Speechify

“So we built a deep learning-based text-to-speech model in 2015 that now reads more than ten million books worth of words to people every year.”

“If you need a wheelchair, you're not going to churn from your wheelchair in the same way that people don't churn from Speechify.”

“Terrible answer. Never would hire that person, not in a million years.”

“But people typically leave not for money. They leave for another thing in their life that's not satisfying them. Because they're, for whatever reason, not satisfied. Figure out what's not satisfying and solve that thing for the purpose. And sometimes it's mone…”

“Now that is exactly what we're doing now, but like front deployed engineers.”

“The things that really move the company are on the other side of five hours of grappling with a hard problem.”

“The places where meetings get Like, become a core of a company is companies that don't know their direction, and so the meetings are used to try and find the direction.”

“The amount that we lost was like a couple of 100,000 dollars, but it was like not crazy.”

“15% of our users come organically from ChatGPT.”

“Every day we upload 1300 to our testing campaigns.”

“Women convert a lot better than men as like audiences. And so right now it's a more, we have more women who use the product than men by like a good margin at the moment.”

“We have 94% of the B to C voice agents market.”

“Coach. So, One thing that we do a lot in Speechify is coach people on how to use AI better. Force them to use AI better.”

“We have more than fifty million people who use Speechify now, the organic B to C product that makes it really easy to read”

“Speechify just finished four and a half years of running profitable.”

“We read more to people than any other AI product in the world.”

“We had two people that we hired from Apple and Snapchat, like very legit people. And they messed up the culture of the company. They would leave at 5:30 p.m. They like didn't live and breathe the company. And in the end, both of them quit after three months.”

“And what we found is it takes about a hundred iterations on a feature before the feature gets adopted. But once it gets adopted really well. So you need to do the same education for your B to C users as you need to do internally.”

“Typically we found, this is my personal experience, when we hire folks from Google, they come from a place of a ton of abundance. So the hunger is not the same hunger. And you need to coach them out of essentially bad habits. And that can infest your culture i…”

“We've only had one real case of attrition at Speechify of someone who, like, worked with us for a long time, and that person went to work for Elon Musk directly.”

“The one that I did not expect to work that worked amazing is I wrote a rap song about my experience with dyslexia and ADHD, and that thing made us like three million, also like three million dollars on a rap song.”

“Companies go through cutting and bulking cycles like bodybuilders do, and there's a reason why cutting and bulking makes sense, because you can't oscillate between the two in one week. You have to commit for like six months to doing one.”

“And we have a principle even inside of engineering, which is you've got to respond to messages within 60 seconds because you want to unblock people.”

“So always at a deadline. And it goes on a calendar with multiple people on the invite. So I'm going to ship this to production by, not Tuesday, by three PM on Tuesday. Invite Rohan Simon-Cliff. Pankaj. So everybody knows. And then sometimes we have calendar in…”

“And my rule is I will read a hundred books about the topic, listen to them with Speechify. I will identify a hundred experts and I'll go and have a conversation with them.”

“And that's why the company has been profitable for so long.”

“Just have a habit of your team sends you their top three priorities every single morning. You don't respond unless the direction is bad, in which case you tell them this is the correct direction to go.”

“So before letting them go, I'd actually rather put them in QA for a little bit, see how well they do in QA. And if they do well there, that tells me something. And we should not get rid of that person. Like, we should just coach them properly. But if they're a…”

“We're one of the 200 companies that's provisioned to test ads on OpenAI.”

“Sometimes people tell you that they're great at AI and then you give them the challenge and they suck. You just got a test. And sometimes people have never used AI before, and you force them to sit with it for 48 hours, and by the end of the 48 hours, they're …”

“And so we have a rule if you ship a PR, To a repo that your team is not responsible for, it gets fast tracked and that team has to very quickly respond because I want to encourage people to learn how to contribute to other code bases.”

“For example, for us, PDF is a complicated concept. Book is not a complicated concept. So our best performing ads include books, not PDFs.”

“One, we try to run ads on Snap all the time. We never succeed.”

“I'm okay spending 500,000 dollars a month making 20,000 pieces of creative because that is a creative investment.”

“Last year we had a 178,000 people who joined, who applied to work at the open engineering positions at Speechify.”

“We test almost a thousand data generated ads a day right now.”

“I have a bent towards older hires. Because they read textbooks about software engineering When they see what's coming out of the LLM output, they understand it.”

“Like you should be the strongest individual contributor in your team if you're leading that team. If you need a people manager, we'll give you a people manager to manage the communication. But I want you mainly hands on keyboard coding, designing, talking to u…”

“We have an asynchronous technical challenge. 19,800 people finished that challenge.”

“Because it's an arbitrage game, if you use the tool that everybody else uses, typically you're not going to win.”

“A prerequisite is we have like this entrance exam. There's this gigantic code repo that you need to read very fast. You can't read it with your eyes. You have to use AI. You have to use LLMs to ship all the features, and then you get a test at the end, which i…”

“By four and a half years in, yeah, we were making more than five million dollars at that point.”

“I made sure that everybody in our company, no matter how senior they got, always were still not just a fat general sitting at the back, but a warrior who could take out their sword and actually hands on keyboard by the ads.”

“We're getting to the point where soon we're going to spend more on tokens than we spend on actual salaries.”

“For the like standard model is anywhere from 70 to a hundred bucks. And for the V three fast model, it's like 30 bucks.”

“And we got it to the point of single digit dollars for a million characters.”

“It took us four and a half years to find real product market fit.”

“When we were 21 people at Speechify, 18 of the folks at the company were previously either CEO, CTO, or VP of engineering at the last company.”

“The newest Speechify Simba 3.2 model is ranked number one in the world for quality. Above all the frontier labs, 10 x more affordable and stuff like 11 Labs.”

“Like I have one of my best engineers right now is not even writing models. He's making synthetic data sets to train models.”

“At Speechify, we still use K-eighty's for a lot of specific operations for inference, and we use older models of GPUs constantly.”

“They have to be able to orchestrate agents well, and if they're not doing that, It's kind of not worth to have the person.”

“We will let people go if they just go Bananas with something for no reason.”

“So we'll buy multiple racks of Rubens. And on top of that, we'll buy B 300, which are like the newest form of Blackwells because we can get them earlier.”

“The realization moment was when we realized that we had really talented engineers who were essentially moving at one seventh of the speed they could have if they had to compute one to one with their creativity and ideas.”

“11 charges 10, a hundred dollars per million characters. The OpenAI model on the benchmarks cost a 196 dollars per million characters. So ours, when we sell it to other B to B companies now, we just launched our API, Simba, 3.2, it costs 10 dollars per million…”

“11 Labs, huge credit to them, leapfrogged us because they sell to B to B. Well, historically we've only sold to B to C.”

“If you want to do large scale training like we do, You need the memory to be co-located with a large cluster of GPUs. I can't just rent from Google or Microsoft or even base 10 and run the size of training that I want because I need a gigantic memory card next…”

“The first one is if you go to the App Store and you search text-to-speech, Speechify has 98% of the installs in text-to-speech for B to C.”

“We're very willing to pay a hundred K per month extra to get them earlier.”

“So one thing that we have changed about our hiring in the last even six months Is we really cared that you read a ton of textbooks about software engineering and that your handcrafted code was amazing. I still care that you read a lot of textbooks about softwa…”

“11 Labs, I think Piotrek at 11 Labs literally bought a bunch of GPUs early, early on and set them up in his house, and then they just kept building bigger and bigger and bigger clusters. They do the same thing that we do.”

“I think it's a strategic mistake for Speechify to go to B to B.”