ElevenLabs

Use credits for creators and simple plans for consumers. Creators on the ElevenLabs app already think in credits and characters, so it keeps about five credit tiers starting at $5 a month, while a parent playing fan fiction at night should not think about tokens.

Both apps run on the same costly API; real AI cost of goods still exists but should not be how you talk to consumers. Phil Carter cites ElevenLabs among AI companies using multiple tiers to protect against compute costs.

Subversive▶ 25:50

Every launch starts as a Twitter thread before anything is built. Teams write the core value promise, turn it into a thread with a headline and hooks, then into a video of motion graphics or people on camera. If a feature makes a poor thread, "you better have a really good reason" to build it.

David Barnard reports ElevenLabs writes the tweet before building the product, which Phil Carter likens to an Amazon launch memo for short attention spans. Tanmay ran his first launch about six weeks after joining with just a doc, with scripts reviewed by the creative team and threads by the head of growth.

Subversive▶ 27:40

ElevenLabs voice can multiply ads that feature real people. Phil Carter recommends using ElevenLabs to translate one real spokesperson into 30 languages as part of testing hundreds of ad creatives a month.

He warns against fully AI-generated ads because of quality, FTC risk on fake testimonials and backlash.

Subversive▶ 13:00

ElevenLabs builds apps so it becomes a product and brand, not just an API. CEO Mati Staniszewski's view is that a generational company builds the product layer on its own research, with a brand both B2B and B2C users know. So ElevenLabs launched first-party apps: ElevenReader for listeners and the ElevenLabs app for creators.

The company began with text-to-speech, voice clones and dubbing, and now builds speech-to-text, sound effects and music models plus a voice agent platform. Jack McDermott expects more labs to do the same, citing Claude Cowork built from Claude Code and ChatGPT Health. "From early on, the aim was to go beyond being just an API that was called upon."

Sub Club▶ 48:10

App teams are the heaviest API users, so their feedback improves the models. Internal product teams are among the most intensive consumers of ElevenLabs' APIs, and their data on latency, adoption and developer experience flows back to the model teams. In return they get early model access and an effectively unlimited compute budget to test with.

McDermott calls consumer work playing on hard mode with the lab as a tailwind. Owning the models also lets ElevenReader offer voices cheaper and across more languages than Audible or Speechify, and voice gains for a receptionist or a Duolingo tutor carry over to a bedtime story. The main cost is focus across consumer apps and large enterprises at once.

Sub Club▶ 43:31

The creator app was built after search demand landed on the wrong product. Before the ElevenLabs app existed, people searching for ElevenLabs text to speech landed on ElevenReader and were confused. Building a creator app captured that demand and clarified what a text-to-speech creator wants versus a consumer.

The creator app makes voice, video, image and sound effects on a phone, creators use it to grow audiences on TikTok and Instagram, and it is growing extremely quickly. No figures are given.

Sub Club▶ 55:40

Users stuffing a demo box with manuscripts revealed the long-form audio niche. People pasted hundreds of thousands of characters, often their own manuscripts, into a 200-character text-to-speech demo box. That pointed ElevenReader, launched in summer 2024, at fan fiction, web novels and other long-form stories with no professional audiobook.

Fewer than 5% of ebooks are recorded as audiobooks, and far fewer in long-tail languages. No figures are given for Reader's users, revenue or retention.

Sub Club▶ 15:30

Iconic and locally familiar voices give listeners a reason to choose Reader. ElevenReader signed paid, exclusive partnerships with Michael Caine, Maya Angelou and Judy Garland and their estates, and signed the Polish voice of the Harry Potter audiobooks, which shot to the top of the charts with Polish users.

Poland is a market where the company has deep roots. "Voice is really a personal and an intimate experience."

Sub Club▶ 4:10

Test long listening sessions, because voice models can drift. Burt Reynolds' voice morphed into a demonic sound after half an hour or more of playback, and the clips went viral in the community before the team fixed it.

"We've worked really hard to keep the demons at bay since then."

Sub Club▶ 7:20

ElevenReader launched on Reddit and Discord word of mouth before adding paid. The team shipped the product and leaned heavily on viral and organic growth through Reddit and Discord, then layered on partnerships and paid as traction appeared, growing from four developers and one designer into a fuller growth team.

Paid is treated as an amplifier of brand and organic work, not the foundation.

Sub Club▶ 16:27

Reach book communities with UGC, not polished AI-lab ads. To reach romance, fiction and fan fiction listeners, ElevenReader ran paid and influencer campaigns on BookTok and Bookstagram, leading with UGC rather than the polished style typical of an AI lab.

These communities already held latent demand and active conversations about audiobooks. McDermott calls the UGC approach key in the early stages.

Sub Club▶ 13:30

A first-of-its-kind partnership can create a launch moment. ElevenReader made Melania Trump's ebook available in her voice in English, Spanish, Portuguese and Hindi, which McDermott calls probably a first of its kind in audiobooks.

He acknowledges it was controversial but valuable as a distinctive launch moment.

Sub Club▶ 24:00

Collapsing ElevenReader's paid tiers into one unlimited plan was the year's biggest win. ElevenReader started with a free tier, a $5 to $10 mid-tier and a heavy-use tier. Tests and interviews showed consumers see price as free or paid, cheap or expensive, judged against Spotify and Netflix, so the team cut to one free plan and one unlimited ultra plan priced around hours of listening.

Dropping a whole tier and switching from credits to hours of listening was called the biggest win of the year for ElevenReader, though no conversion or revenue figures are given. "Consumers tend to view pricing much more in a binary way."

Sub Club▶ 54:30

Use credits for creators and simple plans for consumers. Creators on the ElevenLabs app already think in credits and characters, so it keeps about five credit tiers starting at $5 a month, while a parent playing fan fiction at night should not think about tokens.

Both apps run on the same costly API; real AI cost of goods still exists but should not be how you talk to consumers. Phil Carter cites ElevenLabs among AI companies using multiple tiers to protect against compute costs.

Sub Club▶ 19:30

Keep the best voices on the free plan and charge for time. ElevenReader does not downgrade voice quality for free users; hundreds of thousands use the free plan, and the paid plan allows up to 24 hours a day of listening.

The team evolved into paid plans rather than gating quality first. "We don't downgrade your voice experience."

Sub Club▶ 18:00

Mobile gets one opinionated plan while web keeps many. On web ElevenLabs sells upwards of five or six plans from $5 to a couple thousand dollars; on mobile it offers one core subscription, with pay-as-you-go rolling out, and tests incentives around that single price.

Downgrades, discounts, expirations and syncing plans across Apple, Google and Stripe are hard, and a small screen rewards a clear offer. The same rule applies to features: bring only what resonates on web rather than squishing the whole product onto a phone.

Sub Club▶ 58:10

ElevenLabs runs as 10 to 12 small startups that ship fast. Self-contained pods of six to eight people hold development, design, growth and product for one product, and ideas and tests are expected to ship the same day. The mobile team is about 12 people in a company of about 300 heading to 400.

Shared roles for brand, SEO, ops and web set standards so pods don't reinvent them. The downside is coordination and coherence across teams as the company grows into many products.

Sub Club▶ 36:00

Benchmarks like a one-week reverse trial did not hold for ElevenReader. McDermott says steady optimization from onboarding through the trial offer has delivered compounding wins, but many conventional benchmarks failed, so the team tests from first principles.

"I've been surprised by as many things that I've had to unlearn... as those that have been confirmed."

Sub Club▶ 36:57

ElevenLabs builds apps so it becomes a product and brand, not just an API. CEO Mati Staniszewski's view is that a generational company builds the product layer on its own research, with a brand both B2B and B2C users know. So ElevenLabs launched first-party apps: ElevenReader for listeners and the ElevenLabs app for creators.

The company began with text-to-speech, voice clones and dubbing, and now builds speech-to-text, sound effects and music models plus a voice agent platform. Jack McDermott expects more labs to do the same, citing Claude Cowork built from Claude Code and ChatGPT Health. "From early on, the aim was to go beyond being just an API that was called upon."

Sub Club

App teams are the heaviest API users, so their feedback improves the models. Internal product teams are among the most intensive consumers of ElevenLabs' APIs, and their data on latency, adoption and developer experience flows back to the model teams. In return they get early model access and an effectively unlimited compute budget to test with.

McDermott calls consumer work playing on hard mode with the lab as a tailwind. Owning the models also lets ElevenReader offer voices cheaper and across more languages than Audible or Speechify, and voice gains for a receptionist or a Duolingo tutor carry over to a bedtime story. The main cost is focus across consumer apps and large enterprises at once.

Sub Club

Spin out a separate app only when its audience needs its own brand. Tanmay Jain saw Canva Stories win Apple awards yet never crack acquisition, retention or brand awareness, because users wanted Canva and it was folded back in. ElevenLabs runs a horizontal platform, so the flagship app and ElevenReader keep separate brands while sharing a design system, tech stack, developers, MMP event mapping and UGC learnings.

Two apps give roughly double or quadruple the surface area for learning, such as running paywall tests in parallel. Separate apps over a super app also let each price in its own users' terms. A branded store search for "ElevenLabs" deliberately leads somewhere different from a search for audiobook apps.

Sub Club▶ 56:05

The moat is speed, ecosystem and brand, not model quality alone. Expecting large labs to ship good-enough voice models, ElevenLabs doubles down on voice, adds platform depth like custom evals, workflows and integrations in voice agents, and compounds earned media so it is the first name people think of for voice.

"Every time someone thinks about voice or speech and audio, 11 Labs is the name that floats to mind immediately."

Sub Club▶ 58:46

The aha is bringing your own file and picking a voice. ElevenReader's aha is uploading your own ebook, EPUB or PDF and choosing the voice to hear it in, so the team skipped goal-setting quizzes and got users to a voice quickly.

They found the aha by watching Reddit, Discord and X, where users share screenshots of surprising voices reading their content.

Sub Club▶ 32:30

A catalog of 600+ voices works because voice preference is a long tail. ElevenReader offers over 600 voices where most apps offer none or two or three. No voice has more than 10 to 12% of use, and the median voice gets about 1 to 2%.

Selection itself is a value driver and a shareable moment.

Sub Club▶ 49:50

Collapsing ElevenReader's paid tiers into one unlimited plan was the year's biggest win. ElevenReader started with a free tier, a $5 to $10 mid-tier and a heavy-use tier. Tests and interviews showed consumers see price as free or paid, cheap or expensive, judged against Spotify and Netflix, so the team cut to one free plan and one unlimited ultra plan priced around hours of listening.

Dropping a whole tier and switching from credits to hours of listening was called the biggest win of the year for ElevenReader, though no conversion or revenue figures are given. "Consumers tend to view pricing much more in a binary way."

Sub Club▶ 8:16

Use credits for creators and simple plans for consumers. Creators on the ElevenLabs app already think in credits and characters, so it keeps about five credit tiers starting at $5 a month, while a parent playing fan fiction at night should not think about tokens.

Both apps run on the same costly API; real AI cost of goods still exists but should not be how you talk to consumers. Phil Carter cites ElevenLabs among AI companies using multiple tiers to protect against compute costs.

Sub Club▶ 8:16

Mobile gets one opinionated plan while web keeps many. On web ElevenLabs sells upwards of five or six plans from $5 to a couple thousand dollars; on mobile it offers one core subscription, with pay-as-you-go rolling out, and tests incentives around that single price.

Downgrades, discounts, expirations and syncing plans across Apple, Google and Stripe are hard, and a small screen rewards a clear offer. The same rule applies to features: bring only what resonates on web rather than squishing the whole product onto a phone.

Sub Club▶ 53:10

Cutting price will not fix retention if users don't care. From his founder days, Tanmay Jain learned that dropping prices after churn may keep users a month longer, but people who don't value the product leave anyway, while those who do will pay a lot.

Sub Club▶ 1:02:28

Every launch starts as a Twitter thread before anything is built. Teams write the core value promise, turn it into a thread with a headline and hooks, then into a video of motion graphics or people on camera. If a feature makes a poor thread, "you better have a really good reason" to build it.

David Barnard reports ElevenLabs writes the tweet before building the product, which Phil Carter likens to an Amazon launch memo for short attention spans. Tanmay ran his first launch about six weeks after joining with just a doc, with scripts reviewed by the creative team and threads by the head of growth.

Sub Club▶ 18:35

Spend one to two weeks refining the story, because nobody else will tell it. With a new state-of-the-art benchmark every week, ElevenLabs spends one to two weeks on its storyline before building launch assets, and keeps an in-house creator and motion design team to produce them.

App Store and Play Store featuring is welcome but not the main lever. At Canva, a video launch took a 12-month run-up and three months of messaging aimed at one date; ElevenLabs instead launches fast and often.

Sub Club▶ 21:19

Monthly launches build earned media that makes paid cheaper. Each launch earns more eyeballs for the next, fills branded search (the first place paid starts) and makes ad viewers more likely to convert. Tanmay is confident it lowers CPA and CPI but admits it has not been measured.

Sub Club▶ 25:44

Run launches and steady paid UA as two parallel tracks. Teams launch roughly monthly for viral upside while paid runs as a forecastable engine on Meta, TikTok, Apple Search Ads, Google Play and Google Ads, each with its own product, message and model.

Brand built by launches means paid need not hit day zero, day six or week-one payback, so the team can test new creatives and accept higher CAC. Each launch is matched to the channels that fit it; LinkedIn may not matter for a short-form creator app.

Sub Club▶ 29:27

Build from what only you can do, not "X for Y" copies. Features copied one-for-one from other products on a "wouldn't it be cool if" basis tended to fail, while ideas rooted in unique capabilities compounded.

Still bias toward action, since a launch that looks gimmicky in hindsight still yields information.

Sub Club▶ 33:56

ElevenLabs runs as 10 to 12 small startups that ship fast. Self-contained pods of six to eight people hold development, design, growth and product for one product, and ideas and tests are expected to ship the same day. The mobile team is about 12 people in a company of about 300 heading to 400.

Shared roles for brand, SEO, ops and web set standards so pods don't reinvent them. The downside is coordination and coherence across teams as the company grows into many products.

Sub Club▶ 44:51

Hire engineers who build side projects and want end-to-end ownership. ElevenLabs screens public GitHub profiles for side projects and obsession with a user pain point, and recruits from companies with strong product cultures, then pairs them with former founders.

Tanmay left big-company PR and legal reviews for a role where he merges PRs, reviews Figma designs and ships paywall experiments himself.

Sub Club▶ 37:43

Test a plain sign-up baseline before tuning the onboarding quiz. The team's biggest fail was spending many cycles on onboarding quizzes and goal setting without testing a sign-up straight into the product. Simpler onboarding with fewer screens to the aha proved the bigger win.

Tanmay's suggestion: "just strip it out entirely for 10% for 5 days and see what's the opportunity size."

Sub Club▶ 13:53

Find your product's natural shareable moment rather than copying Wrapped. Wrapped worked by turning passive consumption into an identity-linked moment; ElevenLabs looks for milestones users are proud of and expects several iterations, since if people don't share, it isn't a shareable moment.

Examples given include the GitHub skyline and YouTube 100k. In ElevenReader, users already share screenshots of voices reading their own files.

Sub Club▶ 36:40

ElevenLabs still runs classic go-to-market and is 50% enterprise. Jason Lemkin cites ElevenLabs as a fast grower that still runs classic go-to-market plays and is already at 50% enterprise sales.

Lenny's Podcast