Luke Harries
Key Views & Dialogues
20Growth: The $6.6B Growth Engine Behind ElevenLabs | Why ElevenLabs Do Not Have PMs | The 7 Part Launch Playbook to Crush All Launches with Luke Harries, Head of Growth @ ElevenLabs
ElevenLabs scales across developers, enterprises, creators, and consumers through embedded product growth leads, horizontal channel specialists, and engineers who own the product loop without traditional PMs. Its launch playbook concentrates every asset on one claim, with video typically drawing 200,000-700,000 views, while CAC payback targets of 12-24 months—and potentially 36 for enterprise—frame how aggressively growth can accelerate.
View summary (7)
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.
Rippling vs. Deel Lawsuit: WTF Happens Now? The Future of the Late Stage Private Market
Chime’s IPO could price 50–55% below its $25B private round, exposing how ratchets transfer late-stage valuation risk while its 75% debit-interchange revenue remains regulatory-sensitive. IPO timing is now a cost-of-capital decision, with LP pressure rising, M&A buyers refusing to step up, and top exits increasingly reflecting longer private compounding rather than broad market improvement. The AI opportunity favors model providers such as Anthropic, but MCP could turn SaaS applications into databases; Chegg’s collapse and a potentially 10-year replacement cycle define the risk.
View summary (7)
- Chime’s IPO reprices the entire late stage. The company is genuinely good — $1.67B 2024 revenue growing 30%+, 8.66M actives, two-thirds using it as their primary account — but with the last private round at $25B and a likely IPO valuation “plus or minus 10 billion,” it’s going public roughly 50-55% below the last round. The caveat investors should clock: 75% of revenue is debit interchange riding Durbin amendment arbitrage through sub-$10B partner banks, and “if you’re Jamie Diamond, you wake up every morning spitting mad.”
- Ratchets mean the 2021 vintage never raised what it thought it raised. Rory’s framing of mandatory-conversion price protection: “You didn’t raise money at 25 billion. You thought you raised money at 25 billion. But in fact, if you go public at 12, you raised money at 12 and you just didn’t know it” — the true cost of that capital was double what the founder believed. Fully protected late-stage investing becomes “the world’s best business”: the only risk is overpaying, “and now it can’t go wrong anymore.” Harry’s conclusion: “seed is for suckers.”
- The IPO decision is purely a cost-of-capital decision. Lenders and financial players should go public because they need continuous capital access; Stripe, SpaceX, OpenAI and Anthropic have “infinite private capital at dirt cheap rates,” so why bother. Meanwhile the LP pressure is building — a retired LP described as one of the top 10 of all time told Jason flatly: “These guys just got to sell or go IPO. It’s just time” — and that cascades LP → GP → CEO.
- This is not the M&A year. Two of Jason’s portfolio companies got $500M offers; both acquirers walked rather than step up, one buying a ~$100M smaller competitor instead. Rory’s darkest line of the episode: “We make an embarrassingly large percentage of our money once every seven years when you’re in the white heat of must-acquire must-own high growth venture assets… That’s probably not this year.”
- Top exits grew sharply — but only because companies stay private longer. The VenCap (likely) analysis shows 99th-percentile exit value growing from $1.4B (2005-09) to $10.2B (2020-24), yet Luke’s read is it’s “just math”: compounding plus dispersion. If Google had IPO’d one year later, the record 2004 exit would have been ~$140B, not $23B. The actionable version: statistically 80% of the time you should sell at the last round price, but the 20% covers everything — and Luke’s new seed-manager rule is “at 2 billion, sell unless you’re 100% sure” it’s a SpaceX.
- Rippling wins the Deel lawsuit — and the counterclaims signal it. Luke puts it at 100%; the others agree Rippling likely prevails if it goes to court. Luke’s litigation mechanic: counterclaims work as damage offsets even when they couldn’t be brought alone, so “all those counterclaims, ironically, are a sign they’re going to lose.” Rory’s universal litigation advice: lawyers change their tune “right around $2 million of legal expenses,” so “settle this thing by Friday” — especially with allegations that “could be interpreted in a criminal fashion.”
- The AI trade: own the anchor tenants, fear the database-ification of SaaS. Anthropic’s run rate went $1B (Q4'24) → $2B (Q1'25) with customers up 8x; OpenAI will burn “at least another 44 billion” to profitability in 2029; one growth investor is trying to buy “every employee’s options” in Anthropic with $5B of LP supply. Meanwhile if MCP really works, Box, HubSpot and Salesforce “just become databases” — though Rory says the replacement cycle runs 10+ years, and Oracle proves a database can mint 43% operating margins. Chegg — $12B to $95M — shows what happens when you’re on the wrong side.