Skip to content
YC Root AccessYC Root Access

This AI Startup Is Taking Over Phone Sales

In this episode of Founder Firesides, YC Managing Partner Jared Friedman talks to the founders of Simple AI (S24), Catheryn Li & Zach Kamran, who just raised a seed round of $14M. Simple AI gives businesses an AI sales agent that handles inbound calls end-to-end and outperforms their human reps. Apply to Y Combinator: https://www.ycombinator.com/apply Work at a startup: https://www.ycombinator.com/jobs

Jared FriedmanhostZach KamranguestCatheryn Liguest
Feb 18, 202637mWatch on YouTube ↗

EVERY SPOKEN WORD

  1. 0:00 – 0:05

    Intro

    1. JF

      [upbeat music]

  2. 0:05 – 0:54

    Simple AI: an AI voice agent that completes sales calls end-to-end

    1. JF

      I'm really excited to be sitting down today with Cat and Zach, the founders of Simple AI. Welcome, guys. Thanks so much for joining us.

    2. ZK

      Yeah. Thanks, Jared. It's good to be back.

    3. CL

      Thanks for having us. It's so fun being on this side of things.

    4. JF

      Before we get started, why don't you tell everyone what is Simple AI?

    5. CL

      So we're building AI voice that sells. What that means is our AI sounds very natural, and it helps businesses handle sales phone calls. It turns out that a lot of major iconic brands still do the majority of their revenue over the phone today. Our AI sits in front of your phone system. When someone calls in, it says, "Hi, thank you for calling XYZ. How can I help you?" And it's able to take you through a journey where it can explain all the products, um, answer your questions about anything, take your billing information, take your shipping information, and handle the entire sales process end-to-end with no human intervention.

  3. 0:54 – 1:44

    Iconic customer proof: Omaha Steaks runs its main phone line on Simple AI

    1. JF

      We mentioned there's iconic brands who are using Simple AI. Can you give an, an example or two? Who's, who's using this now?

    2. CL

      Today, Simple AI sells everything from steak to self-storage to home insurance. One of our customers is this brand in the US called Omaha Steaks. They've been around for 100 years. They started as a family butcher's business a century ago, and they sell steak over the phone.

    3. JF

      And this is, like, a big company.

    4. CL

      Big company.

    5. JF

      Okay. And if I call Omaha Steaks now, there's a good chance that, like, Simple AI is gonna pick up and-

    6. CL

      It's not a good chance

    7. JF

      ... take my order.

    8. CL

      If you call the number on their website, it will go to Simple AI.

    9. JF

      That's cool. You guys had a very interesting backstory to how you ended up starting this company. So I wanna rewind back to the beginning now and tell everybody how you ended up starting this startup.

  4. 1:44 – 3:39

    Founders’ origin story: meeting and building software inside Y Combinator

    1. CL

      So Zach and I actually met while working at YC with Jared.

    2. ZK

      [laughs]

    3. CL

      Um, we worked for it together for many years. I ended up running engineering and product for a software team at YC, and I met Zach there. We worked together for three years. I knew this guy was a cracked programmer.

    4. ZK

      [laughs]

    5. CL

      And [laughs] I was like, "I have to work with him."

    6. JF

      I think a lot of people don't realize that YC even has a software team or what it would mean to be a programmer at YC. Like-

    7. CL

      Yeah

    8. JF

      ... that sentence probably makes no sense in a lot of people's brains. So how about you guys just give, like, a couple sentences of context about, like, how it was that you were working at YC as programmers and-

    9. ZK

      Yeah. So YC actually has, like, a, a ton of software, um, s- that all of its founders use. So I was working on a team called the, the Batch Team, and the product we were working on is called Bookface, which is basically like Facebook for all the founders who do YC. And it has a ton of resources, and also is just a place for all the founders to connect and talk to each other and meet other founders.

    10. CL

      I think software is actually a secret weapon of YC's. In order for YC to be able to invest in so many really amazing companies, it-- you really have to have software to even handle applications, and interviews, and legal diligence, and accounting. And taxes are really complicated because YC's portfolio is, like, thousands of startups and who knows what happened to them. They could have died or gotten acquired or went public. Um, and it's all very confusing. So I think software is just, like, really crazy superpower that YC uniquely has. Um, and then YC has all this really cool, unique data and insight. Probably no other institution in the world has as much data as YC does on early-stage startups and how they're formed and how they ended up.

    11. JF

      Yeah. It was awesome getting to work with you guys, and I'm so excited that you decided to branch off and start your own company, which is what we should talk about next.

    12. CL

      [laughs] Yeah.

    13. JF

      So now you guys are running this awesome company called Simple AI.

    14. ZK

      Yeah.

    15. JF

      Can you tell us about it?

  5. 3:39 – 4:47

    Catching the startup bug: deciding to build after years around founders

    1. CL

      Yeah. I think it was just really hard not to catch the [laughs] startup bug at YC.

    2. ZK

      [laughs]

    3. CL

      You meet all these wonderful people. I think both we-- Zach and I both spent four years at YC, and so we met founders both in the program while we were there, but also all these crazy alums who would come back and talk about the early days of their company. And so I think it was very inspiring to hear how Airbnb was just three guys, an idea, no customers, no revenue, and today it's a public company. Um, and I think we both were extremely inspired by that. I think at YC you also get to see some really cool innovations before the rest of the, the public does. Um, for example, OpenAI was actually started as a com- a team within YC called YC Research. And so being at YC, we had access to all this really, really cool technology that the workers were-

    4. JF

      So we have a pretty good track record with spin-outs.

    5. CL

      [laughs] Exactly.

    6. ZK

      Yeah. [laughs]

    7. CL

      Yeah, exactly. Um, and so we got to see all these cool, like, research previews of the things that OpenAI were doing and that Claude were doing.

    8. ZK

      Yeah.

    9. CL

      Um, and so I think we knew that it was time. There was something really, really big we could do here.

    10. JF

      How did you guys come up with the idea? How did you decide what to work on?

    11. ZK

      Oh, my gosh.

    12. CL

      Yeah. [laughs]

    13. ZK

      That's a great question.

    14. CL

      That's a good story.

    15. ZK

      [laughs]

  6. 4:47 – 5:49

    Idea #1: building a “better Siri” (and why it confused users)

    1. CL

      Um, when we started the company, I had insisted that we should work on consumer, and Zach probably remembers this very well.

    2. ZK

      True story. Yes.

    3. CL

      Um, my experience was largely in consumer. So at YC, I built a lot of products for founders, like Startup School and YC's co-founder matching site. We actually collaborated on launching YC's mobile app together, if you remember that. [laughs]

    4. ZK

      Mm-hmm.

    5. CL

      And so I think a lot of my experience was building, um, for people like founders. Before YC, I actually worked at Meta. So I worked on a bunch of back-end systems for Instagram Stories and Facebook News Feed, both of which are very large consumer, like, high DAU products. So that was what my experience was, and I thought my passion was as well. Um, so I was just like, "We have to work on consumer. Nobody is really servicing consumer. Like, there's all this crazy AI technology."

    6. ZK

      Yeah.

    7. CL

      "LLMs are so powerful, and yet Siri sucks."

    8. ZK

      Yeah.

    9. CL

      Like, frustratingly, Siri can't do anything for you. And so it just felt like there was this, um, lack of really great, um, AI consumer products, and we-- so we started by venturing there.

    10. ZK

      Yeah.

  7. 5:49 – 8:26

    Idea #2 goes viral: AI that makes outbound calls for consumers

    1. CL

      The first thing we tried to build was a better Siri, and basically you could tell it to do things for you. Um, in addition to asking it for advice or chat, you could also ask it to, "Hey, can you call my Uber home?" or, "Can you order DoorDash for me? Can you check my calendar?" All these things.Um, and it turned out that when the app could do 50 things, it was very confusing to everyone what it could and couldn't do. For us, we knew exactly what it could do 'cause we wrote the code for it.

    2. JF

      [laughs]

    3. CL

      But no one else could figure out, you know, where are the limits of this? Okay, I can call an Uber Home. Can I add in a second stop? Can I cancel it? Can I change the ride? Who knows? So we decided to focus on just one thing and be really, really good at it. And the one thing that all of our users gave us tremendous feedback about was this voice AI capability. We could use voice AI technology to make phone calls on your behalf. So we could, you know, call restaurants to get you reservations, and call your doctor to set appointments. And this ended up being really cool, 'cause people would use it to do all kinds of crazy stuff that they could not do before. Um, one user actually sent us an email that was a selfie of them buying a car that they had Simple negotiate, which was really cool.

    4. JF

      Wait, wait, wait. So they told Simple, "Call this dealership and buy me this-"

    5. CL

      No, not just this dealership

    6. JF

      ... okay. I didn't-

    7. CL

      "Simple, call these 10 dealerships."

    8. JF

      Oh, 10 dealerships-

    9. CL

      And find-

    10. JF

      ... and find me the best price for this car."

    11. ZK

      Yes.

    12. CL

      Yes.

    13. ZK

      Yes.

    14. JF

      Interesting.

    15. CL

      He knew exactly what car he wanted to buy, so I wanted-

    16. JF

      Yeah

    17. CL

      ... the, you know, this model and this year.

    18. JF

      Yeah, that makes complete sense.

    19. ZK

      Yeah.

    20. JF

      It's, like, horrible to haggle with a bunch of dealerships.

    21. ZK

      [laughs]

    22. JF

      Okay, yeah.

    23. CL

      And haggle with everybody. He ended up choosing the cheapest one, went there, bought it, sent us a picture.

    24. ZK

      Yeah, I think we saved him-

    25. CL

      And that was really amazing

    26. ZK

      ... like, thousands of dollars.

    27. JF

      Okay.

    28. CL

      Yeah. And time.

    29. ZK

      And, and huge amounts of time.

    30. CL

      Like, imagine going to all these different stores, right?

  8. 8:26 – 9:30

    Why the viral consumer product didn’t monetize: novelty vs repeat usage

    1. ZK

      Yeah, so the thing is, we had all these people using it for, I mean, thousands of different things. And what ended up happening is you would use it, and you would be like, "That was the coolest thing ever. Oh my gosh." And then you would never use it again.

    2. CL

      [laughs]

    3. ZK

      Because you don't need to buy a car that often.

    4. JF

      [laughs]

    5. ZK

      And, like, you don't really need to, like, call Priceline that often. And so when you do use it, it was really, really cool, and maybe you would even use it for fun, too, and have it call your friends and whatever.

    6. CL

      Which people did.

    7. ZK

      Which people did do.

    8. CL

      They would prank call everybody they knew.

    9. ZK

      S- so much fun to do that.

    10. JF

      [laughs]

    11. ZK

      Um, but at the end of the day, you're just not really gonna pay for something you use once a year, even if it does save you, like, $600.

    12. JF

      Yeah.

    13. ZK

      Um, but what keep, what kept happening was these people would try it, and then we would talk to all of our users. We would send each of them an email and talk to as many of them as we could. And they would start telling us, "Hey, your app was super fun," or, "Hey, I used it for this thing. It got me this reservation. That was awesome, but I'm definitely not gonna pay you any money for it."

    14. JF

      [laughs]

    15. ZK

      And we'd be like, "Yeah, we know. Like, we, we track the stats. Like, yeah, you haven't opened it since then." [laughs]

    16. JF

      Right.

  9. 9:30 – 11:27

    The pivot trigger: inbound business demand—and a prank call that landed Omaha Steaks

    1. CL

      But on the other hand, we actually got a bunch of inbound from people who had tried the app and told us, "Hey, your voice thing is so cool."

    2. ZK

      Yeah.

    3. CL

      "Can I use it for my business?" And so I think we had this light bulb moment where at first I was a bit resistant to it. We kind of just thought, like, well, you know, there are all these other services that do this. There are a bunch of other YC companies, in fact, that build AI voice for business. Um-

    4. JF

      Use it for my business so I can call a bunch of people?

    5. CL

      No, for inbound. Like, my business gets inbound, or I'm the C blank O of this other company, and we get tons of inbound calls.

    6. JF

      That's interesting. How did they, like, connect the one thing to the other thing? 'Cause the consumer product was really, like, an outbound-

    7. CL

      Yes

    8. JF

      ... calling system, not an inbound calling system.

    9. ZK

      Yeah.

    10. CL

      Correct.

    11. JF

      And it was the same people who were using the consumer idea.

    12. CL

      Yes.

    13. ZK

      Yeah.

    14. JF

      Okay.

    15. CL

      Yes. So I think it, you m- it might take some creativity, but the first thing that you do when you sign up for our consumer app is we use your phone number as a test, so you can hear what it sounds like, and you can experience it for yourself. And so when people heard this example, they basically went, like, "Oh, shit, this technology is really cool. This voice is so realistic, and I can just use it for all of these other things I have to do." This is actually how we got our very first big customer.

    16. ZK

      Yeah.

    17. CL

      The CEO of Omaha Steaks, which is a company that sells lots of steak over the phone, actually used our mobile app to prank call his COO when they were commuting to work together.

    18. JF

      [laughs]

    19. CL

      And they realized this would completely change their business.

    20. ZK

      Yeah.

    21. JF

      Wow.

    22. ZK

      I-

    23. JF

      So that's how you got the Omaha Steaks deal.

    24. CL

      Yeah. [laughs]

    25. ZK

      Yeah. I think it was, it was such a hair-on-fire problem for them, that they were able to, like, imagine this consumer app solving their, like, totally different thing in their call center.

    26. JF

      Interesting.

    27. ZK

      Which was, like, really, really cool. It, like, really showed us, like, oh, they, like, really care about [laughs] this problem. Like, this is a really big pain for them.

    28. JF

      Okay, so explain what idea number three is, which is, like, the current Simple AI.

  10. 11:27 – 14:01

    Phone sales is mission-critical: Omaha Steaks’ holiday surge and staffing crisis

    1. CL

      So now we're building Simple AI, which is AI voice that sells. And the way it got started was we were talking to-

    2. JF

      What does that mean, AI voice that sells?

    3. CL

      Yeah. So a lot of companies like Omaha Steaks actually still a lot of, still do a lot of their revenue over the phone, it turns out. Um, businesses that either sell, you know, complicated items or expensive items, there are a lot of reasons why someone wants to talk to somebody before they buy something. And then in the case of Omaha Steaks, they are a very, um, gifting-oriented business. They sell a lot of steak over the holidays. And so they have this crazy problem where in October, they 15x their entire workforce in order to gear up for Thanksgiving and Christmas

    4. JF

      Whoa, so they do like so many sales in Christmas that they need to like dramatically scale up their sales team-

    5. CL

      Literally 15 answers

    6. JF

      ... with a bunch of temp workers who are gonna just answer phone calls for like two months and then don't have a job again in-

    7. CL

      Correct.

    8. ZK

      Yes.

    9. CL

      Yes

    10. JF

      ... in January.

    11. ZK

      Yes.

    12. JF

      Wow.

    13. CL

      And they have to hire all those people and train them in October so that they're ready to take the calls through Thanksgiving and Christmas. But also, these are people that are down to do a temp job for just November and December, and so they tend to not show up to their shifts, and they don't show up to their training. Even if they do show up to their shifts, they might not have paid attention during training, so they don't know how to use the software.

    14. JF

      [laughs]

    15. CL

      And so all these people are calling Omaha Steaks going, "Hey, I saw your holiday campaign. I wanna buy your holiday package." And the- these temp workers don't know how to take those calls.

    16. JF

      Wow.

    17. CL

      And so Omaha is losing tons and tons of revenue over this.

    18. JF

      That sounds like a hair-on-fire problem.

    19. CL

      [laughs] Yeah, it is.

    20. ZK

      Yeah.

    21. JF

      Okay.

    22. ZK

      Yeah.

    23. JF

      Okay, and so the Omaha Steaks CEO uses Simple AI version two to prank call his COO.

    24. CL

      [laughs] Correct.

    25. ZK

      [laughs]

    26. JF

      And then at some point he's like, "Wait, could you guys help me with all my holiday calls?"

    27. CL

      Exactly.

    28. JF

      How- how did that happen?

    29. CL

      I think we at first were unsure because we had known about all these other solutions out there, including other YC startups that do literally AI voice for business.

    30. JF

      Mm-hmm.

  11. 14:01 – 16:31

    Deploying in the real world: integrating with 1990s-era systems and on-prem constraints

    1. JF

      I'm guessing the Omaha Steaks guys are not like exceptionally technical.

    2. CL

      They don't have an engineering team.

    3. ZK

      Mm-mm.

    4. CL

      But they have an IT team.

    5. ZK

      Yeah.

    6. JF

      Okay.

    7. CL

      And they've actually-

    8. ZK

      Yeah

    9. CL

      ... self-created all this crazy software from the '90s.

    10. JF

      Huh.

    11. CL

      Um, [laughs] and so we actually spent two weeks in Omaha, Nebraska, working with the team and helping them get set up and understanding deeply what their business- how it works and all that. A lot of their software is actually like AS/400 terminals. Like, that's what their marketing team uses to create promotions. That's what all [laughs] their sales team is-

    12. ZK

      Yeah. But the thing is, this is-

    13. JF

      Can you describe what this terminal is-

    14. ZK

      Yeah

    15. JF

      ... for people under the age of 50 who are listening? [laughs]

    16. CL

      Um, so imagine like- imagine an old movie that has crazy hackers in it, and instead of like a- a high-tech screen, you have this big box that's gray.

    17. ZK

      It's gr- gr- like a green screen.

    18. CL

      Like, like... Yeah. Okay.

    19. ZK

      It's like a-

    20. CL

      It's like- like a big round box.

    21. JF

      Like- like green text on a black background.

    22. CL

      Yes.

    23. ZK

      Yes. Yes, I agree.

    24. CL

      It's like a big box, and in the middle there's this little black, uh, screen thing, and then there's text that's all mono font.

    25. ZK

      [laughs]

    26. CL

      Um, you can't even click anywhere. You can- actually can only like tab and space and enter and stuff.

    27. ZK

      And there- there's not database tables. There's just files. It's like... And you can't-

    28. CL

      Oh my God

    29. ZK

      ... you can't add columns and, like... It's crazy.

    30. JF

      And this is how they run their advertising?

  12. 16:31 – 20:26

    From product catalogs to SKUs: teaching the agent to sell complex inventories

    1. CL

      It's to make sure we can do things like fetch their customer information. If someone calls in, we need to know, is this a past customer or a new customer? Did they buy things with us in the past? What did they buy in the past? And that's just step one, right?

    2. JF

      Okay.

    3. CL

      In order for our AI to work really well, um, we have to also do things like take your order and place the order-

    4. ZK

      Yeah

    5. CL

      ... 'cause we are building a sales agent, right?

    6. ZK

      Right.

    7. CL

      So we have to take all your order. There's a ton of complexity about around the items themselves. Um, Omaha Steaks, for example, has many versions of filet mignon. If you ask me for filet mignon, we have a six-ounce, seven-ounce, and eight-ounce option. We have a four-count, six-count, and eight-count option. And then we have all these different packages. So during the holidays, they might have promotions for 20% off this thing or like pick 6 out of these 20 items. There are a lot of different, um, very complicated structures around their items and SKUs. They have elemental items and combination items. They have different discounts and promotions.

    8. JF

      And you had to teach your agent all of those things in order for it to be able to handle these sales calls. How did you like suck the knowledge out of the people at Omaha Steaks in a, like inject it into the agent?

    9. CL

      We spent two weeks living in Omaha, Nebraska, doing daily stand-up and planning with everybody on their team.

    10. JF

      Yeah.

    11. CL

      So we would talk to their product managers, project managers, their engineers, their CIO, who has been with the company for 25 years, um, people on the marketing side, so sales and marketing team. I think their business is very marketing-driven, and so they do a lot of experimentation around what packages are we selling this season and how much do we want to price that. They actually have different source codes for even different parts of the country. So they might send the same package, you know, flyer to Missouri, but then also New Jersey, but they show different things and different discounts and all that stuff.

    12. ZK

      Yeah. I think what's really interesting is we also learned, like-Not only is Omaha Steaks really, really complex, but actually like all of our customers who we talk to-

    13. JF

      Mm

    14. ZK

      ... they all will tell you like, "Oh, well guys, sorry, like we have a really, really complicated system." [laughs]

    15. JF

      [laughs] Yeah.

    16. ZK

      And like they actually all do. Like it's all is pretty complicated, but there are abstractions that like work across them that I think are specific to sales and e-commerce and selling things.

    17. JF

      Yeah. Who were the other customers? It's not just Omaha Steaks.

    18. CL

      No.

    19. JF

      You guys have other-

    20. CL

      Yeah

    21. JF

      ... other customers too.

    22. CL

      Today, Simple AI sells steak, home insurance, self-storage, jewelry, like anything you can imagine where someone might call in to ask questions about, um, or make their purchases, we do.

    23. JF

      Is it similar, companies with similar models to Omaha Steaks, like direct-to-consumer businesses that have high-ticket items where their clientele is calling in on like a 1-800 [laughs] number and placing orders over the phone?

    24. CL

      That's exactly right. Yeah.

    25. JF

      Okay. Yeah.

    26. ZK

      Mm.

    27. CL

      We mostly work with direct-to-consumer companies.

    28. ZK

      Yeah, yeah. And it could be like the product is complicated, so like maybe you sell auto parts and someone calls in 'cause they wanna buy oil, but there are like thousands of different types-

    29. JF

      Mm

    30. ZK

      ... of car oil, you know? And like I think even people in San Francisco would like call AutoZone 'cause they're like, "I don't know, man, like my car's broken. Like I need oil for it."

  13. 20:26 – 23:54

    Performance edge: 30% better upsell than trained reps + personalization experiments

    1. JF

      How do the AI agents compare to like a human answering the same phone call?

    2. CL

      Our AI is actually really, really good at sales. A big part of companies like Omaha Steaks and their business, the way they might make money is actually by upselling.

    3. JF

      Yeah.

    4. CL

      So maybe you called in, you wanted to buy this holiday special that you saw on TV, right? Um, the job of a sales agent at Omaha Steaks is actually to then say, "Okay, great. I have your order. Oh, by the way, your order qualifies for a great add-on today. For only $40 more, you get free shipping and six burgers and blah, blah, blah, blah, blah."

    5. JF

      Yeah.

    6. CL

      And that's very important to their business.

    7. JF

      Yeah.

    8. CL

      'Cause even a 20, $40 lift on every order, that adds up really quickly.

    9. ZK

      It's a huge amount of money.

    10. CL

      Right?

    11. ZK

      Yeah.

    12. CL

      And so upsell rate is something that these businesses care a lot about.

    13. JF

      Yeah. Yeah.

    14. CL

      Our AI is 30% better at upselling than trained live reps. Let's not even talk about-

    15. JF

      Seriously?

    16. CL

      ... the temp workers, the trained live reps.

    17. JF

      Like the full-time people who are there-

    18. CL

      Yes

    19. JF

      ... year-round.

    20. ZK

      Who are there year-round-

    21. JF

      Okay

    22. ZK

      ... have been doing this for years.

    23. JF

      Wow.

    24. CL

      And there's a lot of reasons for it.

    25. JF

      Why?

    26. CL

      So for one, I think that every call center has some distribution and variation on how good their agents are, right? So you might have three or five superstars that are really, really good at upselling.

    27. JF

      Mm-hmm.

    28. CL

      And then you have other people who are kind of like, oh, okay, they're just kinda here for their nine-to-five. They don't really care that much. And I think the magic of AI and software is we can selectively train on the best agents so that-

    29. JF

      Mm

    30. CL

      ... every single call you take is taken by your best agent.

  14. 23:54 – 26:56

    Beyond labor replacement: better customer experience through memory and conversation

    1. CL

      I think what's really cool is people who are on the other side of the line, like the end customers, they're actually having a better experience.

    2. JF

      Yeah.

    3. CL

      'Cause one of the problems with call centers is that you're always trying to decrease your average handle time. That is a stat that these people care about, and the reason is because there's always people waiting on hold, and so the faster you finish your call, the more calls you can actually take.

    4. JF

      Oh, interesting.

    5. ZK

      Yeah.

    6. CL

      But now-

    7. JF

      But if you're not in a hurry, you can focus on upselling to the max.

    8. CL

      Exactly.

    9. JF

      Right. Yeah.

    10. CL

      And not even just upselling, you can provide the best experience by asking about their day and, you know, just having conversations. Sometimes people just wanna talk to someone.

    11. JF

      Interesting.

    12. CL

      That's it.

    13. JF

      Does your agent do that? Is it instructed to basically just like-

    14. ZK

      Have a conversation with the person as if, you know, the agent had all the time in the world.

    15. JF

      Yeah.

    16. ZK

      Like, you know, "Oh," like, "What are you buying this steak for? Oh, what are you..."

    17. JF

      Yeah.

    18. ZK

      Like, "What else are you having for Thanksgiving?" You know? [laughs]

    19. JF

      Yeah.

    20. CL

      No, no, seriously, it's so amazing. Customers will literally tell us their whole life story. Like, there was one time we tried to upsell some woman a chicken package because that's what the marketing team was promoting, and she responds by telling us, "Oh no, dear, my husband grew up on a farm when he was a kid, and he had to kill chickens when he was younger, so we don't eat chicken anymore." Great, I will never sell you chicken again.

    21. ZK

      Oh.

    22. JF

      Yeah.

    23. ZK

      Do you have a way to basically persist that into their non-database database?

    24. CL

      [laughs] Well-

    25. ZK

      Well...

    26. CL

      In our, in our database. [laughs]

    27. ZK

      Yeah, in our real-

    28. JF

      Yeah.

    29. ZK

      Our real database. [laughs]

    30. JF

      In your real database.

  15. 26:56 – 29:00

    Company momentum: launch timing, funding, hiring, and building premium-quality voice AI

    1. JF

      So when did you guys launch the, like, the current version of Simple AI, the sort of, like, turning point when, with Omaha Steaks?

    2. CL

      February last year.

    3. JF

      Okay, so like one year ago. Where's the company at now?

    4. CL

      In the last four months we went from two people, just me and Zach, to 10. So we've grown as a team. We have an office in San Francisco, and we raised $14 million to keep working on this.

    5. JF

      And this is brand new, right?

    6. ZK

      Yeah.

    7. JF

      Congratulations.

    8. CL

      Thank you.

    9. ZK

      Thank you. [laughs]

    10. JF

      Yeah.

    11. CL

      Yeah, we're super excited. I think it's very clear to us that this idea has legs. I think people really want what we're building, and I'm very confident that we're building the best-in-class product as well. I think what's so cool about building in the sales realm and not some other voice AI, like support for example, is that because we're helping businesses increase revenue, we don't have to be the cheapest option.

    12. JF

      Yeah.

    13. CL

      And so that means we're in turn able to spend money on R&D, making our agent the best, highest quality performing agent.

    14. JF

      Mm-hmm.

    15. CL

      We can do things like spend money to decrease the latency, for example. I think a lot of the other voice AI options out there are working to make their agent as cheap as possible and then pass on the savings to their business. We're doing the exact opposite. We can spend lots of money on making the AI sound really natural, behave really well, and also respond very quickly.

    16. JF

      Mm-hmm.

    17. CL

      Um, and that turns out to be, like, a really great experience for our customers and then their end customers.

    18. JF

      It's also probably a fun engineering problem to work on.

    19. CL

      It's way more fun. [laughs]

    20. ZK

      Incredibly fun engineering problem to work on.

    21. JF

      Yeah.

    22. ZK

      Yeah.

    23. JF

      Yeah.

    24. ZK

      I mean, so much more interesting than, like, a, like a crud app. [laughs]

    25. CL

      [laughs]

    26. JF

      Yeah.

    27. ZK

      Yeah. Yeah. I, I actually think, like, we have some of the most interesting engineering problems, like, of, of any company, any startup right now. Yeah.

    28. JF

      Are you guys hiring more people now that you've grown to 10?

    29. CL

      We are, yes.

    30. JF

      Okay. Who, who are you hiring?

  16. 29:00 – 35:18

    Technical moat: latency, custom models, evaluation, and guardrails for reliability

    1. ZK

      The thing about what we're building is it touches so many parts of your business, and it's a really complex product because it's real time. Our goal is to respond to our customers when you're talking on the phone in under 600 milliseconds, which is really not that much time. And so there's a ton of software to build, and I think a ton of really interesting software to build. Um, and our goal is really just to build, like, the best product out there, where when you talk to our voice agent you don't say, "Oh, I wanna talk to a human." You say, like, "Oh, thank God it's, like, Simple AI agent." And, like, I think we want our customers to feel the same way, where, like, when you see a demo of the product you're like, "Wow," like, "I, I have to have this." Um, and I think we're, we're close, but, like, no one, no one on the market is there yet, and, like, I think we're, we're hopefully the closest. And so we're spending a lot of time on just core engineering and making the product amazing.

    2. JF

      Do you wanna talk about any of the technical problems that you're working on or the roadmap?

    3. ZK

      Yeah.

    4. JF

      What if- Like, if people go to work at Simple AI, what will they get to work on?

    5. ZK

      I mean, one of the biggest problems for voice agents is latency. So how do you make the conversation sound really natural while doing all these, like, really complex things, while selling up to, like, 100,000 different products and, like, answering questions about a bunch of different things? And so we're doing a lot of interesting stuff to make that work really amazing, all the way from building a full evaluation suite so when you're building these agents you can test them and have them simulate, like, thousands of calls to know that, like, when you put them in production they're gonna work really well, to doing RL and fine-tuning our own models so that instead of using the, like, you know, OpenAI or Anthropic models that take a second to respond-We can respond in 100 milliseconds-

    6. JF

      Wow

    7. ZK

      ... because we have a model that's trained for each individual customer and each individual use case done automatically, so our customers know nothing now. When I-

    8. JF

      Okay

    9. ZK

      ... when I think about our whole product, our goal is that when customers come to us with res- questions, we can just tell them yes, and it sort of like magically works.

    10. JF

      Hm.

    11. ZK

      And behind the scenes, like they don't need to know what's happening, but it is like really, really complicated. Um, and so we're working on a ton of software to get latency down, and then deal with all these other like long tail voice agent problems. So for example, like transcription models are really bad at transcribing addresses.

    12. JF

      Mm.

    13. ZK

      It turns out when you're selling people things, [laughs] you collect a lot of addresses.

    14. JF

      Addresses, yes.

    15. ZK

      Um-

    16. JF

      And it's very annoying.

    17. ZK

      And it's so annoying. There's nothing more annoying than having to, to repeat-

    18. JF

      Repeat your address

    19. ZK

      ... your address three times.

    20. JF

      Yes, totally.

    21. ZK

      Um, and our solution is we just, we'll do it ourself, and so we'll train our own model that's just like incredible at transcribing addresses.

    22. JF

      Addresses.

    23. ZK

      And whenever you're transcribing an address, the customer will be like, "Oh, cool. It worked." Behind the scenes they will not know that there [laughs] were hundreds of engineering hours spent to make it work magically.

    24. JF

      See, I think that's such an interesting example of like the moat of companies like SimpleAI. Like a common question that we get when we talk to aspiring-

    25. ZK

      Mm

    26. JF

      ... startup founders is like, "Oh, these agent companies, you could just like build it in a weekend. There's no d- d- defensibility."

    27. ZK

      Yeah.

    28. JF

      And it's like, yeah, you probably could build like a very-

    29. CL

      There's 50% there.

    30. JF

      Yeah.

  17. 35:18 – 37:13

    Founder lessons: don’t over-plan—talk to users early and follow demand

    1. JF

      So before we wrap up, guys, you guys have a, had a very interesting startup journey, first working at, [laughs] at YC on the software team where you shipped a ton of the like core software that's like essential for, for YC. You guys met each other there, then you decided to like break off and do your own company, and then you went through really like three ideas before you landed on the thing that actually worked, and it took like over six months to really like get there.

    2. ZK

      Yeah.

    3. JF

      Um, so you guys have been like deeply immersed in startup stuff for like many years now.

    4. ZK

      Mm-hmm.

    5. JF

      What have you learnt on the journey? What advice would you guys have for other people starting companies?

    6. ZK

      Yeah. I, I think most founders feel like they have to have their idea too figured out before they just go and like try to get customers and build things, and I think like for Cat and I, like we've always just followed what was like the most interesting and what our users actually wanted. So like when we were building the consumer app, like when we first got started, we had like 10 users. But like we would still talk to them and like treat them like we had a million users and like the product actually mattered a lot. Um, and like same for the business product where like, you know, we weren't like experts in call centers four months ago, but like we knew that it was a real problem and we knew like we would figure it out. Um, and I think like working at YC, we saw all these founders tell their stories where they would have been working on something entirely different, like the Brex founders were working on VR goggles.

    7. JF

      [laughs]

    8. ZK

      And then they just like realized that like, "Oh, all the people in my batch need credit cards," and no one will give them a credit card. But [laughs] they didn't know that much about credit cards.

    9. JF

      Yeah.

    10. ZK

      And they just like went and did it and were like, "I'm smart. I'll figure it out." And I, I kind of think like a lot of great founders just do that, and so I think we, we've tried to do that along the way.

    11. JF

      I love that. Thank you guys so much for coming back to YC.

    12. ZK

      Yeah. Thanks, Jared.

    13. CL

      Thanks for having us.

    14. ZK

      It's awesome to be back.

    15. JF

      [laughs] [outro music]

Episode duration: 37:13

Install uListen for AI-powered chat & search across the full episode — Get Full Transcript

Transcript of episode 6iilze3aDkU

Get more out of YouTube videos.

High quality summaries for YouTube videos. Accurate transcripts to search & find moments. Powered by ChatGPT & Claude AI.

iOSAndroidClaudeChrome