Summary

Luke Harries originally dismissed Mati’s ElevenLabs plan as “a terrible go-to-market plan.” The company instead built foundational audio models and sold them across developers, enterprises, creators, and consumers. Luke still advises most founders to focus, but says exceptional models, natural demand, financing, and founder-type product owners let ElevenLabs shard itself into discrete businesses without losing velocity.
ElevenLabs scales growth through a matrix of embedded product teams and horizontal channel specialists. Each product has a growth lead acting as its CMO, while experts in performance marketing, SEO, and other channels work across the portfolio; enterprise marketing alone was expected to reach 20 people, versus five to 10 for the mobile app. Luke’s default startup sequence is much leaner: hire one product-fluent generalist growth marketer, then a front-end-leaning growth engineer who can expose value through landing pages, mini-tools, SEO, and automated outreach.
The launch machine begins with one primary claim, builds every asset around it, and manufactures “surround sound” across every available channel. For ElevenLabs Speech-to-Text, the repeated claim was “the most accurate speech-to-text model,” supported by diarization, character-level timestamps, and 99 languages. Tier-one launches get a sharp first tweet, motion-design video, technical blog, ubiquitous cross-posting, and immediate employee amplification; Luke’s instruction is blunt: “You need to be loud” and shamelessly DM the network required to trigger distribution.
Video remains ElevenLabs’s most effective launch medium, with 200,000 to 700,000 views typical and almost all creative attention concentrated in the first 30 seconds. Luke favors motion design for major launches, screen shares for technical buyers, and founder-led films only when the format suits the brand and can hold attention; a motion-design project typically costs $5,000-$10,000. He acknowledges supply could eventually erode video’s advantage, but currently sees no better format for compressing a complex value proposition.
Growth efficiency is managed through CAC payback, not a deceptively precise CAC:LTV model. ElevenLabs accepts roughly 12-24 months, potentially 36 for durable enterprise contracts, and Luke argues that teams comfortably above target should accelerate immediately rather than raising spend by an arbitrary 20% weekly. Individual channels usually get more expensive, but new channels, better activation, virality, conversion, and consumer-to-enterprise expansion can keep blended CAC flat.
Eliminating traditional PMs lets engineers own the entire product loop while former PMs migrate into growth. Engineers choose features, design, ship, and analyze results; an engineering lead owns product quality while a growth lead owns awareness and acquisition, with both sharing activation and retention. Luke estimates 60%-70% of core engineering code is AI-written, though sensitive research code is excluded, and expects PMs increasingly to become either marketing-capable growth leaders or product engineers using tools such as Cursor and Lovable.
Luke reversed his view of conversational AI after helping about 20 customers build voice agents. He had believed sub-200-millisecond interactions would be impossible, but says, “I was completely wrong.” ElevenLabs subsequently productized conversational AI as a platform while explicitly leaving avatar applications to partners such as HeyGen, Synthesia, and Captions. His clearest labor implication is that inbound SDR work—largely BANT data collection—can move into conversational agents that qualify buyers and route them directly to AEs.

Deep dive

1. Luke’s best lessons came from opportunities he initially misread

Luke met ElevenLabs’s Mati at a Cambridge hackathon when they were 19. Their team won Microsoft’s prize, sold the Xboxes on eBay for about £300, and the event produced Luke’s first Microsoft job plus a friendship sustained through twice-yearly startup conversations. Luke also notes that the organizer, Filip, later founded Wordware, which he thinks raised about $40 million—the largest YC raise ever.
Seven years later, Mati pitched ElevenLabs while they swam in the 11-degree Hampstead Heath ponds: build the world’s best audio models, then sell them to listeners, creators, developers, and eventually enterprises. Luke’s response was categorical—“That is a terrible go-to-market plan”—so he passed on investing; six months later, Mati called after reaching one million users and recruited him to lead growth.
Fella produced the more consequential correction. Luke and Richie entered YC with browser automation before GPT-3 existed, abandoned it, briefly ran a nonprofit COVID-testing clinic, and distributed Curative’s first tests. Their free beta accidentally drew a car park full of San Francisco tech billionaires in Lamborghinis and Ferraris; Luke’s niche lesson was not to beta-test pandemic testing. Curative subsequently became, in Luke’s telling, one of the fastest companies to $100 million in revenue.
After reading the semaglutide paper on the day it came out, Fella built a clinic around medication that reduced body weight by 15%. Revenue reached roughly $300,000 before flattening; Luke, impatient at around $200,000 annually, left while Richie stayed. The company later exceeded $30 million a year, leaving Luke with the lesson: “Commit for the long term, be patient”—especially when the market wave is still forming.

2. ElevenLabs makes horizontal expansion work by sharding the company

Luke contrasts ElevenLabs with PostHog’s cleaner model: one product-engineer ICP buying multiple compounding tools. ElevenLabs instead begins with horizontal audio capabilities—text-to-speech, speech-to-text, sound effects, voice cloning, and voice isolation—then packages them as APIs, enterprise workflows, conversational AI, a Reader app, and creator products.
Harry’s pushback is the standard one: “Focus.” Luke agrees that most founders should choose one ICP and build adjacent products around it, and says he is “not sure I’d recommend” ElevenLabs’s model. Its exception is rooted in best-in-class models generating natural demand, the ability to raise money, and “founder-type people” who independently own each product.
The operating answer is discrete consumer, creator, developer, and enterprise teams, each with embedded growth capabilities. Product growth leads act as CMOs for their product, while horizontal specialists—including a performance leader formerly at Shopify and an SEO specialist formerly at Canva—supply channel depth.
The scale is deliberately uneven: enterprise marketing alone was expected to reach 20 people by year-end, while mobile would have a separate five-to-10-person growth team. This is a portfolio of product businesses connected by shared expertise and foundational models.

3. The first growth hires must combine judgment with the ability to ship

When Luke joined, growth consisted of three junior specialists, and he had never held a formal growth role. His recommended first hire is nevertheless a generalist who owns positioning with the founder, awareness, channel experiments, and conversion—and, above all, deeply understands the product and its users.
A pure product marketer may create elegant messaging nobody hears; a purely quantitative channel operator may acquire traffic that never resonates. For a technical product, the generalist must be technical; for consumer, they must understand consumer behavior and virality.
Hire two should be a front-end-leaning growth engineer: “hacky,” metrics-oriented, and able to build landing pages, SEO tools, mini-tools, and automated outreach. Motion designers and back-end growth engineers are other useful hires as needs emerge.

4. Video wins launches in the first 30 seconds

ElevenLabs treats video as the keystone launch asset, especially motion design that can compress abstract value propositions and UI into an attention-holding story. Luke’s warning to founders: a five-minute mission monologue will not retain viewers “unless you have the editing skills of MrBeast.”
Nearly all creative attention should go into the first 30 seconds: enter quickly, state the core value, and only then extend toward five minutes for viewers wanting depth. The three formats are motion design for major or complex launches, founder-led storytelling, and fast screen shares for technical audiences that want product craft and detail.
Luke’s pointed example is Superhuman’s recent founder-led launch: the staircase looked beautiful, but by the time Rahul reached the bottom, “you’ve moved on.” Harry liked the cinematic staging; Luke’s objection was distributional, not aesthetic—the attention economy punishes delayed substance.
A freelance motion-design project normally costs $5,000-$10,000, while ElevenLabs launches commonly receive 200,000-700,000 views. Early contractor failures taught Luke to internalize video quickly because major launches often allow only about one week between product readiness and publication.

5. Every major launch runs through one message and many surfaces

ElevenLabs has three launch tiers: a new model or product line gets tier-one treatment; a meaningful customer feature is tier two; minor changes go to the changelog. The product growth lead first defines the audience, KPIs, core value proposition, and the short, consistent language everyone will repeat.
Speech-to-Text’s primary claim was “the most accurate speech-to-text model.” Diarization, character-level timestamps, and support for 99 languages were supporting claims—not competing headlines. Luke insists on one primary message that founders, staff, social posts, and collateral can repeat verbatim.
The first asset is the tweet thread because it forces the hook into one line: “Introducing the world’s best speech-to-text model.” Supporting detail follows, paired with the launch video; the penultimate tweet must remain strong, while the CTA and external link belong in the last tweet. Luke cites Elon Musk’s claim that X actively downranks posts that put the link in the first tweet.
That message becomes motion video, technical blog, and posts across X, LinkedIn, Bluesky, Threads, Product Hunt, Reddit, and Hacker News. An internal amplification channel coordinates early likes, comments, reposts, and employee threads to create “surround sound” and give the algorithms the initial engagement they reward.

6. Shameless distribution is a startup advantage, not an embarrassment

Founders often tell Luke ElevenLabs can launch loudly because it has a 200-person team, prominent investors, and influential friends. His reply is to inventory every reachable person: he recently worked with someone who had roughly 3,000 contacts across five years of Gmail, before adding X followers and LinkedIn connections for manual launch DMs.
Harry reinforces the unscalable tactic: he personally messaged his first 50,000 X followers and cross-promoted the newsletter for 15 minutes daily, helping it reach several hundred thousand subscribers. Luke’s rule is simple: “You need to give your launch a big boost for these algorithms to actually care.”
Channel focus and ubiquity are compatible. ElevenLabs perfects X for creators and developers and LinkedIn for employees, partners, and enterprise prospects, then reposts the work elsewhere—where neglected platforms may be easier to own.

7. Tool pages may outlive conventional SEO content

Technical launch posts still matter because technical audiences need benchmarks, implementation detail, and some “secret sauce”; they also attract backlinks from press, social distribution, and other sites. Luke distinguishes that function from generic long-form SEO articles, which he expects ChatGPT-style answers increasingly to displace.
Even that decline is not immediate: Luke says more than 70% of Zapier’s SEO traffic still reaches its blog. His higher-conviction call is that interactive tool pages will persist “for at least five years” because they require engineering, proprietary data, or dynamic utility rather than scraped prose.
Search “text-to-speech Spanish,” in Luke’s example, and ElevenLabs offers a box that accepts Spanish text, lets the visitor choose a voice, and plays the result. The enterprise equivalent is not exposing the whole product: identify “small bits of value” that deliver the wow moment before login.
ElevenLabs also hired its first in-house creator after discovering that his independent YouTube video ranked above the company’s own for “ElevenLabs.” The mandate spans TikTok, YouTube, Instagram Reels, and YouTube Shorts—bringing a proven external creator into the company.

8. Channel ownership begins when a weak signal becomes repeatable

Luke’s largest channel mistake was waiting too long to hire dedicated owners despite strong product-market fit. An engineer built the affiliate program in one week around 18 months earlier; untouched since, it was still producing tens of thousands of dollars in monthly recurring revenue.
His counterfactual is to staff any channel showing life with one person accountable for that product, channel, and KPI set. The objective is not broad marketing busyness but “laser-focused” ownership once the company sees signs of life.
Direct-response channels can be judged through unit economics; social, brand, and enterprise campaigns require fuzzy attribution over longer horizons. For a San Francisco push spanning billboards, podcasts, newsletters, and events, Luke would compare aggregate lead lift against comparable markets such as New York or Seattle rather than pretend each touch has an isolated ROI.
Enterprise marketing’s North Star is marketing-sourced sales-qualified leads: marketing identifies a lead, gets a call booked, and an SDR validates the opportunity. Webinars still contribute, though Luke argues the stale name is the problem—“Academy” better signals useful programming and email capture.

9. CAC payback gives permission to accelerate

Harry challenges CAC:LTV as transient and easily distorted by changing acquisition prices and product expansion. Luke agrees operationally: ElevenLabs manages CAC payback, with targets around 12-24 months and potentially 36 months for enterprise buyers likely to retain or sign multi-year contracts.
When a channel clears its threshold, Luke rejects timid 20%-per-week spend increases: “Put your foot on the gas as quickly as possible.” Falling below target can still be a deliberate bet on future expansion, but strong present economics should be treated as explicit permission to compound demand.
CAC usually rises within a mature channel as the obvious audience saturates. Blended CAC can remain flat through new channels, stronger virality, improved activation and paid conversion, and expansion into higher-value products—including ElevenLabs’s “funky move” of using cheap consumer acquisition as an entry toward creator or enterprise usage.

10. Enterprise growth is both top-down and bottoms-up

Luke “really hate[s] the word enterprise” because it hides the buyer. ElevenLabs found that the actual people were usually engineering managers or product leads, sometimes CIOs and CTOs—personas overlapping substantially with the developer audience and therefore reachable without defaulting to white papers.
Rather than choose consumer entry or executive sales, ElevenLabs does both. Enterprise marketing runs ABM, executive dinners, events, and webinar-like programs against marketing-sourced SQLs; developer advocates pursue broad awareness through hackathons and other events.
Brand should begin with the founder and community, not wait for billboards. Over time, companies may choose to allocate 20% or 70% of spend to awareness depending on performance elsewhere, but authenticity matters: Mati leans into firesides and large events, while Luke and the developer team carry the mission through channels they enjoy.
Founder brand is also a dangerous dopamine loop. Because startups are “10-year, 20-year-long games,” Luke advises founders to use channels that energize them rather than force daily posting and viral clips that distract them from building.

11. Counter-positioning works, but an early tag can become permanent

Luke defines counter-positioning as turning an incumbent’s central strength into a weakness. Brex celebrated card points and spending; Ramp argued points encourage waste and redirected the promise toward software and savings, allowing it to answer Brex from the opposite side of every message.
He sees TBPN using the same device: embracing ads against All-In’s no-ads posture, explicitly pro-tech against skeptical traditional media, and raw live output against polished programming. Harry’s reservation is that its channel-market fit may resemble Clubhouse—excellent X clips and volume, but weak underlying views and retention; Luke’s answer is “we will see” because the product is still evolving.
Bad press does exist when trust drives enterprise sales. Luke questions “Cheat on Anything” marketing built around a founder claiming he cheated through Amazon, Palantir, and Google offers; Harry suggests experienced reps and credible customers could reset the story in six to 12 months, but Luke says buyers may permanently remember the cheating tag.
The broader warning is that brands are difficult to retag—Harry cites Calm’s struggle to transcend “the meditation company.” ElevenLabs therefore wants to be intentionally known as the trusted player in audio for both voice actors and large enterprises.

12. Engineers own product; growth owns awareness and acquisition

ElevenLabs has no traditional PMs because “the engineers are building the product, and they should be responsible for the product.” Product engineers own the roadmap, move from idea through implementation and measurement, and avoid the approvals that break context and slow iteration.
Product and marketing are partly fused into growth, staffed heavily by former PMs. Each product pairs an engineering lead responsible for product quality with a growth lead responsible for awareness and acquisition; they collaborate on activation and retention.
Lean growth teams have neither time nor authority to drift into wireframes and roadmap management. Engineers are screened through a three-stage product challenge: research and select features, sketch the experience in Figma, then design the back-end and API architecture—testing the full loop beyond coding ability.
Luke expects PMs to migrate either toward growth, combining product judgment with marketing, or toward product engineering through Cursor and Lovable. He estimates 60%-70% of core engineering code is now AI-written; his own coding, about 20% of his time, happens entirely through Cursor, while sensitive research code never goes through an LLM.

13. Fast AI revenue is real only when cohorts retain

Harry raises the “sugar high” concern: generative-AI startups can race to $50 million ARR amid 10-15 plausible competitors per category. Luke’s answer remains conditional—revenue is real when the product solves a real problem and retention is strong—but he thinks investors are oddly slow despite companies reaching $6 million MRR within a year or already passing $10 million.
Luke distinguishes today from the COVID-boom “Tiger Global play”: expanding model capabilities now enable products that previously could not exist. If $100 million ARR was once the rough IPO threshold, entire cohorts are moving toward it unusually quickly, even if choosing category winners has become harder.
Harry recalls Lovable reporting roughly 86% retention and notes that 14% monthly churn would look terrible in traditional SaaS. Luke reframes the unit: an enterprise account can reach 150% NRR while many individual Slack users churn; similarly, a prosumer at 87% revenue retention may share a Lovable-built product that recruits enough new users to push the overall account-level NRR above 100%.
Luke is bullish on Lovable owning more of software creation as model intelligence rises beyond the roughly 60%-70% AI-generated code already seen at ElevenLabs. The challenges are making the product secure, getting enterprises to build with it, and supporting full apps; he sees Lovable as well placed to own more of that core stack and suggests Retool may be asleep at the wheel.

14. The sharpest growth doctrine is product first, copy second

For B2B founders, Luke’s costliest mistake is paid acquisition before product-market fit: it consumes engineering time on funnels, creatives, and “meta metrics” when launches plus a remarkable product can establish initial demand. Consumer is the caveat because distribution itself may be existential, justifying earlier performance tests.
Harry suggests paid spend can test messaging, colors, fonts, and titles; Luke replies, “I’ve never seen that work,” then asks whether early 20VC needed A/B tests or simply better episodes. Harry concedes the point: build a better product first, then lean into paid.
Organic LinkedIn is Luke’s most underappreciated channel because X pits founders against the world’s best writers, while LinkedIn pits them against “Freddy from Deloitte who got a promotion.” LinkedIn ads are the polluted inverse: excellent targeting, prohibitive CPMs, and long enterprise cycles make outstanding organic content the better foundation.
Copywriting is the foundational growth skill across ads, posts, blogs, threads, and landing pages. Luke’s related launch rule is equally uncompromising: do not market unfinished products—Apple’s promotion of Apple Intelligence before users could access clear value is his leading counterexample.

15. Conversational AI changed Luke’s mind about the application layer

Luke entered ElevenLabs convinced natural voice agents would fail because latency could not fall below 200 milliseconds. After helping about 20 customers build conversational agents, his view flipped: “I was completely wrong,” and ElevenLabs turned the repeated implementation pattern into its conversational-AI platform.
He now prefers AI customer support in some cases because agents know policies, diagnose problems, and escalate without requiring social niceties. The clearest near-term displacement is inbound SDR work: BANT qualification is largely data collection that a conversational agent can complete before routing a credible buyer directly to an AE.
Infrastructure providers can move into applications, but boundaries preserve partner trust. ElevenLabs wants to own end-to-end conversational-agent infrastructure while explicitly avoiding full avatar platforms, leaving that layer to customers such as HeyGen, Synthesia, and Captions.
Voice cloning does not erase permission requirements: ElevenLabs does safety work to ensure voices are used where their owners have given permission. Luke’s favorite recent growth strategy remains Bryan Johnson’s authentic combination of controversy, the “Don’t Die” message, community, and constant participation—the larger principle being that distribution compounds when the founder’s public behavior and actual belief are inseparable.

Alex Rampell: The Best Founders Materialise Capital, Customers & Labour | The Future of Venture Randy Baron's "Spicy" Victoria PLC Pitch