ElevenLabs: Building an AI Sales Machine & Why We Set a 20x Sales Quota

Summary

• Outbound is dead unless it’s human. Carles has tried a large number of AI go-to-market tools and none work — “they see everything as a transaction,” and buyers can smell mass AI outreach, with outbound email response rates now “less than 0.01%.” ElevenLabs instead spent engineering headcount building in-house revenue agents — an AI SDR for inbound, an AI proposals manager scanning the web for RFPs, an AI customer-success manager drafting personalized emails — and “that has closed deals for us already.” The rule: “it’s human if it’s perceived as human.”
• The AI-native sales org is smaller, not bigger. Carles targets a 50% productivity improvement explicitly so he can hire fewer, better-paid people, pays full commission on deals AI agents close “as if a human actually closed it,” and would rather sign a million-dollar commission check than cap earnings — because “every single million has $33 million in extra valuation” for the company.
• Comp mechanics behind the 20x quota: two employees hit their entire full-year quota in February (“the Mount Olympus of sales”), commissions run 5% on everything sold with accelerators at 1.1x–1.5x above quota, and pilots pay nothing — “it’s not adding to our valuation as a company… then why should we pay it?” Quotas must be challenging but fair because good salespeople “are moved by the coin” and without the challenge “they’re just going to be slacking.”
• Customer success is a revenue function now, not “complete BS.” The Snowflake-era view (charge professional services, CS is a cost center) made sense then; today “anyone can spin up a competitor of your product in the next two days,” so CS must drive expansion, cross-sell, and retention — a pure services model “becomes a transaction.”
• Go-to-market is portfolio construction: open markets in parallel, not sequentially, because 100 competitors arrive within a month; “test 100 things to find the three, four, five, six that actually perform.” India verticalized too early and “depressed our revenues for a single quarter — an absolute disaster.” Carles scaled the revenue org from zero to over $350M in ARR; ElevenLabs could hit $1B revenue by year-end “if we were creative”; the sales team doubles from 130 toward ~250 this year.
• The customer-support paradox: Harry calls the category uninvestable (Sierra, Dacorn (likely Decagon), 16 providers raising $75M+ in 18 months); Carles wouldn’t personally invest either — yet it’s ElevenLabs’ “fastest product in terms of revenues that we’ve ever had,” and “we power all of them,” competitors included — the Nvidia position. The CIOs-and-CSOs must-buy-AI window is only 18–24 months, and he agrees “100%.”
• Substitution risk is overrated; brand is the moat. He half-believed open-source commoditization last year and changed his mind — enterprises try open models, “lose three months and then they come back.” Brand cuts enterprise sales cycles “1 million percent,” and today’s procurement blue chips are OpenAI, Anthropic, and Cursor.
• The macro call: a next wave of foundational-model companies is coming, and OpenAI, Anthropic, Google — and ElevenLabs — “will end up buying all of them… a few billion here, a few billion there.” Asked to pick: buy Anthropic at 500 over OpenAI at 8:30 — Anthropic is “spread too thin… they need to start from scratch.” “Claude is my best friend.”

Deep dive

1. Outbound is dead — ElevenLabs replaced it with in-house AI revenue agents

• Carles’s verdict on the AI SDR category is blunt: he’s tried “a fair large number” of AI go-to-market tools and “they don’t work… they see everything as a transaction.” The tools message everyone as if everyone wants a message, and recipients can perceive mass-sent AI email instantly — outbound response rates have dropped “to the lowest of any point in time… less than 0.01%.” Hence: “Outbound is dead unless you do it with humans or unless we do it humanly.”
• What he built instead took persistence — two-plus years convincing ElevenLabs to hire engineers for revenue agents, greenlit only last year. The stack: an AI SDR handling inbound (“super good results”), an AI proposals manager that scans the web for RFPs and RFIs and scores them, and an AI customer success manager living in email that reads all customer data and contract tiers, then proactively drafts emails the human CSM edits before sending — with originals, edits, and responses stored to fine-tune tone per customer, per language.
• The distinction that makes it work: “It’s human if it’s perceived as human.” Each customer gets a slightly different message in a different tone. The AI CSM “has closed deals for us already” — and Carles pays commission on AI-closed upsells “as if a human actually closed it.”
• The endgame is concentration, not headcount: his target is a 50% productivity improvement — “because then it means I can hire less people. I prefer to manage a smaller team that gets super well compensated.” Instead of doubling or 5x-ing over two years, retain only “extremely top talent that are performing at Eleven that no one else is performing.”

2. The comp system behind the 20x quota: 5% on everything, accelerators above, nothing on pilots

• Two employees had achieved the entire full-year quota in February; Carles posted in Slack that they had “reached the Mount Olympus of sales at ElevenLabs.” Harry’s pushback — didn’t you massively miss-set the quota then? — draws his quota philosophy: challenging but fair, and if people miss because it’s genuinely too high, “do the right thing and lower it or compensate them correctly.” But good salespeople “want to be the best ones… they’re moved by the coin,” and without a big challenge “they’re just going to be slacking.”
• Mechanics: 5% commission on anything you sell, then accelerators at 1.1x, 1.2x, 1.3x, 1.5x and so on above quota, plus spiffs on individual products the company wants pushed that quarter. His caveat: over-incentivize and “you create the wrong behavior, which is people try to sell whatever it is to unlock it.”
• No commissions on pilots — “it’s not adding to our valuation as a company… then why should we pay it?” Payouts trigger on signed annual contracts; anything expanded over the next 12 months earns commission, and designated strategic accounts (top 20–30 in a market) pay for two years. The math justifying generosity: “every single million has $33 million in extra valuation for the company… if we sign a million-dollar commission check, I’m the happiest person alive.”

3. Customer success is a money-generation function — the Snowflake view was era-specific

• Harry invokes Chris Dagnall (Snowflake) — customer success is “complete BS,” charge professional services. Carles’s rebuttal is era-specific: “In the world where Snowflake grew up, that made sense. In the world we are in today with AI, that doesn’t make any sense. Because anyone can spin up a competitor of your product in the next two days.” Close the first contract as fast and deep as possible; CS is what expands and retains it. A pure pay-per-penny services model “becomes a transaction. You’re not building a community, you’re not retaining for the long term.”
• On the classic warning — never let your farmer face someone else’s hunter — he agrees hunters are vital but adds the failure mode: “a hunter that is not well managed will create more damage than opportunities.” Hit the sub-companies of one holding with unaligned pricing and value stories, and each discrepancy “ends up coming back to hunt you down” at renewal. Hunters need to be kept in check together with customer success.

4. Go-to-market is portfolio construction — parallel markets, whales plus liquidity

• The traditional VC advice — win one market deeply, then move to the next — “doesn’t work” anymore: “you’re going to have 100 competitors in the next month.” His frame: “go-to-market is similar to investing in venture capital… you need to test 100 things to find the three, four, five, six that actually perform” — markets, channels, self-service, resellers, cloud ecosystem, grants, affiliates. Forecasting in this mode is “just impossible”: “I only need one of those to work really well to give me another 100 million dollars in revenues.”
• He keeps a weekly calendar block called “pipeline construction”: each account exec needs the ability to close big whales in enterprise and enough liquidity to close deals every week — “because people were losing the confidence.”
• India is the cautionary tale: verticalizing the sales team too early “depressed our revenues for a single quarter and it was an absolute disaster” — people weren’t passionate enough, sales cycles were long, the team was too small. They reset to horizontal with negotiated named-account lists, and “people started closing deals in the first month… it was day and night. We had changed the culture in the team in the local market by getting it wrong.”
• Government is the deliberate slow bet: sticky, a societal benefit (“I don’t care if I don’t make enough money to justify it or I lose money”), and hard to enter — “the moment you actually get in, you never leave that industry.” Failed bets get owned too: media & entertainment studios didn’t work early on (“the industry might not be ready”), so ElevenLabs pivoted the ICP to media-creation platforms, then bet early on agentic systems — now “fantastically huge.” On $1B revenue by year-end: “Who knows… if we were creative we can do it. If not, it will be a little bit later. That’s fine.”

5. The customer-support paradox: uninvestable as a startup — and ElevenLabs’ fastest product ever

• Harry’s thesis: customer support is uninvestable — Sierra and Dacorn (likely Decagon) own the brand and funding, “16 providers have raised over 75 million in the last 18 months,” incumbents from Salesforce to Zendesk crowd in, and OpenAI may enter. Carles half-agrees: “I would not personally invest” — but for established, fast-growing companies moving into the space, the opportunity is big.
• ElevenLabs is exhibit A: “the majority of our customers start with customer support and we make an absolute ton of money on that… it’s the fastest product in terms of revenues that we’ve ever had. It’s just insane.” The kicker: “we power all of them” — the competitors too, plus the API foundational-model layer. He names the position himself: like Nvidia, powering everyone while competing with everyone.
• Before launching the agents product, he personally called the biggest agent platforms running on ElevenLabs: “Guys, FYI, in the next couple of months we’re going to be launching our agents product. We’re going to be competing with you.” Every founder said “welcome on board.” Transparency plus deal-by-deal coexistence “is when you end up creating a good ecosystem.” And he agrees “100%” with Harry’s urgency: an 18–24 month window while CIOs and CSOs say they have to have AI — “everyone is selling agents left and right. We’ve had the best quarter ever.”
• On powering Hollywood’s decline: he rejects the framing — “Hollywood will come out even stronger.” The broken part is the model, not the tech: a big-budget production is like a startup demanding “$100 million pre-seed just for this idea” with five years to ship. AI voices and video let studios launch ideas cheaply, prove engagement — he pitches a cost-per-engagement framework — “and if it works well, let’s scale it. Then you have the fifty-million-dollar budget.”

6. Substitution risk is overrated — hassle is the moat, brand is the accelerant

• Jason Lampkin’s “year of substitution” thesis — great product, gets expensive, swap for the cheaper 80% version — gets a flat “No.” Carles admits he changed his mind here: last year he half-believed open-source models would commoditize voice. What he realized: even at equal quality, enterprises must maintain, scale, and keep funding open models when priorities shift. “You try to operationalize it and they’ve lost three months and then they come back.”
• Does brand reduce enterprise sales cycles? “Yes, 1 million percent. And whoever tells you no is just lying.” The IBM ideal — no one gets fired for buying you — has three claimants today: OpenAI, Anthropic, and Cursor. “Cursor have killed enterprise unbelievably well.” Twitter’s “no one uses them anymore, we all use Claude Code (likely)” misses the ground truth: “go into HSBC in Hong Kong, Barclays Bank in Swansea — I promise you they’re using Cursor in their local dev shop.”
• On the SaaS-pocalypse — companies building their own software — “a very large portion is overexaggerated, but there is a real fact that people will end up building their own applications.” He’d custom-build a CRM (“it’s just a pool of data”) though ElevenLabs still runs Salesforce — and “Lovable has the best CRM that I’ve ever seen. Those guys have really smashed it.” He wouldn’t rebuild Google Drive or Gmail.

7. Culture: public leaderboards, brutal offers, 20-minute hiring reads

• A bot posts per-rep quota attainment year-to-date and forecasts into the sales channel every week — anyone can see who’s bottom. He’s unbothered by demoralization: “by definition, you don’t go into sales if you’re not willing to take the harsh feedback.” But the number isn’t the whole read — someone building strategic accounts or real pipeline gets patience; the true tell is “if I open a calendar and someone has a full calendar that is empty.” His view on the profession: a very large portion of reps today are softer — “they just want a normal job. And that’s okay… it is not what I’m hiring for.”
• Every offer carries the warning: “ElevenLabs is going to be extremely difficult… You’re going to work a huge amount of hours. I expect full commitment.” The self-filtering is why sales churn “is actually not that big” — and why he believes his people genuinely are 4x better, against Harry’s direct challenge: “I do 100% believe it. I would take them to war, every single one of them.” The team earns it by challenging him daily — which market to open, whether 20x is right, whether to sell to competitors — “it forces the entire business to rethink our position every single day.”
• Hiring is fast-twitch: high energy, sharp, clearly researched — or the quieter deep type, “less obvious, but once they do, it’s just insanely good.” “I don’t need to spend more than 20 minutes to know if I want to hire someone” — he writes his feedback into Ashby mid-conversation, and hates recording interviews because “people behave differently if they think they’re being recorded.” Sales calls are the opposite: Gong mandatory, auto-filling Salesforce. The scaling risk in doubling the team from 130 (adding ~120, kept under 250): dilution — “if you’re not up front about the expectation, people come with different expectations.” He nods to Marc Andreessen’s claim that companies are overstaffed by 25%+, but warns understaffed reps with too much pipeline “end up not closing nothing.”

8. Budget is never the constraint — proof is

• “If you want to do something, budget is never going to be a problem. And that’s been since the beginning.” The real gate is evidence: take $50k–200k and prove it on the side — “don’t ask me for a million dollars, don’t ask me to dedicate an entire team… but if you prove me you can make it work, we’ll scale it and we’ll give you all of the budget that you want.”
• The F1 sponsorship (Audi) was the exception that took long deliberation with no data. His branding rule: premium branding on the worst team beats secondary branding on the #1 — more exposure, more creative freedom, and it’s a bet: “every single sports championship Audi has entered, they’ve ended up winning within five years.” Building the car and engine from scratch against the incumbents “is who we are as a company” — and who Revolut is.
• Marketing ROI ranking: exec dinners win — $3,000–5,000 for ~15 people, and inviting competing buyers from the same ICP manufactures FOMO in the room (“they know each other… that always works really well”). Conferences have no ROI; most trade shows don’t have good ROI. The conclusion: build your own events — the ElevenLabs Summit in London — with content that is deliberately “not salesy”: Mati does the keynote, then partners (BCG, NTT) take the stage. “It’s less about ElevenLabs, it’s about them… you bring partners to validate that what you’re doing is right.”

9. CVCs as a distribution weapon; language is the tax on going global

• His controversial favorite channel: corporate VCs — Woven Capital (Toyota), Deutsche Telekom’s T-Capital, Telefonica, NTT DOCOMO Ventures, Liberty Global on the cap table. “They help you navigate big brands. They are your champions internally.” The structure has teeth: allocation is tied to pipeline — “for every million dollars you want to invest, you need to bring X amount of revenues in the next 12 or 24 months” — with penalties if it doesn’t happen: “We buy you out.” The win-win: they help close contracts, their stake appreciates, and ElevenLabs “learned everything about the telco industry” — and automotive — from the inside.
• Partner ecosystems generally: not a silver bullet. “If you’re thinking you’re going to be making another 20% of revenues in two quarters, you’re wrong.” They take time, dedicated people, and incentives on both sides; track SQLs first, then migrate the metric to revenue per SQL. Salesforce is “the reference” — an ecosystem that both receives deals and sends them, even billing through partners in markets it can’t bill directly.
• The internationalization mea culpa — “entirely my fault”: believing the whole world signs English-only contracts. Less than 5% of Latin America speaks English; France won’t sign without French-law contracts. He opened Japan himself without speaking Japanese, flying out constantly for two years before hiring rep Sang Wong and GM Jim — “really hard,” echoing UiPath’s experience. Korea got a proper thesis: you must sell in Korean, but the content produced targets international markets, so map the languages Korean producers care about.

10. The investor lens: be helpful, test GTM hunger — and buy Anthropic over OpenAI

• Should operators invest simultaneously? He initially says “Yes” repeatedly, then qualifies that it is not always so — founders want backing from the best operators. Harry’s pushback is sharp: you’ve promised your employer your best efforts; if one of his own sellers were reading about drone cost curves in the bath, “you’d be like, dude, your turf is France and it’s media.” Carles’s answer: it works only if “you prove every single day that you are the hardest working person in the office.” His advice to operator-angels: “Be helpful. The money that you’re investing is not worth anything for any company” — the #1 ask from portfolio companies is go-to-market and sales hiring.
• His biggest investing mistake: backing founders hungry to build product but “not hungry to iterate on the go-to-market.” Now he tests it in the first conversation — a founder who says “I’m only thinking about building research” gets “I love your product… it’s just not for me.” Slow burners are still fine (his Linear-style cases): “not everyone’s market grows at the same pace.” Forced to go all-in on one non-ElevenLabs portfolio company: Theker Robotics in Spain — “the most geniusest guys that I’ve seen in robotics ever.”
• The forward call: “I am actually looking forward to the next wave of foundational model companies” — and he expects OpenAI, Anthropic, Google, and ElevenLabs itself to “end up buying all of the new foundational model companies… paying just a few billion here, a few billion there and swapping them in.” Buy OpenAI at 8:30 or Anthropic at 500? “Anthropic. And I put it online multiple times.” Anthropic is “spread too thin… they need to start from scratch” — and he flags the same stretched-too-thin risk for ElevenLabs. Personally: “Claude is my best friend. I use it all the time” — funds, ElevenLabs, everything.
• The Mati Staniszewski (likely) decision framework he most values: listen, decide, “acknowledge if he got something wrong and then change it immediately.” The ratio: “you should be getting it right 99% in the 5% of things that are really mega important. In the 95% remaining, you should be getting it mostly wrong.” Happiness in 2026: three unicorns in the fund and ElevenLabs crossing $1B in revenues — with a longer-run target of seven unicorns (“David Beckham… he was number seven”).