Pod Insights The Twenty Minute VC (20VC): Venture Capital | Startup Funding | The Pitch 20Growth: The Death of Growth Teams? | How Hubspot Use AI to Triple Email Conversion | The Future of AI SEO | Why Prompt Engineering is the New Coding | What Every CMO Needs to Know About AI in 2025 About this episode Kieran Flanagan is the CMO at HubSpot, where he’s led the transformation of their growth strategy from SEO-led to multi-channel and AI-powered. Formerly SVP of Marketing, he helped scale HubSpot’s user base to millions and revenue past $2B. Before HubSpot, he drove breakout growth at Marketo and Salesforce. Kieran is one of the most respected voices in SaaS marketing and a pioneer in growth-driven content strategy. Agenda: 00:03 – The Death of Growth Teams? Kieran's Wild Prediction 06:44 – AI Innovation Pods: The New Org Structure for Startups 10:18 – Email Personalization That Tripled Conversions 13:21 – From Software Budget to Labor Budget: The Shift is Happening 16:35 – The Big Lie: Why Autonomous Agents Still Suck 19:24 – The Secret Sauce Behind HubSpot’s Email AI Stack 21:44 – Segment-Based Marketing Is Dead. Enter Micro Audiences. 24:15 – Content Collapse: Why Google Organic Is Getting Torched 30:52 – The Future of AI SEO: 1 Product, 100 Pages, Infinite Prompts 33:16 – Memory = Moat: Why ChatGPT Is Becoming Unbeatable 35:46 – Prompt Engineering is the New Coding: Here's How to Win 41:03 – The Death of the Middle Manager Marketer 46:17 – OpenAI vs. Anthropic: Kieran’s $400M Bet 48:00 – Europe Is Falling Behind: The Harsh Truth on Regulation 52:39 – CMO Playbook 2025: Micro-Audiences, Creator-Led, AI at Scale
Transcript 16,513 words · 2 speakers
Speaker names marked ? are inferred from speaking time and the show’s known hosts — they are not voice-verified.
0:00 Harry Stebbings? Welcome back to 20 VC with me, Harry Stebbings, and this is a very special episode of 20 Growth as we welcome back one of my favourite growth minds. So for those that don't know, 20 Growth is the monthly show where we sit down with the best growth leaders to unpack how they think, learn, build the best growth organisations today. And rejoining me in the hot seat is Kieran Flanagan. As I said, I think he's one of the greatest minds in growth today. He runs all things growth and especially AI implementation at HubSpot. Today was a very natural, free-flowing conversation that originated from one of his LinkedIn posts, which I loved, and I so appreciate him rejoining me in the hot seat today to really delve into how AI changes the growth landscape as we know it.
0:43 Harry Stebbings? But before we dive into the show today, SecureFrame empowers businesses to build trust with customers by simplifying information security and compliance through AI and automation. Thousands of fast-growing businesses, including NASDAQ, AngelList, Doodle and Coda, trust SecureFrame to expedite their compliance journey for global security and privacy standards such as SOC 2 and ISO 27001, CMMC, NIST standards and more. Backed by top-tier investors and corporations like Google and Kleiner Perkins, the company is among Forbes' list of the top 100 startup employers for 2024, G2's best software awards for highest satisfaction products, and a recipient of the 2024 Cyber Security Excellence Awards, something I definitely never got in school myself.
1:28 Harry Stebbings? Learn more today at SecureFrame.com. Now, let's switch gears to the future of mobile and how one company is turning your screen into real returns. We spend nearly half our waking hours glued to our phones, upwards of 50 hours every week. Recently, one company transforming this reality stood out so much I personally became a shareholder: Mode Mobile. Mode Mobile created the Earn Phone, the smartphone that pays you for daily activities. Instead of big tech profiting billions from our attention, Mode returns over $325 million directly to users through earnings and savings. Mode's revenues surged an incredible 32,481%.
2:10 Harry Stebbings? In three years, recognized by Deloitte as 2023's fastest growing software company in North America. And here's why I'm excited. Mode's equity offerings have raised over $30 million from 20,000 plus retail investors. One of 2025's standout public raises. And you can now join me as a shareholder with as little as $1,000 at invest dot mode mobile dot com forward slash 20 VC for a limited time unlock up to 100% bonus shares and a free earn phone. Email us for the investor brief at 20 vc at mode mobile dot com or check out invest dot mode mobile dot com forward slash 20 VC. We've talked about making money from your phone. Now let's talk about making your product do exactly what you want it to.
2:55 Harry Stebbings? Whether it's the software you build or the software you buy, your tech stack should be creating results, not creating roadblocks. Well, Pendo's no code software experience management platform makes your software better with tools to see where users get stuck, guide them with in-app messaging and constantly improving your UI. It's so easy that over 14,000 businesses use Pendo to increase revenue, lower costs and reduce risks. Businesses love the control, engineers love the freedom, everyone wins. Start for free today at pendo.io forward slash 20 product. You have now arrived at your destination. Kieran, dude, I always love our conversations. So first, thank you so much for joining me today.
3:35 Speaker 01 I'm happy to be back on the show. Thanks for asking me back on.
3:38 Harry Stebbings? Dude, we were just chatting now and you did give like mic drop moments like straight away. And I think shit, we're not recording. And so you said to me just now that you thought in the future, growth teams would become redundant. And I thought, holy shit, what a great start to the show. What do you mean by that? And how do you envision the growth team of the future then?
3:57 Speaker 01 Okay, cue everyone throwing pitches and forks and knives at me here for saying this. So growth teams are primarily like focused on how do we make the go to market more touchless? How do we help onboard people? How do we get people to use the product? How do we get people to buy in a more touchless way? And what I think AI does is extrapolate all that away to just like, how do we make our go to market much? More AI centric, like regardless if you are a touchless business model, you are a human led business model. What I see is AI is starting to collapse all of that into a single function, which is like an AI innovation pod. And the AI innovation pod is trying to figure out, how do we deploy AI across that go to market to make it a much more efficient go to market?
4:39 Speaker 01 And that we are using AI where it's needed to make humans more efficient. And then we are using AI where it's needed to take humans out of that process. The difference for growth is growth traditionally would not work on the human driven aspects. We wouldn't try to figure out, how do I make the sales team more efficient? How do I make the customer success, the customer support team more efficient? How do I deploy AI to do those types of things? It would mostly stick to the product led growth part, which is like the onboard and the activation, the upgrade points. And I think AI changes that a lot. Like I think you need one team who can like switch between both of those things. And I'll give you an example, right? Like people are saying, well, that's just doesn't make any sense. Let me give you a clear example.
5:19 Speaker 01 We had growth forever trying to like figure out onboard and like tool tips and product tours and all of these different things. But what happens when your entire onboard and experience, your buying experience is done through an AI multimodal agent? Right? I go to someone's site, multimodal allows someone to conversate with you via text, audio, they can see your screen, guide you around the product. And that multimodal agent can stay with you throughout the onboard and experience, the upgrade experience. And actually, even when you're a customer, who owns that experience, right? Like what's the handover, what part does growth own, what part does customer success own, what part does customer support own, what part does sales own? To me, an AI innovation pod should be the pod that's trying to integrate that experience across your go to market.
6:00 Harry Stebbings? So have you disbanded the growth team at HubSpot then and created the AI innovation pods?
6:04 Speaker 01 No, we haven't.
6:07 Speaker 01 Which is why I think I'm going to get in trouble for saying this. So we, we have a, a growth team that are very much focused on product led. They do really great work. We have an internal team that is trying to work across AI innovation across like humans, sales, customer success, support. We are starting to try to bring some of these experiences together. So like we have people who are working on different AI chat multimodal components, whether they're touchless. Or whether they actually help serve the, you know, human led part of the go to market. So we haven't integrated them, but I do think forward looking thing in the future, like on the horizon. I can see those things starting to merge. You know, the big part for me is that having one team who could think about the touchless aspect of it, the human aspect of it across your entire go to market
6:50 Speaker 01 will make more sense in an AI driven world versus the way that we've kind of thought about it traditionally is like, hey, like we have a growth team that works on. How do we extract humans out of this loop and make our experience and make the ability to do this touch as much easier? And then have another team trying to figure out how we make all humans much more efficient.
7:05 Harry Stebbings? So if I'm an early stage founder now, starting teams from afresh without any historical baggage or team debt, so to speak. Do I start with an AI innovation pod or do I still start with a growth team and layer on an AI innovation pod?
7:19 Speaker 01 And I think just for like growth teams screaming at me and just saying like, this is typically what we do. I agree. Like, I actually think growth teams could evolve into this AI innovation pod. And if I was talking to founders and I have done a lot of this talking to founders about how they start to deploy AI across their business. For me, you have an AI ops team that can actually help you ingest all of your unstructured and structured data that can provide that foundational layer that you can build AI experiences on. And then like if we even remove names, because I think that's where people really start to get a little bit touchy about like, oh, like I'm a growth team. I own this. I'm on this. I own this. Like if we just remove names, right? If we have, we think about AI and how AI is going to benefit go to markets.
7:58 Speaker 01 The ability to ingest all of this data and for the big part for AI is the unstructured data, like your sales transcripts, your chat logs, like all of these different things we could never really use before. And you have that data layer and you have incredible context about your customers and you have all of these live signals. And then you have a team that sits on top of that, that is able to innovate on how they actually use that data to make using your product much easier, to make extracting value much easier, to make selling your product much easier, to make supporting your product much easier. I think that is where we're going to go in the future and that you have a singular team looking across the entire go to market, whether that is the self-serve part, the sales-led part, the customer success part, the customer support part. Because
8:40 Speaker 01 Again, AI is able to extrapolate away all of these traditional handover points. Oh, now you're with the sales team. Now you're with the customer support team. Now you're with the customer success team. And actually collapse those things and make that experience a little bit more ubiquitous. And I think that's one of the things that I have started to see, not just really within HubSpot, but even outside of HubSpot and the way that other teams are doing what we would have called, quote unquote, traditional growth.
9:03 Harry Stebbings? When you think about like AI GTM tooling and you think about where it's valuable today versus where it's not, and it's not quite meeting the muster, where is AI tooling 100% super valuable and where is it not? Like, you know, we've seen a lot of AI SDRs, very few of them have actually made the grade, for example. How do you think about where it's valuable and where it's not today in GTM?
9:26 Speaker 01 So what we've seen is a ton of value in personalization. People know that use case, but it really does work. And so we saw, we've seen increases of like triple conversion rate uplifts in email. By just being able to personalize that to the person, again, using your structured and unstructured data, having a great data layer that you can actually extract all of that information and put it into email. We are starting to be able to like automate a lot of the traditional prospecting that sales did through email. So this stuff does actually work. And so you would say, well, why haven't some of the, you know, traditional off the shelf tools work for SDR? Because that's like an SDR motion, right? Like the ability to capture data, the ability to do outreach personalization. One of the things I do think that is a struggle for AI tooling, and I'll get into some other use cases,
10:09 Speaker 01 but one of the struggles for AI tooling is that you have to build prompts that will work across all companies, right? So like if I'm an AI SDR tool, I'm trying to build a prompt and a prompt is the way that I'm going to craft the email that will work across hundreds of customers. What we have found is the more that we customize the prompt for our needs and the way that we customize that for our needs, the results are much, much better. And so I do wonder if these kind of AI tools are ever going to be able to personalize a lot of this stuff in a way that the company needs versus the company building some of these things itself, right? And I think that's one of the differences between using off the shelf software and customizing some of it to yourself. We've seen a lot of impact. In that use case, customer support, I think that's a known use case, right?
10:54 Speaker 01 Like, it's probably one of the first ones that a lot of companies start with, that you can actually handle a lot of your customer support with AI. It works pretty, pretty well. We're actually starting to sell through chat as well now. If you kind of move into the future, the team that used to do customer support, they can also sell if you have AI doing both of those things. As long as you have good intent switching, the AI can pick up and signal. It's just a support conversation or a sales conversation. Your customer support team that used to be on chat on your website, used to be in chat in your premium product. If you're starting to integrate AI into that and make AI the kind of first touch to conversate with the user, you can now have your customers chat team, like act like a sales team because they can pick up on those cues. And so that's one of the ones that we have seen work.
11:34 Harry Stebbings? Listen, for me, the big question as an investor is, are we going to see the transition of budget for tooling like this from software budgets to labor budgets? And are we going to eat into labor budgets with software? If so, for me as a venture investor, fantastic news. If not, it's not as exciting as we thought it was. For you as HubSpot, are you willing to pay more or to spend what was labor budget on software tooling in that way?
12:00 Speaker 01 AI allows companies to do two things. You can be much more efficient, so you can do some of the things you used to do with less humans. Or you can actually be much more growth orientated and say, well, we're going to like just do more things with the humans that we have, right? So now, a good example of that is I can do a lot of the low end or non-complex customer support with AI. And I can say I can eat, I can just take that cost upfront. I can actually do that with less humans and I can save that capital. Or I can use that to deploy my customer support team and to do more white glove onboard and for our premier accounts, right? Same thing with the kind of BDRs, right? Like, I can get more coverage by integrating AI across my BDR team. BDRs can put the same amount of meetings with less amount of people. Or do I just want to take that capital and deploy it and generate many more meetings, right?
12:45 Speaker 01 Do I want to grow much faster? We are still in the phase of piloting a lot of this AI software and we don't really look at it as in a cost of headcount. We don't say like, we can either have this tool or we can have this many headcounts.
13:00 Harry Stebbings? You reducing your customer support teams?
13:02 Speaker 01 We are redeploying those teams to do other important work.
13:06 Harry Stebbings? Got you.
13:06 Speaker 01 Because you can move your humans up the value chain. I think that's one of the things that you can make a choice to do is that you can say, well, we do no longer need people to do this work. So we can move them into work that we think is higher value for whatever part of your go to market you want to do that for.
13:25 Harry Stebbings? Do you not buy that functions will be smaller in the future then?
13:28 Speaker 01 Uh, yes, I think every function is going to be smaller in the future. I think that's a hard thing to even like, like you're in this space a lot. People are really concerned about their jobs. I try to figure out the most optimistic view of this for folks. Like, what is the way that you can talk about AI and not come back to the fact that if we are being really honest with each other, teams are going to be much, much smaller.
13:48 Harry Stebbings? I think a really good example is like for our media teams, like, you know, the clips that we make from shows like this are bluntly just to make, pretty rote and pretty like the same, actually. Everyone sees these clips on social and it takes days and days, whatever. And actually you don't have the chance to be way more creative in storytelling around how to promote the show, telling Kieran's story ahead of time. You're allowed to do so much more creativity when you don't have to do the process like content creation side.
14:14 Speaker 01 Would you then say you, you like, you take the AI savings there so the AI can do some of the mundane work? And then you have the humans do like you have your team do.
14:21 Harry Stebbings? Way more creative storytelling stuff. Yeah.
14:23 Speaker 01 That's, that comes back to the thing I was like, it's going to become down to a company's choice and how they, because when I think about AI, AI gives you time back. It depends how you as a company want to redeploy that time. Some companies will make the choice to redeploy that time by just having better CAC, right? Like we're having better unit economics and just doing the same amount of things with less amount of people. I think a lot of smart companies will redeploy that time to figure out how they can actually have their team do higher value work. And I think your example is a good one where there's a lot of AI content tools. These AI content tools are not good if you just use them to create content and publish it. They are really good if you want to have a first version of something and then have the person spend way more time on trying to figure out how to craft that into something that's unique.
15:05 Harry Stebbings? But just so I get like a yes or no here, just because I'm intrigued as an investor. Do you think we will see AI tools make the transition from software budgets to human labor budgets?
15:15 Speaker 01 Yes, I think we're going to price agents against human labor costs, outcomes.
15:20 Harry Stebbings? What's the biggest BS or hype in AI GCM tools? On the flip side, where we mentioned real value there in personalization, in customer support. Where is the reality not quite what perception is?
15:34 Speaker 01 Autonomous multitask agents. First of all, I think sometimes people are not sure what an agent actually is. Like, an agent needs to have data, tools, context. We've seen a lot of hype around agents this year. And all I can talk about is the kind of external companies that I've spent time with and how they're deploying AI. And even internally, how we're kind of using agents. I think there are some good use cases for agents. But I do think that in a lot of cases today, AI is still not reliable enough. Where you can guarantee an autonomous agent can do multitask goal in a reliable way. Where it will kind of give you the same thing back each and every time. I don't know if you've you've spent much time playing with some of these agents, but they are inconsistent.
16:17 Speaker 01 They are unreliable. They do not always work. And the models are getting much better. Like, I think Claude, which for me is trying to be the backbone of agents. I think that's one of the things they want to do. Their OPIS has obviously seemed to have made some improvements. Like, if you look at the coding agents. But in terms of go to market, I don't think that I have spent or seen a lot of great agent first companies. Be as reliable and consistent as they need to be.
16:42 Harry Stebbings? Can I ask you, when you look at what you've tried and deployed within HubSpot that's failed? On the flip side of the customer support doing well, personalization, you said the kind of three digit improvement in conversion. Amazing. On the flip side of what you've tried that hasn't worked internally. What has that been on an AI GTM tooling perspective? And what did you learn?
17:01 Speaker 01 So I think one of the things we've learned is, this is so mundane, but it is so true. The better your data, the better your results, right? I think what we have learned is like, the more you can combine data together from different sources, internal and external. The better results you're going to have each and every time with the any kind of AI initiative you're doing. And so our ability to train. Our AI chat agents, right? We have an AI chat agent that is able to sell now. It's booking meetings, it's doing sales. When I look at our path there, our ability to refine, to train that agent using the data we were capturing is when we saw like leaps and bounds of improvements.
17:43 Speaker 01 Same thing with customer support agent, like the more you eat up the, what we call deflection rate, which is like people come to your support agent, conversating with the agent, you can solve that problem. The more complex they get, those problems, the more you actually need to incorporate lots of different data. We had one big increase in an experiment we ran where we were able to put in all of the development docs and developer docs and all the external information around how you kind of do the nuts and bolts of development HubSpot. And the quicker the agent can kind of learn, the better the data sources. In personalization, it's been incredible, like how every kind of new data source we find that actually enhances our ability to talk to that person via AI has just seen huge spikes in increases in conversion rate.
18:26 Speaker 01 And if you had told me 12 months ago that we could use AI to like 5X the conversion rate on email, because I'm like, oh, like there's only so much personalization you can do in an email.
18:35 Harry Stebbings? How do you literally do that? For founders that are listening going, wow, that sounds fucking great. How do I do that? How do you literally do that? Who do you use? How do you structure it?
18:43 Speaker 01 I think there's a couple of tips that we have found, which is if you divide a single email into a collection of different prompts, and so you have a prompt for the subject line, you have a prompt for the intro, you have a prompt for the introduction, you have a prompt for the meat of the email, you have the prompt for the last part of the email. You can iterate much better on the separate parts of the email. And so one of the things that I'll tell you is traditionally, when you look at email and the results of email performance metrics, you'll look at like open rate, you'll look at click through rate, conversion rate into something. And that's a kind of like traditional metrics in the way we look at things. Where AI is able to shift things is you probably get inundated with outbound emails all of the time, right? And there's probably like 1% of those emails that you actually do react to
19:27 Speaker 01 and you actually do something with are a small amount. And when you actually extrapolate why that is, it's because there's a certain style that will work for. You and that style is actually like how someone conversates with you. And what AI is incredible at doing, and I think people have still not captured caught on to this. And this is like the been the kind of thing I've been obsessed with for like eight to 10 months. Is AI is able to extrapolate styles. So it's able to say, hey, like, there's a certain style that makes up this email. And I can replicate that style for you. And so when you look at email performance metrics in the future.
20:01 Harry Stebbings? How homogenous is style to conversion, though? Like, is what I like and what converts me homogeneous with actually a lot of other people? Or are people very individualistic in their styles?
20:12 Speaker 01 I think people are more individualistic than we think. And we typically have done segment level personalization, right? So like, what have we done in B two B? We collate lots of people together and we say you have similar characteristics. Now, what similar characteristics really are we talking about? Job title, seniority and industry, right? It's not actually nothing to do with you as a person. It's everything to do with the data that we have that we can map you into a segment. And the reason we do segment level marketing is because we need a broad enough group of people to go after to make it economically feasible for us to have a certain size of marketing team to market to you, right? We can't market to like 300 people who have like very, very fine. We can get to like the finer details of really who you are.
20:53 Speaker 01 We can't market to 300 people because it would take too many humans to actually build a marketing team to be able to do that. But AI is actually able to do that. So like in your scenario, you would say, okay, like Harry has reacted to like three to five of these different emails. And every time I write an email, which is like the first thing I told you, like you, you have these different prompts. You create a first version of that email. You can actually send that email through a style generator, generate a style for you. And then actually on the other end, send you that email. And so that to me is how AI can change things because now I can actually enable my team to market to far smaller groups of people who I can really decipher what are their styles and characteristics that will make them react to something. And so again, the data really matters.
21:36 Speaker 01 It really matters how you actually think through things in a very like process oriented way. And understand like what AI is enabling us to do, which I think what AI is going to enable. What to do is tailor our experiences in a much different way.
21:50 Harry Stebbings? We both understand economics. When you massively increase supply or make supply infinite, the price goes down and you would expect conversion to go down as well. Do we lose efficiency or utility value of outbound email when supply with AI is literally infinite?
22:08 Speaker 01 We have data on this, right? So like I've seen data to say it would takes many more emails now to book the same amount of meetings.
22:15 Harry Stebbings? How many more just like?
22:16 Speaker 01 I would say like rough ballpark, like three to five X.
22:19 Harry Stebbings? Three to five X.
22:21 Speaker 01 More email because to your point, what's AI enabling people to do? It's enabling everyone to do like pretty good email. All right. That's the challenge with AIs. It takes people from like not very good at something. And you had leverage because you were very good and it takes everyone up to like still pretty good. Now, there's still always going to be exceptional. There's always going to be people who understand more how to use this technology more creatively. So does outbound get much, much harder? Yes, but how you use data creatively will can still separate you from most other folks. Because most other folks will take what AI gives them, do the same thing as everyone else and say, well, that's good enough. And I think coming back to the point we were talking about, your media team is like, there's going to be in every single discipline in terms of go to market 20% that really matters.
23:06 Speaker 01 And it's like, can you use AI to do most of the other 80%? And have the exceptional humans figure out how to make the 20% much better, much different, much more unique.
23:16 Harry Stebbings? I agree. Sadly, what worries me is like when you have such an influx of supply, even if you're truly graded it, you get lost in a world of just infinite supply. And that becomes a challenging problem. It's a bit like venture. If you're a boutique great provider, you can still lose when the world of mega, mega AUM machines.
23:33 Speaker 01 So I think there's a lot of unknowns with how supply is going to increase. Like even one of the things that I think a lot about is like Anton, the founder of Loveable tweeted, I think two days ago, that last month, 10% of all websites and apps got built on Loveable. Like 10% and a, what was it? 5% and 10% in a single month. What actually is more scary is that there's an infinite amount of software now and there is a decrease. Recent amount of distribution channels to actually market that software through. We're seeing a reduction in search, which I know we're going to get into the ability to like rank for that on Google. We're seeing increased paid CAC prices. So actually you have like one chart that's going off the scale. It's in terms of like exponential growth and you have another chart, which is like my ability to reach and market these people. To show them my software is like actually declining over time.
24:19 Speaker 01 That's a pretty troubling trend as well for people who want to like grow their business, grow a product online.
24:25 Harry Stebbings? Dude, I have to go to that. You actually cited it in a LinkedIn post that I loved and you actually cited CloudFlare CEO. You shared amazing but brutal stats and it was kind of called the content collapse, I think. I wanted to talk about this. What is the content collapse? And what does it mean for organic content in your eyes?
24:41 Speaker 01 Yeah. So I'll share the stats because I thought the stats were some of the best stats I've read that showed actually what was happening. So the CloudFlare CEO, two things he talked about, he had this data point, which was scraped to visit ratio. And script to visit ratio is a really good way to think about this, which is how much data do these companies take? And then how much visits they give you back? So the open web has kind of been built on Google. And if you look at the data, I think it's around 63 to 67% of all referral traffic across websites comes from Google. So Google like is still kind of the open web. A lot of it used to be through social media companies as well and social network platforms. But they have reduced the amount of traffic they send elsewhere. They're trying to keep people within their own platforms. Now what's happening is, you know, Google had this relationship with publishers.
25:24 Speaker 01 I'll take your content, we'll take your content and we'll give you visits back. And that relationship is not holding anymore because 10 years ago, Google would scrape two pieces of content and give you a visit back. That's a pretty good relationship. Most people were happy with that. Six months ago, which things have been changing pretty rapidly, that was three X. So with six pieces of content they scraped, they give you one visit back. Now it's like 18 pages they scrape and they give you one visit back. And the reason for that is because of AI overviews. AI overviews is it's going to become Google's default way for you to like get your answers. So anyone who's been on Google, ask a question and the little AI box comes up and tries to answer that question for you. That's on the over like 35% of all searches today have some sort of AI overviews.
26:05 Speaker 01 And so that's rolling that much more rapidly. Their last conference, Google said that they were going to make AI mode, which is really this, the default search experience. They've better tested that in the US. Now they're better testing that in India. And so where will we end up? Well, let's look at ChatGPT. ChatGPT is a dominant AI assistant. It's the way most people would actually use AI to get search results. I think it has eighty percent of all market share for in the AI assistant space. Six months ago, it was two hundred and fifty pieces of content scraped and to give you one visit back. Today, it's one thousand five hundred pieces of content scraped and to give you one visit back. And there's a big reason for this that I had called out years ago when I first saw ChatGPT is when I started talking about the fact that this would replace Google really early, after I had really used ChatGPT for the first time,
26:50 Speaker 01 because I always will say that users will default to fast and easy over everything. Now, people would say to me, no, because it hallucinates and people will not trust it. And I was like, no, they will eventually just default to easy and fast and that's all they'll care about.
27:05 Harry Stebbings? I don't think it gets more and more accurate with the
27:07 Speaker 01 Yeah, and it gets more accurate. ChatGPT is a good indication of where we're going, right? Like, we're going to go to a world where you can get all of the answers you need without ever having to click on any of these websites. Now, what happens to the internet because of that? I'm not sure. It's actually a hard thing to think through because if the incentive to publish content disappears, why do people publish content? And that's what the CEO of Cloudflare was saying, that they did launch a product that allowed people to opt in and monetize their content if AI bots were going to crawl it. But I do think there's going to have to be a new sort of relationship between publishers and these AI assistants to make it worth actually even publishing content.
27:44 Harry Stebbings? Well, my question to you is this, and HubSpot is a business that was built on organic content to a large part. What the fuck do you do then? In the nicest way possible. I don't mean that badly, but like with that ratio, holy shit, me and Jason at SaaS to talk about this a lot, like, what the fuck?
27:59 Speaker 01 I still think there's going to be a world where you have to market your business. You know, the last change, and we can argue if it was a bigger change or not, but was the move from offline onto online was like huge change. I'm sure there's a lot of things that people had to figure out as we made that transition. There's going to be a world Where you still have to market your business. Actually, I think marketing and go to market becomes way more important because you have what differs software anymore? Not really anything because software, the cost of code is so much cheaper. People are building so much more product. How do you differentiate? I think you differentiate with how you market and you go to market with your product.
28:33 Harry Stebbings? Do you actually believe in that commoditization of software? I actually don't. When you think about speed of design updates, you know, product detail, like actually we're just gonna have a load more shit software. But for truly great software, I don't think that's like commoditization of software.
28:47 Speaker 01 I think that is true today, but when I look back and I think about the conversations when chat G P T first come out and people kind of talked about, hey, like, it's never going to be good enough to replace the blue links. What I think about is like, where do we end up in two to three years?
29:01 Harry Stebbings? Okay, but there's a bit like this. HubSpot is a very deep product with lots of different elements of the product. Do you think that HubSpot product is going to be commoditized?
29:10 Speaker 01 I do agree that there's going to be strength in platforms, strength in the data, strength in the ecosystem. Like you actually can build still defensible software. I think what I, I guess what I would say is if you are a single point solution, it's a challenge to be in that space. But how many like true platforms are there versus single point solutions? And then also, did all platforms not start as like a single point solution at some point? Like, I still think it makes software many categories, the difference between one product and a not, like very hard spot.
29:38 Harry Stebbings? So you don't then today, when you look at this, you don't say at HubSpot, shit, we should not do as much organic content. Or how does it change the organic content strategy?
29:49 Speaker 01 You should still have a S E O strategy. There's still like value in transactional search. It has not disappeared. It has decreased, but it does not disappear. So we split out search into informational, which is like, I used to create content to teach you something. And if I taught you that thing, you would come to my website and I could convince you to buy something. Then you have transactional, which is you're already looking for a product that solves a problem and we can appear in Google for that kind of problem. Now informational content, I just don't see value in it long term anymore because people's consumer habits are going to change to just get all of that from A I. And that is a challenge for most businesses. There is opportunity to like still appear in the A I. Search assistance for that content. And for transactional, which we'll spend a little more time on, which is, hey, I have a product.
30:32 Speaker 01 And yeah, people are still going to Google, but now a lot of them are going to these AI assistants. How do I make sure that when ChatGPT recommends products, I'm in that recommended list? I think there's a lot of value in that, right? Like ChatGPT has over a billion users, they're growing much faster. They're going to add a freemium paid product, which I think is going to help us market within ChatGPT. So the thing is, if you have a single product, let's say I have a product that did, you know, a sales tool. In Google, you would try to rank that one page. You would say, okay, I'm going to try to rank that one page for like a couple of keywords. The difference in AI search optimization is you conversate with the AI. So now instead of three ways of asking for a product that fits your product, like Google taught us to search in keywords. And so the average page may have like three ways that people would search for that specific products.
31:17 Speaker 01 We're taught to search in keywords. AI assistant is different because you conversate with the AI, you have a real conversation with it. And so actually people could ask questions about your product in hundreds of ways. And actually, what you need to do for AI search optimization is have hundreds of like micro versions of your product page, tailored for all of these different audiences.
31:36 Harry Stebbings? Does that not make the ads business for OpenAI a hundred times better than the ads business for Google then? Because they can monetize all the different micro segments and the micro conversations that you could potentially have.
31:48 Speaker 01 They have memory. I think people like, even when we're trying to like think ahead of, there's a lot of tools now trying to build, like who's the SEM rush for ChatGPT? Who's the equivalent of Ahrefs for ChatGPT?
31:59 Harry Stebbings? He got AI. There you go.
32:00 Speaker 01 There's a lot of great companies. I'm interested in if you spend time with them is like, the hard thing for all of us is like, when you had Google, you could decipher a lot of analysis about how you were doing on Google. The challenge I think is going to be for ChatGPT and why their paid product is going to be so good is because of memory. So your results over time for ChatGPT are going to be so much different than my results. Because I'm going to spend 10 hours a day in there. It has all of the memory. So when I say, what's the best product for X? It's going to say, based upon the history I have about you, which is a lot, your results are this. And it's going to say for you, your results. Or that. And so how does that collate all of that information together?
32:38 Harry Stebbings? This is why I always laugh when I hear about the commoditization of like LLM providers. Cause I'm like, no, if I suddenly move to DeepSeek, it doesn't have any of the memory. Or if I move to Claude, it's a completely different product.
32:49 Speaker 01 Are you saying memory is a network effect? Like memory makes them more defensible?
32:53 Harry Stebbings? Incredible retentive mechanism.
32:55 Speaker 01 Hundred percent.
32:56 Harry Stebbings? One thousand percent better. Completely agree.
32:57 Speaker 01 I think it's the, it's the most defensible part of an LLM. I'm a power user of all of these apps, but I was definitely like in Claude a lot. Cause I felt the Claude latest models at the time were like more creative for some of the marketing tasks I did. O three really changed my behavior. I moved into ChatGPT a lot. I use it every single minute or every single day. I find it really hard to leave ChatGPT because of memory. Like it's just so fine tuned to who I am and what I want to do.
33:22 Harry Stebbings? Do you use Claude at all anymore?
33:24 Speaker 01 I still use Claude. Yeah. I still find a lot of their use cases pretty good. They, we have enterprise seats with most of the LLM providers, but I think a lot of their enterprise use cases are really good, actually.
33:33 Harry Stebbings? Do you think we will use prompting in the same way in two to three years as we do today?
33:39 Speaker 01 Oh, that's a great question. Okay. Prompting is like my thing right now. That's all I do. It's all I post about. It's like the thing I've been obsessed by. And so I hope that it's still going to be a point of leverage for me. I will say like the number one tip I can just give anyone to get better at prompting is ask the AI to create the prompt. Still blows my mind that people don't really do that.
34:00 Harry Stebbings? So why do you recommend that so much and what does it mean?
34:02 Speaker 01 Because it's so good. It's just so good. Now, if you actually have a first conversation with AI, any of them, like I used Claude or ChatGPT, if you have a conversation, you explain the outcome you're trying to get to. If you can actually show the outcome, the best thing to do with AI is actually say, here's an example of this being done before. And here's the exact outcome. And that's the kind of outcome I'm trying to get to. Create me a prompt to get this kind of outcome. It will do a first good draft. I'm pretty deep in prompting, so I don't, wouldn't use the very first draft. But it can get you very, very far. And I'll just give one more tip here. The other thing that is going to really rapidly improve your ability to prompt is you can take these prompting guides. So OpenAI has a great guide. I found it on Reddit. Was had a prompt for the O models, right?
34:47 Speaker 01 And it's this like huge document. It's like eighty pages. And so what I would do with something like that is I'll take that, I'll create a custom GPT. I'll give my custom GPT instructions to say, using this guide, craft me the right prompt, using the right tools to get to my outcome. And then because it's using that guide, it can create a pretty incredible like first draft of a prompt. I'm pretty bullish on prompting as being a way to differentiate your results from everyone else. Where we will be in two to three years, I don't know. Like the models might just be self prompting, self fine tuning, self doing everything.
35:20 Harry Stebbings? Do you train team members on how to prompt?
35:23 Speaker 01 Yes, yeah, like we have made a big investment on how we can enable every employee in HubSpot to be AI first. Like on the marketing team, we have a ton of training materials, knowledge, sharing, all of these different things to try to get people to understand the importance of prompting, all of these kind of AI first disciplines they need to learn.
35:42 Harry Stebbings? How much marketing content is AI generated today?
35:46 Speaker 01 I think on LinkedIn, probably ninety percent.
35:48 Harry Stebbings? Seriously?
35:49 Speaker 01 No, I don't know. Um, I think that it's still easy to identify, like even outside of the M dashes, which is the big funny thing online, like if for people listening, if you see something with an M dash, it's kind of annoying now because I used to use M dashes in my write-in and now I can't use them anymore because people just think it's AI.
36:05 Harry Stebbings? Well, that's an interesting point. Does AI generated content inherently have less value?
36:09 Speaker 01 I would say that if you look at social media avatars who are like AI first, no. People are pretty happy if the content is good enough. They don't care.
36:17 Harry Stebbings? This is my point. What is value in terms of a sort of like a distribution channel, retention and usage of like content? So like, do you stay and watch the video? I stay and watch AI videos just in the same way I would watch normal videos. So to the, and the provider, it is the same.
36:32 Speaker 01 I agree. I don't think the human cares as long as the value is. Did you see the dental ad from VO three, the guy? So there's a dental practitioner in LA. I just got access to VO three today, which is great. I know in the UK they have access for a while. So VO three is Google's latest video model. For me, it's like, if you think about the evolution of AI here, ChatGPT was like the big moment for text. ChatGPT's image N tool was the big moment for images. You could go text to image. Pretty good. And then VO3 is Google's latest video model. You can go text the video and it's very, very good. And there's a dental practitioner in the LA who created this, kind of looked like a vlog of a gorilla, doing a parachute jump, losing his teeth and then going to his practice to get them fixed. Then it went viral. And he apparently booked up his dental practice for years and years on years on end.
37:14 Speaker 01 Now, everyone knew that was generated by VO3 and no one cared. I certainly didn't care because it was a great, great ad. And I love that for two reasons, which is ideas are being democratized by AI, which is the thing I am most passionate about AI for is that great ideas person. It's never been a better time to be an ideas person, right? But it also does show you that people don't care. Good is good. And if you have great taste and you can make the AI generate that taste, then we can maybe talk about can AI really generate tasteful content? I don't think people care.
37:42 Harry Stebbings? Dude, I have so many things that I want to unpack from all of this discussion. We mentioned like the analytics side of seeing where you rank in traditional search. For you as a HubSpot now and as a buyer, are you going to spend a shit ton of money on analytics for LLMs?
37:58 Speaker 01 We use an X funnel to like, start to see how we are tracking against LLMs. Our search team here has transitioned pretty heavily to be a LLM experimentation team and try.
38:07 Harry Stebbings? How much traffic comes from LLMs today? I remember like Vassal posted not too long ago, they've seen a three X increase from a search to LLMs.
38:15 Speaker 01 Yeah, we've seen a five X increase, but it's still less than 1% of our traffic.
38:20 Harry Stebbings? Wow.
38:21 Speaker 01 Because all you can see is referral traffic. The big change is, and it's a change we have to get comfortable with is, share of voice is a more important metric for LLM optimization than traffic.
38:32 Harry Stebbings? What do you mean by that? Sorry.
38:34 Speaker 01 Let's go back to the stat that 1,500 pieces of content get scraped and you send one visitor. And so is there any value in the content being there at all if you get very little amount of visits? Yes, because you have the biggest platform to show your brand. And so it's just like an ad impression.
38:50 Harry Stebbings? But do you actually show your brand if it's like in an article that actually has five other bits of blogs related to it? And there's just like a link there that says hubspot.com.
38:59 Speaker 01 That's for informational search and I agree that there's little impact on being in there if you're just like, I'm. Another blog post that teaching you how to create Facebook ads, or I have another blog post that's teaching you how to like deploy capital and venture. Because I actually think what's going to happen there is the human used to go to Google and consume informational content and learn that and then go and take action. What AI fundamentally changes is AI is taking the informational content and training on that. What the human is asking for is usually, they're not going to ask for, how do I do this? They're going to ask, do this for me. I think that informational layer changes into an action layer. All that informational content is used by AI and people learn over time that I'm just going to go in there and ask the AI to do that for me.
39:41 Speaker 01 Where there is value in being showing up in the LLMs and OpenAI is actually showing up when there's a question, that's a transactional question, a question specifically about, I want to do this thing, what are the five best products? Because then you are actually one of those five best products and that is valuable. And I think that's the thing to really obsess over is those transactional questions. When someone is in the buy-in mode, you want to make sure that your brand appears in those answers.
40:07 Harry Stebbings? The question is, do people even just skip that deliberation process? And say, I want to buy a cattle, buy the best one for me.
40:15 Speaker 01 I think in B to C, yes, like, you know, we've seen like perplexity in some of these other companies start to deploy agents. I know Google is doing that as well. I go through our chat logs all the time for a starter, for our pro, like analyzing those with LLMs with the team. And I can tell you that people have a lot of questions, even in the smaller products, even the products that cost a less amount of money, that I still think we are a ways away from them saying, sure, that would be fine, just go buy HubSpot, right?
40:44 Harry Stebbings? So do you think we fundamentally changed the nature of the content that we create to prioritize LLM-based SEO from traditional SEO with Google?
40:55 Speaker 01 I think the example I give, which is you used to have one product page for your segment level audience. I think you will have 100 product pages tailored to individual audiences. And who knows, like in the future, you might have a product page per company.
41:08 Harry Stebbings? Do you think you'll have a $10 billion LLM analytics company? You know, if you mentioned SEMrush, I love PKI, we just did the seed round. A reason you would not do that deal is SEMrush is 17 years old and worth a billion five.
41:22 Speaker 01 The thing I keep getting stuck on is how do you think about memory? How do how do those companies overcome the personalization of results? Because what they do is they basically will do the you'll add your questions probably. And then they'll take those questions and then they'll poll an instance of chat GPT and tell you if you appeared or not every day. But how do they figure out did it appear for you, me, someone else? That's the part that I struggle with.
41:45 Harry Stebbings? So I think there's kind of three things when you're seed investing, which is like number one team, number two, directionally correct and number three deal. I am not smart enough to predict memory and how it impacts the business model of OpenAI in two to three years. And I know this is a cop out answer completely. I just look at Peak as a seed company and go, holy shit, amazing team.
42:07 Speaker 01 They're smart.
42:07 Harry Stebbings? Directionally correct a hundred percent. Pull is insane. Our revenue scaling is almost unlike anything I've ever seen other than lovable. And the deal was expansive, but not crazy.
42:18 Speaker 01 Yeah.
42:19 Harry Stebbings? So I'm still fortunate enough to be early.
42:21 Speaker 01 I would still do the deal because I think if I was looking to that, I would say. Because the only way that problem gets solved if OpenAI is incentivized to do something that allows companies to poll it in ways where they can get better data. Now, why would OpenAI ever do that? If they want to do freemium paid and they just hired the ex CEO of Instacart. So it's pretty, I think, pretty plausible that they will do a paid tier. They will need to have some of this functionality to allow people to optimize their pay towards audiences that they care about. And if they do that, that is a great thing for all of these companies who also want to be the SEM rusher LLMs. Because they'll have some way of actually saying, well, our audience is this. And based upon that audience, how many times do we appear for these different queries?
43:01 Harry Stebbings? I literally noted down because we had this like discussion thread that I had so many different bits that I want to pull off on. So if you were Google then, what would you do? You've got your golden goose that's at threat.
43:11 Speaker 01 I think they have decided this year that there is no slow rollout, right? I actually think the past 12 months, they've been trying to figure out, can we do a slow rollout here? And they're seeing consumer behavior change pretty rapidly and that they are seeing more and more people gravitate towards like OpenAI and some of the AI assistants. Now, if you actually look at the data, are we seeing that AI? Is mean means people are searching less. No, actually, AI is causing people to search more. So search volume is going up and we would say, well, that's great then. Why are businesses not benefiting from that? Why that is that good for businesses? Is that good for Google?
43:47 Harry Stebbings? When you say search more, do you mean search? How do we define search? That is conversation.
43:52 Speaker 01 Yeah, asking questions. Yeah. The other interesting data point that was the CEO of Cloudera shared was like 2 billion more people have entered the internet. So it should actually be like, if you actually added those things together, it should be the most incredible time for businesses. Again, it comes back to what's actually happening. Well, people don't need to go to the website because AI is really good at answering that question. That doesn't mean that Google can't figure out a ads model that works really well for that. And it should actually allow them to create an ads model to your point for OpenAI. That is much more tailored to the individual. And I still think that they have a history and a skill set of doing that. So to me, I would deploy this stuff way more rapidly. And I would obsess over figuring out what that ads model looks like.
44:36 Speaker 01 That actually integrates into that conversation in a way where it's not working right now. I don't think these ad experiences are very, very good. But the challenge for Google is there's no reason OpenAI can't create a better or an equivalent model. Like they've lost their dominance because page rank no longer really matters. And that was really the thing they built their business on. So I think they have to go all in, deploy it really rapidly, figure out the paid model. And I think what they're trying to do is like go back to their roots and ship fast.
45:03 Harry Stebbings? Yes, hard to do when you're at that scale and size.
45:06 Speaker 01 Hard to do when you're at that scale.
45:07 Harry Stebbings? Totally agree. You can invest in OpenAI at 300 or Anthropic at 60. Which one do you choose?
45:12 Speaker 01 I feel bad about saying that's OpenAI. The opportunities for them to disrupt whole industries is just unrivaled.
45:19 Harry Stebbings? If I could do anything without fear of retribution from my LPs, I'd put my 400 million fund into OpenAI.
45:25 Speaker 01 Yeah.
45:28 Speaker 01 I think they all like create huge billion dollar revenue streams for themselves by accident. You know what I mean? Like they were like, oh yeah, we do all of this now. We didn't even realize that. Let's just spin that out into a business. Like the opportunity is so vast. Obviously they have Zuck taking all of their talent. But like if they can keep their noses in front, then.
45:45 Harry Stebbings? How do you analyze Zuck taking all of their talent? I look at this now and I'm like, does this make me inherently more bullish on Meta moving forward? Or do I go, wow, that's a last ditch attempt at really trying to be one of the AI players when they're not really yet?
46:00 Speaker 01 Yeah, I was talking to someone on WhatsApp this morning and I was trying to figure out what is their, you know, I could reel off the other companies, you know, OpenAI wants to be the kind of personal assistant via AGI to every human on the planet. Claude wants to be the backend for AI and allow companies to build on their infrastructure, be the infrastructure for autonomous agents. Google obviously just want to deploy AI across their ecosystem and stay ahead. I guess Meta's ultimate goal is probably very similar to OpenAI, which is a personal assistant for that billion users. I am a fan of Zuck. I actually like how aggressive he is. I think this is like baller stuff. It does seem like it's a bet it all to win aggressively. The list of people he's got is pretty stunning.
46:39 Speaker 01 Does it make me more bullish in Meta? I think I'm pretty bullish on Meta. I'm pretty bullish on Zuck as a CEO. Now they're kind of like metaverse. You know, he has kind of like gone into some industries and it's not played out, but if you were going to bet on someone, then with the list of people he's just hired, I think it's a pretty good bet to make. I still don't put anyone ahead of OpenAI.
46:58 Harry Stebbings? We mentioned actually like the value of prompts and how that changes moving forward. In terms of model selection, you can now choose, obviously you've always been able to choose which model you want to like deploy your prompt against. Do you think we will have the option to choose which model or do you think that will go over time?
47:13 Speaker 01 I think this feature is even in Microsoft's Azure that it will actually choose models for you based upon the prompt. I actually tried to invest in the company that was going to build us into SaaS, which is like, there's a layer and it will switch between models based upon what the task you're trying to do. The reason you would do this is because there are some tasks that people in companies will probably use reasoning models for, like O3, things like this. It would be very expensive to run those prompts and what you're trying to. To do is probably doable through a cheap open source model. One hundred percent. I think that most companies at some point will have the ability to like switch between open source models, paid models, whatever model is the best model for their task, because it's going to help them optimize their costs.
47:53 Harry Stebbings? You also mentioned using O three and like the rollout into Ireland. EU AI Act has been really prohibitive. Do you think Europe is shooting itself in the foot with the regulatory stance that we're taking around AI?
48:06 Speaker 01 I hate everything about the EU's stance on most of these things. I'm getting VO three now, when that has been out for like three months. We are in a race where I think people who are knowledge workers who are enabled with AI just have such a distinct advantage. And if Europe want to be a hub for excellence and innovation, the rules and regulations we have are forcing founders to still, I think, gravitate towards the US. You would know much better than me. I talked to a great founder, two great founders, over the past month in Europe, and they want to stay in Europe and they want to build great companies in Europe. And they're still contemplating moving because they believe this is just going to slow them down and they're going to be at disadvantage with their competitors in the US.
48:50 Harry Stebbings? Are you not professionally prohibiting yourself by still being in Ireland respectfully? If you're only just getting access to those models now, you could have had access to them three months ago and been three months ahead of where you are today.
49:01 Speaker 01 Yes. Yeah, I think I'm trading off like the ability to do much more, to have access to the best tools on the planet to like live where I live. I think if I was earlier in my career, I would really think about that. I would really think it's like this continent, the place to build your career. I'm still pretty hopeful. Like, there's a ton of great companies being built in Europe. Like, if you go to, if you look at Sweden alone and some of the AI companies being built out of that country, like, there's just so much good stuff happening. And so I do hope that we will see some sense in terms of regulation allowing companies to go much faster here.
49:32 Harry Stebbings? Can I ask you, in terms of AI-driven content or AI-created content, how are you thinking about investing in that today at HubSpot? And how does that compare to your budget for human-created content?
49:42 Speaker 01 I think we're changing where we think the value accrues. And so I talked about the search team. They have a like really Pivoted, they're still, we still do search, it still makes up a big part of our number. We have that team really pivoted to work on AI optimization experiments and we have seen a lot of increases in our awareness or share of voice in L L Ms because of that. I do think AI is a great assistant across media, like you're pretty well skilled in media, you have a media team, you're probably seeing the ability for AI to start to do some first drafts of video, first drafts of scripts, first drafts of like be a good partner in how it can create good media. Do I think AI can create really tasteful content, maybe in some respects, but most of the time not.
50:20 Speaker 01 You still need a human to really inject taste, personality, point of view in a lot of content you create. So we have AI enabled across our media team, like we had bought a media company some years ago called The Hustle. We have a great number of people from that company, but like a lot of great media properties.
50:35 Harry Stebbings? What tools are you using most significantly when it comes to content creation, editing, distribution?
50:43 Speaker 01 We're still using mostly the core L L Ms for a lot of the content creation, customized through prompts. We have started to use like captions, we use HeyGen. So captions is a really great tool. They just released Mirage, which allows you to create really great short form video. We started to play around with that tool. HeyGen's latest models, I'm not sure if you've played with them, they're like pretty incredible. So we started to do a lot of things with avatars. What that's really helping us to do is expand our YouTube footprint in non-English countries much, much quicker and it's working really well. So we have like AI avatars doing our videos. Or we have like experiment with AI avatars in Spanish. We've experimented with AI doing the dubbing in different countries for the videos that we have. That's working really well.
51:24 Harry Stebbings? The other thing is like with the cost reduction, the experimentation ease is so much more easier. And so you can try so much more because it's so much easier to do.
51:33 Speaker 01 Yeah, you can experiment much more rapidly. That is a big part of integrating AI across your teams is you can get feedback much, much quicker because the cost, as you said, to do those things is so much smaller. I think that's actually a really great point because where I think we're going to gravitate towards, I know even the future of marketing or the future of knowledge work. What I see happening is we're going to go to a place where we're going to have a lot of like super ICs and generalists that are enabled. People with AI but can do much, much more. And so you used to have these kind of people who were very disciplined and they did this one kind of thing. And then you had to do this one kind of thing. Whereas with AI, I think you can get people who are like super ICs, much more general. But they can experiment so much, much faster because they have AI assist in them. To me, that is like, again, one of the benefits.
52:17 Speaker 01 It's one of the things if you're not using AI, you can just learn so much slower.
52:21 Harry Stebbings? Going back, you said the LLM traffic, referral traffic, that is, is five X more than it was, but it's still less than 1%. In terms of proportion of referral traffic, you said less than 1% there. What is it in two years?
52:35 Speaker 01 I honestly think that maybe it's like 3 to 5%. Where is the value then? The value is in the impressions. The value is in brand metrics, which people may not be happy about. But I think that a lot of this way we measure this is going to be based upon the traditional brand metrics. It's like not causation, it's correlation.
52:55 Harry Stebbings? So I'm a CMO or CEO managing budgets. Do I go, fuck, we're going to see paid cats go through the roof with infinite supply of content? And distribution just being a shit show. I'm going to move everything to actually offline. Brand marketing, airports, billboards, buses. That's where the shift is. How does that change?
53:16 Speaker 01 If you're a CMO, you would say, okay, the thing I really need to figure out is how to build influence, right? How do I build influence with my target audience? And I need to build influence through AI. And influence through AI means I need to capture a ton of the impressions when someone is asking questions that are related to our product. I need to build influence across media. Media today is actually much more suited for creators and individuals than it is for brands, right? YouTube, podcasting, Substack, these kind of channels, I think are much more skewed towards people who have personalities. I kind of call it personality led growth, where they're like more suited to individuals. And so I think if you're a CMO, you would say, I really need to have some sort of creator first led content strategy that allows me to build creators for my market across these channels.
54:01 Speaker 01 And I want to build influence. And so how do I build influence there? I think there is going to be a lot of value in what we use. To call traditional outboundy type marketing that you can tailor and make much, much better. And I'll give you a good example. I'll give a little shout out to two guys I met in London, actually, called Hyper GTM, because I actually believe account-based marketing is going to be much more valuable with AI than it was previously. I think previously it was very hit or miss. And so the example there is they're able to tailor personal guests. They were able to like figure out they would have a deep research agent, it would pull a lot of data about you and it would figure out like the one personal thing about you that is quirky, that is unique, that is differentiated. And then they have an agent that would send you a gift, right?
54:41 Speaker 01 And that's an example of a kind of traditional marketing tactic that gets way better, way more scalable, because it's tailored to you. And so if I'm a CEO, I'm like, okay, what are the ways that I can now tailor marketing much more to my audience and get much better there? And then the last one is paid advertising is going to become more competitive. There may be new channels coming online. I don't, I kind of posted about this last week, where the real winner of AI search so far is like Quora. Quora is actually, you know, their number of citations are seeing in Google AI overuse has drastically increased, because what I said was there's many ways to ask these questions now and AI wants all of that data. So it's actually showing Quora and a lot of more of the results.
55:21 Speaker 01 So maybe there's like ad platforms that will start to grow and we can leverage that we were not able to leverage in the past. But even the ones that you're leveraging today, AI is able to help you do creative at scale. So instead of having 10 tests running, you can have 100 tests running with different creative, because AI can actually create the creative for you, right? So you can actually start to optimize your paid much more rapidly. I think there's a playbook being built for CMOs. Like I think it's AI search. How do I get a lot of impressions? It's creative-led content marketing. It's how do I look at old historical marketing that maybe was not that tailored, that can be much more tailored? I like to think of that as like micro audiences. How can I go from marketing to 300,000 people to audiences of 300 people, because AI enables me to do that?
56:03 Speaker 01 And then how can I take these traditional performance channels like paid advertising and integrated AI? So I can do it at scale and learn much more rapidly, do a lot more creative testing and really dial in what I'm doing.
56:13 Harry Stebbings? A traditional enterprise CMO is fucked, Karen.
56:16 Speaker 01 Well, I think the CMO that is really fucked is the one who does people management and moves budget around. Like, I think the the future is really for super ICs. I think the CMO can actually do the craft, use AI, ship things, understand how to do things because they're going to have smaller teams. And so I think because they're going to have smaller teams, a lot of this traditional people management stuff that they used to have to do is going to get less important over time. What gets more important is actually the work and how you actually integrate AI to be much, much better at your job and your team's job. When I think about CMO of the future, I think they have to kind of operate some way at both ends of the spectrum, right? So we call marketing art and science. And I think about it more as like engineer and tastemakers.
56:56 Speaker 01 And so i- if you have opposite ends of the spectrum, you have one marketer who comes from a data background, is like very process orientated. That person is going to excel because now they can actually figure out how to integrate AI, build AI workflows to make their team much better, to be able to do work much faster, to be able to learn much faster. But then you still need someone who's a tastemaker because you need to be able to tell incredible stories to stand out through the noise. You have to tell people why your product actually matters and why it's differentiated for that audience. You have to break through all of the kind of noise happening across all of these different channels. And so you kind of have to be like pretty exceptional at both ends of that spectrum. And everything in the middle, which is just like, fine, but I'm like, I'm fine at these things.
57:38 Speaker 01 But the way I have succeeded is because I have big teams and I know how to do a lot of people management. I know how to do a lot of hiring. I think that becomes less important over time.
57:46 Harry Stebbings? I think the core question in AI is something that I said earlier, which is fundamentally, will we see the transition from software budget to labor budget for AI tools? When you think about the core question for you that not enough people ask, what do you think that is?
58:02 Speaker 01 I think I see people who have a lot of assumptions about AI. Like, I believe AI is going to be phenomenal here. I believe AI is going to be incredible here, but they haven't actually deployed anything. They haven't actually tested anything. They haven't actually learned anything. And I do believe that your assumptions drastically change when you start to deploy this stuff. And so who actually wins in that space? It's the team who is able to go much, much faster. And so when I look across companies that I have Talked to, they think of AI as a software project. They're like, oh, we're gonna have our IT team deploy AI here, here, here. Whereas coming way back to the very first, you know, where we started is for me, AI has to be treated like a growth project. Right? You need a growth operating model. You need to ship each week. You need to have like, what did we learn?
58:45 Speaker 01 What are we learning? How do we be much more agile? How do we be much more experimental? It's not a software project. It's not a software deployment because it is still like hit or miss. You get success by learning our email conversion rates that I talked to you about. They were like 10 months in the making. You know, the first three months sucked. I have this thing, which is if you can't do something in six weeks, it's not worth doing. If you're in growth, right? You actually have to have minimal viable experiment. Figure out if you can do that thing and see a success, what I call an exit metric, but see a success metric after six weeks to continue on to the next phase. Now, if I had enforced that for our AI personalization and auto prospect and via AI, we wouldn't have been successful because after six months, the results were not good.
59:28 Harry Stebbings? Why did it suck for six months?
59:30 Speaker 01 Sorry, six weeks, six weeks.
59:32 Harry Stebbings? So why did it suck for six weeks? And is there anything that you could do now to make that better, more efficient? Like for me, tell advise me, I'm at the start of my journey where I'm like, okay, I want to have AI implemented in more of the GTM process. What do you know that can help me ramp faster in that six weeks?
59:52 Speaker 01 The your data layer really matters. Most people do not have good, clean data sources. And I think that's your both your internal data and your external data. For most companies, they can figure out like good external data sources. You can get much more creative with your external data source. Like I can give you one quick example to bring this to life. There's a concept we're working on called microaudiences. And so microaudiences is how do we use data and context tags to divide groups into smaller groups? I've kind of been talking about this a lot throughout this podcast. And so an example would be, and this is just an example of built for illustration purposes internally, is we would take a list of what we call, quote unquote, our product champions. And so product champions are people who use the product, people who have given us a positive MPS, people who have been
1:00:36 Speaker 01 To the product and activate it on the product like there's a big seed list of product champions. Then we say, cool. Now we're going to go and look at the external data and we're going to take all of their job titles. And look at jobs posted in the last 12 months for those job titles. And we've also built a seed list of KPIs that that list are generally in charge of. So we have a seed list of KPIs. So the AI has gone off, said, here's your job titles. Here's all of the job ads posted for those job titles in the last 12 months. Here's my seed list of KPIs. I've seen all of these KPIs and I've stacked ranked them by how many times I've seen them. And so now we have a stacked list of KPIs. And so what we do then is we can now personalize your experience based upon, hey, we actually know based upon this job title. This is a rise in KPI. It's something that people are more and more being tasked with.
1:01:21 Speaker 01 And I can give you a good example is I did this for a VP of sales and the one that actually popped up that I thought was really, really interesting was productivity savings. That was the fastest rising KPIs across job ads that they were being asked to do this in a way that actually saved headcount. They were asked to be more productive. They were asking to automate more of the workflow. That was like a VP of sales. So we would never have usually marketed to a VP of sales in that way. We would have said pipeline. We would have said deals created. We would have said all the traditional KPIs. But because of the way we're leveraging data, we could say, well, there's this rising KPI. There's a group of VPs, marketers who really have to, or VP of sales, we have to think through that. And that we can tailor everything for them, the emails, the ads, the call notes for the salesperson who talks to these folks.
1:02:03 Speaker 01 And so the first thing is like, your data really matters and it's not your just traditional like B to B type data. It's how you can be much more creative around the data sources you capture. The kind of little tip for the email specifically is to divide email into different prompts. The third tip is like, really think about styles, right? There's a style that will probably work for your audience and how you can extrapolate how AI can help you craft styles and that will really work. It's just trial by iteration. Like it really is that mundane. It's like the more you experiment, the more feedback you get, the quicker you're going to get to a positive outcome. And I think you have to have some concerted bets. And I think that's what I've learned is personalization, even if it takes some amount of time for it to work for you. It's a bet we're making because if you believe AI is going to
1:02:46 Speaker 01 Be incredible in certain areas and personalization is one of them. You might as well actually make a real long term bet to actually figure out how it can work for your business.
1:02:54 Harry Stebbings? What bet have you not taken yet that you would most like to take?
1:02:58 Speaker 01 Multimodal chat, we're still at the early innings of how we could replace over time our text based chat with a true multimodal experience, which means that you can flip between conversating via text, you can do a video call, you can actually share your screen and ask some questions when you're in the freemium tier. There's a couple of things, consumer behavior is probably still some way away from talking to an avatar. Now, we did run an experiment, we had people conversate with an avatar at the lower end of our market. When they wanted to do a demo, instead of watching the traditional product video, they actually conversing with an avatar and I learned two things about that. People enjoy talking to the avatar, the average engagement rate was eight minutes, so people actually enjoyed it. And over time, the conversion rate for people who talk to the avatar was much better than watching the video. So people really did get better educated and could make better decisions and they ended up converting much higher.
1:03:42 Speaker 01 But the intense switching is really hard for multimodal chat, which is like, is this a sales conversation? Is this a customer support conversation? I think that takes a little bit of time to like really figure out. And the switching between like text, voice, avatar, share and screen takes a little more time to, it's gonna take some time to figure out.
1:03:58 Harry Stebbings? Dude, I would love to do a quick fire there because I could speak to you all day about this.
1:04:02 Speaker 01 Yeah.
1:04:02 Harry Stebbings? What do you think a growth channel that's dying is over the next two to three years?
1:04:07 Speaker 01 I feel really bad saying this, like it is Google search. I just see, I see like Google search is one of the best channels we had in B2B. Eighty percent of all purchases started on Google search. It's predicted by 2027, ninety five percent will start in LLM. It's like a drastically different world.
1:04:24 Harry Stebbings? What tool other than chat GPT are you obsessed with right now?
1:04:28 Speaker 01 Oh, that's a, that's a great one.
1:04:30 Harry Stebbings? So like for me, it'd be Descript. Descript from an editing perspective is incredible actually, and they're eating more and more of that editing bandwidth away from our teams.
1:04:39 Speaker 01 I can tell you the one I'm going to be obsessed with, but I only recently started getting time to get into it personally was Captions Mirage. We had the founder on the show. That tool to create short form video is awesome. The short form video they created, the ad, I didn't really see the ad they used to launch that. It's like the best short form video that I've seen created through AI.
1:04:59 Harry Stebbings? I was a pre-seed investor in the company.
1:05:01 Speaker 01 There you go.
1:05:02 Harry Stebbings? Super happy about that. Thank you for that. What do you think is the biggest mistake that founders are making today when it comes to AI implementation?
1:05:10 Speaker 01 I think there's a big difference between AI-first companies and non-AI companies and how fast they're going with AI deployment. Is AI a solve for everything? No. Like if you saw Klarna's latest article where they're rolling back on some of the AI progress in favor of humans, you can go too fast, but there's a real balance to strike because I don't think there's ever been a better opportunity to be in the first mover group. Like I think anyone who's in the first mover group and how they find the leverage in AI gets all of the benefit. I think it's probably not going fast enough if you're a non-AI first founder.
1:05:38 Harry Stebbings? I tweeted a year or so ago, "AI services businesses will make more money in the next five years than actual model providers." Many people disagreed with me. I stand by it when you look at a lot of European corporates needing massive handholding help. Do you agree? Or do you think I'm wrong?
1:05:55 Speaker 01 I 100% agree that I think that there's going to be a huge influx of companies who can help companies deploy AI across their businesses. It's still super complex. It's super hard to do. It's, again, I think part of the reason it's so hard to do is the data really is a big thing. I think most companies do not have their data in a state where it's going to be easily and readily usable for AI, especially like just the unstructured piece, right? Like if you're just using kind of the traditional data you have, I will guarantee your results are not going to be that great. The more you can enrich the data, the more you can get the unstructured data, all of these things working, it will really drastically increase your results. But I totally agree with that. I think about that for two ways. And you've probably seen this because you've probably invested in some companies. Also, some of the best founders or ex-service companies, you know, they were doing like some sort of
1:06:37 Speaker 01 like niche thing through an agency or they had some sort of niche service and they just transformed that into an AI piece of software or an AI agent. And they're some of the best startups because they really understand that discipline and they have so much data through the services businesses that they can then leverage for the actually AI training.
1:06:55 Harry Stebbings? I actually love founders that have created agencies. Before, I always think to Desno and at Intercom when I think about agency to startup and it just gives you unlevelled customer insight to product building.
1:07:04 Speaker 01 Yeah, it gives you training data. What are we really doing here? Like, what are all these companies actually doing? They're training AI. Like, the core uniqueness of it really is to train in the AI to get better, and services businesses have so much more data.
1:07:16 Harry Stebbings? Are the qualities of what made a great growth hire in the past very different to what makes a great growth hire in the future?
1:07:24 Speaker 01 When I hire growth people, you want curiosity, you want experimentation, you want someone who's going to go after, and I come and be relentless and persistent. Then the only thing changes is that that person, if you were in a larger company, you hire a growth person to do onboarding, you hire a growth person to do monetization, you hire a growth person to do something else. I think now you can have one person enabled by AI to do all parts of that role. And so I just think that that person can do much more, but the core characteristics should be the same. I actually think AI is a boom for people who are growth oriented. Like, it is a just a an avalanche of momentum for people who are deeply curious and persistent. The best growth professionals I've worked with, they have those characteristics.
1:08:05 Harry Stebbings? Final one, what are you most excited about? When you think about applications of AI today, what are you most excited for? And so like, for me, probably today, it would actually be long-form video editing. We still have a lot of team spent budget-wise and time on long-form video editing. If you could throw in a video file and edit it beautifully, that would be a game changer for our business.
1:08:28 Speaker 01 That would be pretty awesome. There's a couple, I think, the AI's ability to refine your ICP using data. We have, you know, these loose ways of defining our ideal customer profile. I think AI is able to continually refine that based upon external and internal data. I know that's kind of machine learning. Again, I think about the transition from marketing of being this kind of segment-level marketing to microaudiences, and that's how we get to microaudiences. I think the thing I would advocate for is the ability for AI to create templates out of everything. So in your example, you would have AI able to edit that long-form video. But I think the AI's ability to edit the long-form video and then templatize that video, so every time you wanted to recreate it, it recreated it perfectly. That, to me, is like really a really interesting application of AI where AI can decipher what are the building blocks of anything and can replicate that thing for you.
1:09:13 Speaker 01 And it's pretty good at that. Now it's not like perfect, but it's pretty good at that now. I think multimodal 100% changes everything we know about B2B buying because today we have, you know, the marketing team hands off to the sales, the BDR team hands off to sales, the sales team hands off to customer success, customer support show up when there's a problem. But this kind of multimodal interaction where someone can just conversate with the same person across that entirety of the journey and that person has all of the context and understands everything about where you are within that process and they can switch between helping you solve a problem to helping you buy more software. I think that's a big one. The other one, which is just like a really mundane one, but is the thing that I have like really struggled with for a long time is dynamic context windows. I just cannot emphasize how incredible dynamic context windows are. So dynamic context
1:09:57 Speaker 01 windows would be, you know, you go to Claude or OpenAI and you say, okay, well, I want to work with you on this thing. And so whether I create a project or whatever it may be, I upload 10 documents. And so I have to go and find the documents and I upload them and it has that in its context window and then works with you. What I actually am waiting for is when I can say, okay, let's work on this today and it can dynamically go and find that data because it's plugged into my systems and load all of that into the context window. It can find like the relevant data for the thing that we want to work on and load that into the context window. And then whenever I finish that conversation, it can actually just get rid of all of that from its context window. So it dynamically loads the right data into the context window based upon the thing that you were trying to work on. That one for me is like the thing I've been waiting on a long time. You know, the other one I was working waiting on for a long time, which is a
1:10:42 Speaker 01 very mundane one as well, is like Claude's ability to build apps on Claude now and Claude can call its own API automatically. So I don't know if you've tried their new use case they launched last week, but it's pretty incredible. Now you can actually go in, they're kind of building these kind of interactive artifacts and you can just build apps and it will automatically just like call the API and integrate Claude into the app you're building. Does that kill Replin? If I think about all these AI assistant tools and I think lovable are pretty incredible. You know, lovable secret sauce is like whatever they're doing in terms of being able to make incredible front ends, that's what I think really helped them create the zeitgeist. But Replin, Replin have done a lot of the hard work of doing like backend stuff and database integrations and things like that. And so I don't, I don't see Claude as being as good as lovable.
1:11:24 Speaker 01 Or Replit. I think it's much more between Replit and and Lovable. And I think it's like a sad state of affairs where neither of us would probably even mention Google Stitch. Like, why are they so, why are they so bad at this stuff?
1:11:35 Harry Stebbings? Google Stitch?
1:11:36 Speaker 01 It was called Fire Studio and then they rebranded it, but it's basically their app builder.
1:11:40 Harry Stebbings? Wow.
1:11:41 Speaker 01 And the fact you didn't even know about it says everything you need to know.
1:11:44 Harry Stebbings? I don't think Replit and Lovable compete in the same way that people think they do, though. Like, I think Replit is for much more heavy-duty application build-outs, and I think Lovable is much more for lightweight, bluntly, front-end software build-outs. And both have very valuable positions in market, but they're just very different use cases.
1:12:00 Speaker 01 There's also a big enough market for the both of them to be incredibly huge companies because what really this does is turn everyone into a builder. So now you have a whole world of builders who need builder tools, and they're the two de facto tools. I think Replit have just updated their AI coding agent, and it's, I don't know, they've done something. It's like pretty incredible. So these tools are just getting so much better.
1:12:20 Harry Stebbings? Dude, I've absolutely loved this. I can't thank you enough for joining me, for shooting this shit. No structure at all to the way I planned it. But you've been amazing. So thank you so much, dude.
1:12:32 Speaker 01 I love coming on, so I appreciate it.
1:12:36 Harry Stebbings? I just love doing that show. I mean, two things to me that I really love doing this show. One is building amazing relationships with fantastic people like Kieran there. And the second is having very natural and fluid conversations that really change the way you think about things that you thought you knew before. Kieran was amazing there. So appreciate his time. And if you want to check it out, you can find it on Spotify by searching for 20VC. But before we leave you today, SecureFrame empowers businesses to build trust with customers by simplifying information security and compliance through AI and automation. Thousands of fast-growing businesses, including NASDAQ, Angel List, Doodle and Coder, trust SecureFrame to expedite their compliance journey for global security and privacy standards,
1:13:17 Harry Stebbings? such as SOC 2 and ISO 27001, CMMC, MIST standards and more. Backed by top-tier investors and corporations like Google and Kleiner Perkins, the company is among Forbes' list of the top 100 startup employers for 2024. G2's Best Software Awards for Highest Satisfaction Products and a recipient of the 2024 Cyber Security Excellence Awards. Something I definitely never got in school myself. Learn more today at SecureFrame.com. Now let's switch gears to the future of mobile and how one company is turning your screen into real returns. We spend nearly half our waking hours glued to our phones, upwards of 50 hours every week.
1:13:57 Harry Stebbings? Recently, one company transforming this reality stood out so much I personally became a shareholder. Mode Mobile. Mode Mobile created the Earn Phone, the smartphone that pays you for daily activities. Instead of big tech profiting billions from our attention, Mode returns over $325 million directly to users through earnings and savings. Mode's revenues surged an incredible 32,481% in three years. Recognized by Deloitte as 2023's fastest growing software company in North America. And here's why I'm excited. Mode's equity offerings have raised over $30 million from 20,000-plus retail investors, one of 2025's standout public raises.
1:14:40 Harry Stebbings? And you can now join me as a shareholder with as little as $1,000 at invest-modemobile.com/20vc. For a limited time, unlock up to 100% bonus shares and a free Earn Phone. Email us for the investor brief at [email protected] or check out invest-modemobile.com/20vc. We've talked about making money from your phone. Now let's talk about making your product do exactly what you want it to. Whether it's the software you build or the software you buy, your tech stack should be creating results, not creating roadblocks. Well, Pendo's no-code software experience management platform makes your software better with tools to see where users get stuck, guide them with in-app messaging,
1:15:24 Harry Stebbings? and constantly improving your UI. It's so easy that over 14,000 businesses use Pendo to increase revenue, lower costs, and reduce risks. Businesses love the control, engineers love the freedom, everyone wins. Start for free today at pendo.io/20product. As always, I so appreciate all your support and stay tuned for an incredible 20 VC coming on Monday with Vlad Tenev, founder and CEO at Robinhood.
Previous 20VC: Daniel Gross and Nat Friedman: Acquired by Meta | OpenAI’s SBC Bombshell: More Stock Comp Than Revenue | Privat Equity is Back: Olo Bought for $2BN | Microsoft Lays Off 9,000 People: Is This Just the Start | Will Sequoia Part with Shaun Maguire Next 20VC: Vlad Tenev on Robinhood’s $85BN Resurgence | Tokenizing SpaceX & OpenAI | Building Nine Revenue Lines Over $100M | Why Crypto Will Be Robinhood's Biggest Revenue Line | Why Stablecoins Will Replace Banking Rails Page updated September 29, 2026 · transcript and summary are machine-generated
