Latest / The Tech Matchmaker / Why Most AI Voice Agents Fail in Production | Simple AI
Transcript
- Gina: Good morning, everybody. Today we are chatting with Cat Lee, CEO and founder of Simple AI. So, Cat, people would have had to be living under a rock not to hear about everything you have accomplished in such Cat: Yeah. Gina: a short time at Simple AI. We're grateful we were able to make this happen and welcome to the Tech Matchmaker. Cat: Thank you so much for having me, Gina. Super excited to be here today. Gina: Excellent. So let's start with the basics. Tell us a little bit about yourself. Cat: Yeah, I have a pretty nerdy background. So I studied computer science and math at MIT, graduated top of my class, and then I worked in Silicon Valley for a few years. my background is engineering, so I built a bunch of back-end systems for Facebook Newsfeed as well as Instagram stories. So hopefully if you've used either of those products, you've actually touched my work before. from there, I ended up leading engineering and product at a company called Y Combinator. If you haven't heard of YC before, It is basically the most successful startup accelerator in history. A lot of the tech companies you know of and love today, think like Reddit, Twitch, Dropbox, Airbnb, Stripe, these companies were invested in by YC when they were nothing more than just two guys and an idea. And I think that's what really inspired me. Working at YC, I got to meet all these really incredible founders who both came through the batch while I was there and also were came back to talk about the early days of their company. And to listen to all these stories of these great founders going from just a couple of guys with no revenue, no customers, barely an idea, to now today, what is a public company? I think that was really inspiring. Gina: I think it also speaks volume to the saying that it the importance of the people you surround yourself with, right? And and the influence Cat: Really, yeah, but Gina: they have. So and and I'm glad you brought up YC. So you went, as you said, from building tools for founders at YC to becoming one. What was the moment Cat: Yeah. Gina: that you decided to jump? Cat: Yeah, that's a good question. I was there for four years. It's really hard not to catch the bug when you just, you know, get to meet Brian Chesky and Tony Shu and all these amazing people like the Collison Brothers. I think the the moment that really did it for me is Fun fact, Sam Altman, who is the founder and CEO of OpenAI, actually also used to work at YC. So he was running a team called YC Research, which was doing all this really interesting research into all kinds of things like clean energy, space tech, and of course artificial intelligence. At the time, artificial intelligence actually came in a lot of flavors. And the one that kind of took the world by storm today is called Large Language Models, LLMs. And so being at YC, I got to see all this really, really early research previews of LLMs before it really even became ChatGPT And that was really like inspirational to witness. I think Sam spun out YC research into a nonprofit called OpenAI. Long story, it's not a nonprofit anymore, but I actually met my co-founder at YC and the both. Of us saw this just like mind-blowing technology that was being created, and we knew that there was going to be a huge transformation happening in the next few years. Gina: Love it. So give us the simple AI elevator pitch, what the platform does and who it's built for. Cat: Absolutely. So Simple AI is AI voice and other conversational agents for B2C companies that have a ton of customer interactions. So what that means is our AI sounds incredibly natural. It responds quickly, it sounds like a human, and it's able to help companies automate some of their customer conversations, whether that's over the phone, whether that's over email, SMS, or web chat. That's sort of the bread and butter product that we started with. And today we've expanded to all these other contact center-related technologies as well. So our AI is able to help you do automated quality monitoring as well as agent assist. Gina: Excellent. And back to the voice agents, just for a moment. So they're everywhere right now, right? And a lot Cat: Yeah. Gina: of companies have had mixed results. So what are you Cat: Okay. Gina: guys seeing the most common ways businesses might be getting it wrong, so to speak, or Cat: Yes. Gina: having not not having success? And what do you guys do differently? Cat: I love this question because, As you know, the market is saturated with a lot of different vendors offering AI voice. And what's so confusing is that there are a lot of different offerings, but the quality is not all the same between the different vendors. And it's very hard for a buyer to tell what it's going to actually be until they try to work with it, right? I think there are different few different layers. One is the actual voice quality itself. So how natural does it sound? How realistic are the inflections? And then really, really importantly, how fast is the response time? It turns out that is very difficult to engineer and it's what I spend a lot of my team's resource on. Being able to respond quickly, being able to handle interruptions, and then importantly, also just actually saying and doing the right things. It turns out is like non-trivial. If you just slap together like a GBT wrapper and throw it into a voice stack, it will say some crazy stuff. Or it will like not know about your company's voice and tone, things like that. It turns out that there's a lot of engineering involved to make sure that. All the right things are being said, it's following your company's tone and voice. It's not just hallucinating products, but actually checking your product catalog, all those things. So you do need a lot of guardrails there. I think one of the biggest mistakes that people make when they're looking at buying a voice or conversational agent is They just listen to the demo and they think like, this is shiny. It sounds decent. It sounds fine. But they don't look under the hood to know like what is actually the accuracy of this agent. If I, you know, use a bunch of test cases and simulate the agent, is it going to say the right thing 100% of the time? And that's not something that's baked into their evaluation process today, which to be fair, it is very hard to come up with that on your own. When your bread and butter is, let's say, selling jewelry or steaks or storage, like you may not know to check for these things, right? So I think that's one big problem. The other big piece is that a lot of buyers I see, they're mostly concerned with the cost of AI. And so they're looking to buy the cheapest option out there. They're basically evaluating, okay, hey, my guys in the US cost me anywhere from $15 to $30 an hour. My guys in like my outsourced reps cost me, you know, eight to ten dollars, maybe somewhere in that range. I want AI to cost $4 an hour or $6 an hour. So they're looking for the cheapest option. And when you do that, you're actually Actually sacrificing quality. And it comes back to hurt you because your AI ends up being your customers end up way less satisfied with the cheap agent that you bought. And that means that you're losing customers retention, you're losing CSAT. And because they're not getting their problem fixed the first time, they're actually gonna call back again. So you're losing first contact resolution rates and all that stuff. Gina: I know one of the things that surprised me most about your platform was how your agents outperform the trained live reps on conversation. Cat: Yeah. Gina: And so like I was surprised by that. And I know when we chatted about it, sometimes contact center leaders don't, you know, they're they're shocked, they don't believe it. What's the hardest part of getting them to to understand and see those stats and and go, wow, okay. Cat: Yes, that's a great question. I'm doing a terrible job with marketing my own company. You're you're much better at it. so at Simple, we do focus on the sales use case. We think that is a phenomenal use case that both has like really, really high returns for the customer and allows us to work on something interesting. I think when it comes to just peer support, it is about cost cutting, right? Like your customers, when they evaluate you, they're considering the ROI on what is my like human reduction and like labor reduction costs. When you're talking to a sales leader, They're lit, they're still interested in the cost, but they're much more interested in the uplift that they'll get from using your product. So are am I going to make more revenue? Am I going to sell more steak using your AI? I don't just want to reduce people, right? And so a lot of what we do that I think is very different from other vendors is we don't just have a set of prompting to follow your SOP. We actually help design your SOP. So we take a look at all of your past calls and transcripts. One thing that's very important to note is any call center leader is going to have this experience where let's say you have two to three hundred reps in your center, they are not all the same. They do not all perform the same. You're gonna have some superstars that just outperform everybody else, and then you're left trying to figure out how do I bring up the quality of my bottom half. Like my worst sales reps, how do I teach them to be as good as the top reps? And what's so interesting about working with AI agents is the problem completely flipped itself. Now you're thinking, how do I get the AI agents to copy what my best agents are doing? And so a lot of what we do at Simple AI is help you do that analysis to figure out, like, hey, which of your reps are doing the best? What are the patterns that are emerging from your best agents, your best live agents? And then how do we use those examples to then train our model to only do what the best agents do. And then what's interesting. Gina: And I love this great. Cat: sorry, I'm just getting really excited about this. Once we hit that as Gina: I love it. Cat: the baseline, now we can do all sorts of other fun stuff to even further improve the outcomes. So once we've hit a baseline where our AI is as good as your best agent, now we can do more to experiment, right? This is something that you know call center directors could never dream of experimenting with before when they're so busy, you know, training their live reps. Now we can do things like, hey, let's A-B test which voice we're offering to your customer. Do you think they prefer a male or female voice? Do they want a New York? Accent or a Texas accent. And all these things could slightly affect the conversion rate and the upsell rate you're seeing, right? Then we can even do things like experiment with which items are we upselling today or this week. And what exact copy are we using to upsell these items? Traditionally, if you had a new idea for a concept or new copy you wanted to run, it would take weeks or months to propagate that language down to all your agents. And by then you don't even know if your experiment is working or not. With AI, you can instantly run an A/B test in our platform and see the results. Gina: It's phenomenal. And when you talk about the deployment, so I I believe you guys promised like ready, like enterprise ready AI phone agents in days, not weeks. So when you're talking about Cat: Absolutely. Gina: ingesting all that data, how long, like, how long does the actual setup process take? What does that look like? And I know it varies, but on average, what are we looking at for a new customer? Cat: Yeah, it does vary, as you said. It depends a lot on whether our customers are kind of AI ready and they have all that data available, which means like recordings, SOPs, scripting. The more you have, the better it is. I will give you one example. One of our first customers was a company called Omaha Steaks I think you're familiar with that company as well. they are a hundred-year-old company that sells American steaks. They've been around. very, you know, household name. With Omaha Steaks we were able to get live in production in 10 days from the start. You can ask Grant about this. It's such a fun story. And one of the reasons why I think we were able to do it so quickly were because their team actually was like very ready to go. They were ready to work with us. We flew out to Omaha, Nebraska, spent the entire two weeks with their team side by side, sitting together, deeply understanding how their business works, looking at all the documentation, learning from their live reps and all that. There are actually a few different things that need to happen in parallel when you start a new deployment. So one is towards the behavior. So sitting with their QA team and their like agent managing team to figure out what should the agent say. Then we have to work with the IT team to figure out what are all the APIs and integrations we need to support the use case we want to do. So for example, for Omaha, a lot a big part of their use case is taking inbound sales. So if someone calls in, they say, Hi, I want to buy this Father's Day package. I saw there was a sale for it. Great. Let's take down your shipping address, your billing address, your the items you want, etc. And ultimately we want this to be all done with no human intervention. So that means our AI has to be able to look up the customer information, identify is this a new or existing customer, create a new customer if it doesn't exist, be able to access like, hey, they gave me a promo code. Is that promo code you know valid or not? do they have Omaha gold rewards cash or not, like all these things we need integrations with. And then ultimately we need to actually make the order without any human intervention. So there's a lot of IT integration steps there. And then finally, there's the telephony layer as well. A lot of companies have their own CCAS system or telephony system that they already use. And so we want to integrate with them so that make sure we make sure that they have all the reporting they need, right? In this case, Omaha does use Five9 which is a system that we are actually an ISB partner of. And so that ended up making it super easy as well. Gina: Excellent. And I think like security and compliance, SOC2, HIPAA, PI2, maybe not in in Omaha's case, which we should chat with Grant and get both you guys on here one time to talk about the success of that implementation. But how do you guys approach all the different security needs and how important Cat: Mm-hmm. Gina: has that been for enterprise deals? Cat: It's so important. It's of utmost importance I think when you're looking at enterprise customers, they have zero tolerance for security risks. And so as a vendor, you need to make sure that you're compliant with all the basics. So that's like HIPAA compliance. If you're working with anyone in healthcare, that's SOC2 compliance. If you are working with payments and money, which we are, you also have to be PCI compliant in order to be able to handle somebody's credit card. Gina: What's next for you guys? What are you building next? I you mentioned quality at the top of the call. What else are you guys looking at? Cat: Yeah, quality is the number one thing I care about. I think we've come a long way. If you just listen to a demo of our voice versus a year ago, it is so much better. But there's still all these little tweaks. I think In order to make the AI feel as natural as a real human agent, there are all these little details, right? It's sort of like an 80-20 rule where, and nobody's doing this right today, handling all the minute details around, for example, when you get interrupted, the most basic vendors, even if okay, the worst case is they don't let you interrupt at all. They just say their whole thing through. I'm sure you've seen that. Gina: Yeah. Cat: The next level up is you get interrupted, but then you stop speaking right in the middle of what you're doing and the audio just cuts off. That's also super weird. And so we've done a lot to make it feel natural where you actually maybe finish the rest of your word or finish the rest of your sentence and then let them cut you off. So there's all these little things around the quality of the voice. But we're also thinking about expanding to a lot more of the software that a buyer needs, like that a contact center director would need to run their business. And a lot of what we do, like I mentioned before, we're targeting the sales use case specifically. That's not to say we can't do support. We can. Sales is just more interesting to me. And when it comes to sales, there's an entire sort of customer brain behind that, which we are developing and continue to develop. In order to get the conversion and upsell metrics we want. We're not just simply following a scripting that our customers have. We actually provide the customer brain. So whenever someone calls in and tells us stuff, and it turns out by the way, people love telling you stuff on the phone. And so we'll just know all these details about a customer. We know their customers better than they do. So because we're we have these conversations over the phone, their customers will tell us things like, my daughter's soccer practice is on Tuesdays, or I have to take my dog to the vet, or I don't eat chicken because my husband grew up on a farm and he took killed chickens as a kid. So all these details contribute to how we interact with this customer in the future. And it allows us to do more proactive like selling and proactive reaching out, just making sure, like, hey, imagine if you had like a concierge for every little business. So an Omaha Steaks VIP customer could have a texting line with someone that feels like their personal Omaha Steaks account manager or something. Yeah. Gina: shopper. Yeah. That would be wild. And I mean just the the profiling and the specificity you can get to with a customer to make me feel special, right? Or whatever the Cat: Exactly, absolutely. Gina: case may be. Yeah. It's phenomenal. Is there anything, Cat, that I haven't asked you that you would like the audience to know? Cat: I'm just thinking about all these, you've got my brain turning on all these other examples now. Gina: Yeah. Cat: I have a customer that sells jewelry, and I remember the most astonishing and just like such a heartwarming experience where we recommended this pink item to a specific customer because we saw and noticed that all of her past purchases are pink stuff. So when it comes to like earrings, necklaces, bracelets, everything, she loves pink. And so we knew when this new item came in to our inventory, we basically figured out like, you know, Jenny is going to love this item. And we suggested to her and she totally not only did she love the item, she actually loved the personal recommendation and the personal attention that we afforded. And to me, that's everything. Gina: I I couldn't agree more. And and especially in this world of self-serve and and you know everything listening to us to have that personal touch is is phenomenal. I love that. To wrap things up, Cat, I've always been a fan of James Lipton. And so to end the podcast, we always do a little bit of rapid fire. What Cat: Okay. Gina: sound or noise do you love? Cat: What sound or noise? so I grew up playing the piano, so I just I just love like calming piano music. Gina: Love it. What sound or noise do you hate? Cat: This sounds so bad, but anyone singing off-tune just I just can't. Gina: That's great. What profession Cat: I I just can't. Gina: other than your own would you like to attempt? Cat: Mmm, I have wanted to be a singer before. I'm currently in an a cappella group, but nowhere near professional. Gina: Good for you. With all your free time. I love it. Cat: Yes, with all my all the abundant free time I have. Outside of my skill singing. Gina: Now if somebody wants more information on Simple AI, who should they call? Cat: well we do have a website. If you go to usesimple.ai you can also reach out to me, Cat@usesimple.ai or or Gina. Reach out to Gina. Gina: Excellent. Cat, thanks so much for joining us today and and just sharing, you know, your experiences, your customers' experiences, and your vision for the future of AI. It's it's fascinating and I can't wait to see what you guys do next. So thanks so much for your time today. Cat: Thanks so much for having me.