could have before. And you can use things like fake voices and all that sort of stuff. So there'll be a lot of scamming and spamming. I think that's really, really bad and deconstructive. And we should definitely change our legislation and allow us to make it a properly punishable offense. And by properly punishable I mean like carrying a real sentence similar to like a bank robbery or something. But I don't think we should be letting like the potential downsides of this tech limit its potential upsides. I think that's a very pessimistic view to have on society in general. When I look at other things that have been happening with ChatGPT, there is this famous education study by a guy called Benjamin Bloom, he's American pedagogical professor and he produced this study called the two sigma effect where he said basically there is a two standard deviation difference between somebody who's had exposure to mastery coaching one to one versus somebody sitting in a classroom. Two standard deviations. That's just basically a massive, massive difference. Right? And whenever I want to know something new that's complicated these days, I go to ChatGPT and I ask it and it coaches me through it and I realized that we have a massive sense democratized our capacity for education. Actually you can say where do they go wrong in this logic equation? And it will correct you what's wrong with this line of code? Why doesn't this compile. Where am I making a bad inference in this statement? What is wrong with this sentence or paragraph of written English? How could you improve it? All of that is now possible. And whatever at the me is of the world, think about the long, long tail of humanity where they don't even get necessarily great classroom education and now they actually can get world class education for free. And that's just one little example. There will be applications. In medicine there'll be applications. And law, there'll be applications and so many things. Where you ever get a legal contract, Most people can't afford lawyer to read it. You can now paste it into GPT and be like can you explain this to me? Like I'm a 20 year old who hasn't studied a lot in their law because most of us haven't. So I think the upsides are just so massive that I'm so, so wary of people who only preach to the downside. That said, there will be downsides, there will be shortcomings, there will be like nefarious uses as there have been for literally everything. I don't even think there's an option to like down tools. We basically saying, oh well, I guess we should just give up on advancing society because somebody might use an advancement in the wrong direction. To me that feels like extreme safety ism and it's more likely to have like negative repercussions long term. That's kind of my take on.

Just the other day there was a post on LinkedIn complaining about being charged 8 to 10k for personal tax fees and saying hey, we already did the work, but they keep upping the fees. Lawyers, accountants, they keep the meter running. They don't give you a quote before they're going to move on to the next thing. They talk to you and they charge you and this changes the dynamic. Like you said, just feed it and get the information you want. No more fear, uncertainty and doubt and like who should I ask? Just get the insight you want. Not the piece of data or the piece of paper or the long winded thing explained in the most complicated way to scare you even further. So I love this. Well, how can businesses who don't have extensive technical expertise get started assuming, you.

Know, technically that means you don't have engineers in house or whatever. I think the best thing you can do in the short term is let your employees expense a chat GPT plus type thing so they can actually use this tech and ask them to like identify their own efficiencies in their workflows and their own repetitive tasks. See if they can get a significant value from that alone. And I think you will see that a lot of people question that. A lot of people don't also realize how much repetitive tasks they actually have in their day to day life. I think that's like an easy way to sort of test the water is at the beginning of every case or project or consultants, we always do the following four things. Let's see if we can prime ChatGPT to do it for us. We'll give it like the five ingredients and we'll see if it knows how to do it. That's like an easy way to sort of identify those areas and just get the actual human capital efficiency that's there. Beyond that, because you can't build it yourself, you'll be looking at off the shelf tooling. So if your work involves Creating visuals or mood boards. You might want to use tools like say Kive AI. If your work involves producing branding, you might want to look at kittl like you kind of canva, but that uses generative AI. There are lots of off the shelf software you can pay for that will do a lot of your work. Jasper, if you do a lot of writing, there's a lot of stuff out there that can actually help your employees again. And then I think after, after that you are probably then into. We should assess the feasibility of hiring a development firm to build against GPT to replicate parts of what our workflow is to see how much efficiency we can actually call back. That's the three steps I'd say definitely.

And you talked about leveraging GPT4 in Fin. So you guys had access to the OpenAI platform and leveraged some of their libraries or LLMs to build this stuff. Right. So you need people with expertise to do this as well. And you know what's funny is like when we had access to it, I think 2019 or 20, it's basically it leveraged the Y Combinator network and the network to get all these people and effectively I feel like it's all our collective of our data help build this thing that is now for the greater good in a way that's exciting. What are you working on that you're most excited about? Maybe things about Fin, things that are yet to be introduced that you can share.

FID is number one. We're in the late days of the build, so it'll probably be possibly live by the time this goes out. We're very excited about that. I think genuinely this is the single biggest opportunity we've had to transform the world of customer support. And we tend to take it with both hands now. Like the technical features are building the Inbox import and learning from prior behavior and interpreting questions in the right way and all that. We're building all that at the moment that's like honestly it's at least 80% of our headspace is just fin resourcing, fin bringing fin to market, adding the right features, getting customer feedback, looking at the resolution rates, all that sort of stuff. But there's huge opportunities beyond in areas like proactive support. Even in areas like this. Further work we can do in the human support, the agent assist side that we actually have in private beta at the moment, just to make the actual humans a lot more efficient as well. That's generally how I'm thinking about it. We believe the future support of all three things humans proactive and bots. We're working across all three and we're leveraging the generative AI to improve all three in different ways.

Definitely. Where can we follow you? Where are you most active?

So it's just my name, Des Trainer. D s T R a Y N o R and I'm probably most active on Twitter and then LinkedIn. Second, everywhere else I'm there, but who wants to follow me on Instagram or strap? I don't know. But like I'm there too, you know. So yeah, Des Trainer is basically what we're. And then the company is Intercom, obviously. It's just Intercom everywhere, you know.

I asked Chad GPT to make a witty poem about you and it says this. Here's to Des, the digital king who makes witty remarks, will make you sing. In tech, he's a true visionary. In humor, he's a true legendary. I asked them a bunch and they did a great job. Passion for design and UX sets Intercom apart from the rest. A customer centered approach with no pretense elevated the company to be among the best. Awesome. We'll leave it at that. Thank you so much. That's always a pleasure. This is the third time and it's always been one of the best.

I need some traction.

Thank you for listening and we hope you enjoyed this week's episode of the Traction podcast. If you enjoyed the show, please leave us a five star review and you can find more information and all the resources mentioned in today's episode at Boast AI that's b o a s t a I forward slash blog.