Pocket FM Launches Fiction Writing AI Sherpa, Aiming to Enable Solo Studios to Produce Market-Ready Audio Dramas
Pocket FM co-founder and CEO Rohan Nayak announced the launch of the serialized writing system Sherpa, aiming to enable solo studios to produce market-ready audio dramas. He also raised the platform's annual revenue run rate from approximately $250 million to about $500 million, claiming a year-on-year content production increase of around 1200%.
The company states that Sherpa is trained using approximately 5.5 billion hours of playback and minute retention data; over 550,000 creators produce about 2.6 million hours of content annually. Nayak outlined the process in seven steps: a one-sentence idea is questioned for tension and psychology, then the worldview and characters are expanded, followed by breaking down seasons, arcs, episodes, and scenes, drafting each scene, using an editing model to control pacing and dropouts, one-click voiceover publishing to Pocket FM, and then revenue sharing based on performance with automatic localization. He noted that general large models lack long-form narrative rewards, while Sherpa utilizes character state knowledge graphs, layered planning, dynamic re-planning, and a prose engine based on proprietary stories and retention; the evaluation model aligns with human judgment about 80-90% of the time, with a checklist of about 40 items.
The revenue structure has been broken down. The annualized $500 million is calculated by multiplying monthly revenue by 12, not based on locked subscription contracts. TechCrunch reported that about $415 million comes from unlocking individual episodes, and about $85 million from advertising. Approximately 93% of the catalog is AI-supported, with about 99% of new content produced through AI. About 96 works have individual revenues exceeding $1 million, with 13 exceeding $10 million. Forbes estimates that top dramas approach a $90 million scale; the company claims at least five creators have cumulative earnings exceeding $1 million. Nayak stated that 10% of qualified authors earn over $200,000 annually, with three becoming millionaires within two years, and a hit drama contributing over $100 million in revenue.
The production side has shifted from contracted writers to user-generated content. The company previously canceled about 200 writer contracts, reducing production costs to about $30 per hour. Co-founder Prateek Dixit stated that Sherpa is not expected to be profitable in the short term, as it fills the gap where general models can write one or two episodes but fail to scale to hundreds. Paid retention has been managed from about 35% to around 60%. The U.S. remains the primary cash market, while localization in the UK, Germany, and France has reduced the launch cycle from about 12 months to approximately 2 months.
Funding and profits are still tied to old valuations. Founded in 2018 by Nayak, Dixit, and Nishanth KS in Bangalore, the company has raised approximately $196.5 million, with a Series D round of $103 million in March 2024, valuing the company at about $750 million, with investors including Lightspeed, StepStone, Tencent, and Times Internet. The current EBITDA is about 5%, with a target of 15% to 20%, and negotiations for a new round of funding of about $100 million to $120 million at a valuation of about $2 billion are underway. Nayak stated there is no pressure for financing. The parent company Pocket Entertainment is also launching Pocket Saga short dramas and extending audio IP into books, television, and film licensing.
Who is buying and who is selling: Buyers are paying for addictive listening hours on a per-episode basis, while sellers offer an infinitely replicable serialized inventory. The event is driven by the launch of Sherpa, with funds flowing from in-app purchases to computing power, voiceover, and customer acquisition advertising. Beneficiaries are platforms that control distribution and retention data, along with a few algorithmically selected authors; those under pressure are contracted writers submitting episodic content and general large models lacking proprietary playback data. The flywheel is written as: more episodes lead to more dropout coordinates, which retrain Sherpa, attracting more creators.
In publicly audited figures, management mentioned an upcoming disclosure of fiscal year revenue of about $324 million, which does not align with the $500 million annual run rate. There are no corresponding settlements on-chain; all increments are recorded in in-app purchases and advertising inventory.
Source: Public Information
ABAB AI Insight
Pocket FM did not start from a "better model" but from the lack of serialized audio content during subway commutes. The three founders first validated per-episode charging in the Indian market, and in 2022, they rebranded the same formula to enter the U.S. market, with Insta Millionaire and Saving Nora proving that a few cents per episode can generate million-dollar IPs. With the arrival of AI, they are not replacing Netflix but writing the dropout curve that has already been validated into training labels.
The capital path is: using Tencent and Lightspeed's money to acquire U.S. users, then using every minute of fast-forwarding and payment from users to retrain writing agents. Canceling a large number of contracts and reducing production costs to $30 per hour transforms fixed human costs into variable computing costs. Sherpa is open for free to turn global creativity into its own catalog, rather than to sell software subscriptions. If the next round of financing is at a $2 billion valuation, checks will continue to flow towards models and short drama formats, rather than traditional film sets.
The benchmark is not Audible's audiobook shelf but rather a combination of Chinese web literature platforms and short drama distribution, layered with proprietary models. Wattpad hands text over to Hollywood, while Pocket keeps text within its own player and charges per episode. The industry phase has shifted from content expansion to data control: those who have minute-by-minute retention can reward long-form fiction. What general models lack is not writing ability, but the ability to recall secrets buried in episode 12 in episode 317.
Structural judgment belongs to the overlay of technical substitution and the transfer of pricing power. The mechanism is: the economic value of serialization lies in retention rather than the literary quality of individual episodes; once dropout coordinates can be fed back into the model, creative rights shift from studio composition to platform arrangement. A solo studio can produce a hit because distribution, voiceover, translation, and advertising materials have all been captured by the same data loop. The result is not the disappearance of authors, but a steeper distribution of author income: a few who hit the algorithm take home millions, while the supply side expands to 550,000.
ABAB News · Cognitive Laws
To have distribution is to have training data
Charging per episode nurtures the model, and the model in turn nurtures per-episode charging
General models can write, but retention models can write the next episode