Jul 11, 2025 · 1h 15m · 20growth
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
0:00 / 0:00
gold bands on the timeline = statements, start to end. Hover to read, click to jump. CC turns on captions
In this episode of 20 Growth, host Harry Stebbings interviews HubSpot AI implementation lead Kieran Flanagan to explore how artificial intelligence is disrupting product-led growth teams, SEO, marketing stack execution, and executive leadership playbooks.
How this conversation actually went
Every chapter scored 0–10 on four independent dynamics. Hover any point for the reasoning behind the score. Harry holds 25.4% of the talking time here. How this is scored →
Harry as informed peer 5.6 Guest teaching 5.4 Guest disagreement 2.1 Harry pushing back 4.3
speaking balance: gold is Harry, purple is the guest (3 minute bins)
Sharpest disagreement ▶ 28:33 Kieran challenges software defensibility premise
Kieran directly counters the host's belief in software longevity, asserting that single point solutions face collapse due to cheap code and shrinking distribution channels.
Hardest push from Harry ▶ 15:05 Harry forces a yes or no on budget shiftsHarry refuses Kieran's diplomatic framing around headcount redeployment, explicitly demanding a binary yes or no answer on whether software budgets will eat labor budgets.
Biggest teaching moment ▶ 25:00 Kieran breaks down Cloudflare scrape-to-visit metricsKieran educates the host on the content collapse using Cloudflare metrics, explaining how scrape-to-visit ratios exploded from 2:1 to 18:1 on Google and 1500:1 on ChatGPT.
Harry holds his own ▶ 28:33 Harry argues against software commoditization using design craftHarry steps in with sharp venture perspective, arguing that elite product execution and design detail prevent high-end software from becoming commoditized despite cheap code.
the scores for every segment, with the reasoning behind each
| Chapter | Topic | Harry as informed peer | Guest teaching | Guest disagreement | Harry pushing back | Why |
|---|---|---|---|---|---|---|
| Reimagining Growth Teams: From PLG to AI Innovation Pods | 3 | 5 | 2 | 4 | Harry challenges Kieran on whether HubSpot has actually implemented his thesis by asking if they disbanded their growth team. Kieran explains his vision for AI innovation pods replacing traditional touchless growth structures while acknowledging HubSpot hasn't fully integrated them yet. | |
| AI Tooling Realities: Personalization vs. Overhyped Agents | 5 | 4 | 3 | 6 | Harry presses hard from an investor standpoint, repeatedly pushing Kieran for a direct yes or no on whether AI tooling will eat labor budgets and reduce headcount. Kieran provides nuanced pushback, arguing that time saved is often redeployed to higher value tasks rather than immediate cuts. | |
| Advanced Prompting and Data Strategies for Email Optimization | 4 | 6 | 1 | 3 | Kieran schools the host on tactical AI email optimization, breaking down prompt decomposition and style extrapolation that tripled conversion rates. Harry asks tactical clarifying questions regarding style homogeneity versus individual conversion dynamics. | |
| The Content Collapse, Search Disruption, and Infinite Supply | 7 | 6 | 4 | 6 | Harry uses economic principles to challenge Kieran's assertion that software is becoming commoditized, defending product nuance and design depth. Kieran counters with brutal Cloudflare metrics illustrating the content collapse and exploding scrape-to-visit ratios. | |
| LLM Defensibility, Memory Moats, and Prompting Mastery | 6 | 5 | 1 | 2 | Harry demonstrates strong venture thesis by highlighting LLM memory as an unbeatable retention moat against model commoditization. Kieran strongly agrees and shares operational tips on prompt engineering using custom GPTs and vendor documentation. | |
| AI-Generated Creative Media and Human Taste | 7 | 5 | 2 | 5 | Harry brings deep venture knowledge to discuss valuation multiples and seed metrics for LLM analytics startups. He then pushes Kieran on whether staying in Ireland professionally hinders him due to delayed EU AI model rollouts. | |
| Modern Marketing Stacks, Super ICs, and the Future CMO Playbook | 6 | 6 | 2 | 4 | Harry makes a bold declaration that traditional enterprise CMOs focused solely on people management are obsolete. Kieran elaborates on the new CMO playbook, emphasizing the need for technical process engineering paired with human taste. | |
| Deployment Frameworks, Multimodal Buy-In, and AI App Builders | 7 | 6 | 2 | 4 | Harry asserts his public thesis that AI service agencies will generate more revenue than model providers, standing firm against past pushback. Kieran agrees and breaks down advanced micro-audience targeting and dynamic context window workflows. |
Statements from this episode (32)
Flanagan: AI will merge growth and GTM teams into innovation pods
“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 like go to market to make it a much more efficient go to market?”
Flanagan: HubSpot maintains separate PLG and AI innovation teams
“No, we haven't. Which is why I think I'm going to get in trouble for saying this. So we have 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 hu…”
Flanagan: AI will eliminate traditional GTM handovers across sales and support
“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 again, AI is …”
Flanagan: Customized AI prompts outperform off-the-shelf GTM software
“What we have found is the more that we customized a prompt for our needs and the way that we customize that for our needs, the results are much, much better.”
Flanagan: HubSpot is redeploying support teams affected by AI, not cutting them
“We are redeploying those teams to do other important work.”
Flanagan: Every corporate function will be smaller in the future
“I think every function is going to be smaller in the future.”
Flanagan: AI agents will be priced against human labor costs
“Yes. I think we're going to price agents against human labor costs, outcomes.”
Flanagan: AI isn't reliable enough yet for autonomous multi-task 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 bac…”
Flanagan: HubSpot's AI chat agent is booking meetings and closing sales
“We have an AI chat agent that is able to sell now. It's booked in meetings. It's doing sales.”
Flanagan: HubSpot achieved 5x email conversion uplift using AI
“And if you had told me 12 months ago that we could use AI to like five X the conversion rate on email, cause I'm like, ah, look, there's only so much personalization you can do in an email.”
Flanagan: Break AI email creation into separate prompts for each section
“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 …”
Flanagan: AI enables economical marketing to micro-segments under 300 people
“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, Very, very fine. We c…”
Flanagan: AI Overviews now appear in over 35% of Google searches
“That's on over, like, 35% of all searches today have some sort of AI overviews experience, and so that's rolling out much more rapidly.”
Flanagan: ChatGPT scrapes 1,500 web pages for every single referral visit
“Six months ago, it was 250 pieces of content scraped, and they give you one visit back. Today, it's 1500 pieces of content scraped, and they give you one visit back.”
Flanagan: Informational content marketing will hold no long-term value due to AI
“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 AI.”
Flanagan: AI search optimization requires creating hundreds of micro product pages
“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.”
Flanagan: Memory is an LLM's most defensible feature
“I think it's the most defensible part of an LLM.”
Flanagan: Ask AI to write its own prompts for better results
“The number one tip I can just give anyone to get better at prompting is to ask the AI to create the prompt.”
Flanagan: Viral AI gorilla ad booked LA dentist for years
“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 go into his practice to get them fixed. And it went viral. And he apparently booked up his dental practice f…”
Flanagan: HubSpot LLM traffic grew 5x but remains under 1%
“We've seen a five X increase, but it's still less than one percent of our traffic.”
Flanagan: Share of voice matters more than referral traffic in LLM search
“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.”
Stebbings: I would put my entire $400M fund into OpenAI
“If I could do anything without fear of retribution from my LPs, I'd put my four hundred million fund into OpenAI.”
Flanagan: Enterprise AI stacks will dynamic-route between open and paid models
“I think that most companies at some point will have a 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.”
Flanagan: EU AI regulations are driving European founders to the US
“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.”
Flanagan: AI will empower generalist 'Super ICs' over narrow specialists
“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 with AI, but can do much, much more.”
Flanagan: LLM referral traffic will reach 3% to 5% in two years
“I honestly think that maybe it's like three to five percent.”
Flanagan: Quora citations in Google AI Overviews have drastically increased
“The real winner of AI search so far is like Quora. Quora is actually, you know, Their number of citations are seen in Google AI overviews has drastically increased because what I said was there's many ways to ask these questions now and AI wants all of that da…”
Flanagan: Traditional management-focused CMOs will become obsolete
“Well, I think the CMO that is really fucked is the one who does people management and moves budget around, right? I think 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 th…”
Flanagan: Growth experiments failing to show metrics in six weeks should be dropped
“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…”
Flanagan: 95% of B2B buying journeys will start in LLMs by 2027
“Google search is one of the best channels we had in B to B. 80% of all purchases started on Google search. It's predicted by 2027, 95% will start in LLM.”
Flanagan: Claude's app builder is inferior to Lovable and Replit
“I don't see Claude as being as good as Lovable or Repplet. I think it's much more between Repplet and Lovable”
Stebbings: Replit handles backend builds while Lovable focuses on frontends
“I think Rapplet 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.”