🎧🍌 The $750 Billion AI Opportunity in Customer Service | Mike Murchison, Co-founder and CEO of Ada
Why AI is still underhyped, how large companies stand to benefit most from AI, why companies talk to customers less as they grow (and how to flip that), and advice for anyone building AI agents today
Mike Murchison is the Co-founder and CEO of Ada, the AI-powered customer service automation platform.
Mike and Ada have quietly built one of the largest AI-native businesses. The nature of its product and scale puts Mike at the forefront at how AI is changing software and labor markets, and this conversation felt like both a glimpse into the future, as well as a look into the past, at a story of pure grit and determination, pivoting to a new market after working seven customer service jobs at once.
We talk about why management capabilities becomes even more important in AI-native companies, how customer service is changing from a cost center to a revenue driver, and how to talk to customers more as you scale.
We also get into why AI is still underhyped, what truly AI native software looks like, the realities of selling enterprise AI software right now, and advice for anyone building an AI agent from scratch today.
A quick request - if you want to help more people find the show - please comment + reply to these posts on X and LinkedIn
Thanks to Boris Wertz at Version One and Fahd Ananta at Roach Capital for their help brainstorming topics for Mike!
Timestamps to jump in:
3:49 Making customer service extraordinary for everyone
5:44 Management becomes more important in AI-first companies
12:13 From customer inquiry to solution in production, fully autonomously
16:01 Why companies talk to customers less as they grow
20:36 Creating new products from customer service data
22:45 Broken incentives in customer service
26:10 Working 7 customer service agent jobs at once for a year
37:19 Why pivoting to Ada felt like failure
46:11 How Mike would build an AI agent from scratch today
49:15 Ways AI will change how we build and manage companies
56:44 Why the best managers are great users of AI
1:00:27 How the top 1% of people are using LLMs
1:06:22 Realities of selling enterprise AI software today
1:11:02 Building a sales team from scratch
1:15:21 Reflecting on Ada’s scale + doubling the last six months
1:16:41 Biggest software category of all-time ($750B)
1:19:51 Why AI is still under hyped
1:21:01 Ego is the biggest inhibitor to AI adoption
1:23:33 How AI will fuel explosion of creativity and productivity
1:25:20 Large companies will benefit the most from AI
1:27:41 Multi-modal language models and autonomous computers
Referenced:
Try Ada
Find Mike on X / Twitter and LinkedIn
👉 Stream on Apple, Spotify, and YouTube
Transcript - (read on Rev)
Find transcripts of all prior episodes here.
Turner Novak:
Mike, welcome to the show.
Mike Murchison:
Great to be here, Turner. Thanks for having me.
Turner Novak:
Yeah, thanks for coming on. You have a really cool product. Can you just explain really quick what Ada is for people who aren't familiar?
Mike Murchison:
Sure. Ada is an AI customer service company and our mission is to make customer service extraordinary for everyone. Our core view is that the quality of customer service that we all experience around the world today as consumers is abysmally low. And that AI, when properly deployed is going to forever change that. And so our product helps businesses hire AI customer service agents, understand their performance and rapidly coach them to improve such that they meet and increasingly actually exceed human performance. And we're doing that again in pursuit of upleveling what we can all expect when we either call a company, text it, email it, however we communicate with businesses daily, we're upleveling the quality of that experience.
Turner Novak:
So when you say hire an AI customer service agent, is this a human who utilizes AI? Is it literally an AI-powered agent that kind of lives in the ether? What is it exactly?
Mike Murchison:
It is by the sort of technical definition of an agent, it is a non-human entity that is powered by a group of large language models that is autonomously acting on your behalf to help your customers in any channel. And increasingly, this is being personified as an employee, a member of your team. Our objective when we work with our customers, the many hundreds of enterprise brands we work with, is to be their number one customer service employee. And what's so interesting about this is that we actually find that the businesses who treat their AI agent that we power most like a true member of their team, those are the ones that get the best results. And so there's this interesting phenomenon that we talk about a lot inside data, which is that if you look at all the most valuable companies in the world today, call it the Fortune 100, what they all have in common, one thing they have all in common is very strong management practices.
Our conviction and what we're increasingly seeing in our customer base is that the next generation of companies who properly adopt AI, achieve a level of efficiency that wasn't possible before, increase their growth rates in a manner that I think is going to catch people by surprise. They too are going to share something in common. They will also have extremely strong management capabilities, but those management capabilities will be not only limited to human employees, they'll extend to non-human employees too.
Turner Novak:
So managing basically different series of LLM prompts are all structured together. Am I thinking about that in the right way?
Mike Murchison:
No, I think it's actually the user experience is very, very akin to managing a human employee who you interface with in Slack. You email with someone who you have a one-on-one with regularly. We don't expose the... We can expose the underlying models of course, but the average user of Ada is an AI agent manager who lives inside our application. And that person is regularly reviewing the performance of their AI agent and instructing how it can improve in written language.
Turner Novak:
This is a human that is giving feedback to the AI agent?
Mike Murchison:
It's a human that's giving feedback and it's increasingly a team of humans who are formerly contact center reps, often the top performing members of the CX organization that get promoted into this new AI role. And that role is focused on maximizing the performance of their new team member. And it's pretty interesting, I think we've talked about this you and I, I think in the past where any sort of wave of extreme technological disruption of which clearly we're living through on with AI, it's very clear to everyone what roles are being disrupted. And it's very unclear to everyone what new roles are being created. And what's been so fascinating about our journey with Ada so far is that we have a front row seat into some of the new roles that are being created by this technological wave. And those are very specifically AI management roles.
And we play a key role in helping this next new gen of employees, human employees succeed in this new role. And so it's really exciting. It's very, very amazing to see the career trajectory of our customers. Again, many of whom are reactively responding to customer service inquiries in traditional CS rules. And now we're proactively managing AI agents and they as a result have become these AI experts inside their organizations. I think it's kind of like if you were to join a web team in 1998, people would be like, "What is the web?" And "What are you doing?" And it turns out that was an extremely valuable place to be working.
Turner Novak:
Can you maybe give me just an inside look at somebody who is one of these sort of AI customer service managers? What does my day look like? Or how am I interfacing with the agent? To kind of get an idea because I don't know about people listening, but to me I'm like, "Still sounds a little abstract." If you can make it super practical, what's actually happening on the day-to-day.
Mike Murchison:
Very specifically, if I'm working with Ada and I'm an AI agent manager, I will open up my Ada dashboard. Inside my dashboard, I will see how many conversations my AI agent has had with my customers in the last day, 30 days, year, whatever, however I want to filter. And I can see the performance of those conversations. How many of them did it autonomously resolve? How many of them did it do so with a five star experience? We focus inside Ada on maximizing the number of five star resolutions that you power. And then what are the opportunities I have to improve my employee? And once I have a quick overview of my performance where I'm performing well, where I'm not, Ada instructs you, provides guidance on, "Hey, here are the specific areas where I'm struggling." I'm involving a lot of our team to resolve issues around this particular type of order refund that our business is trying to deal with in our APAC and our Asian business.
Turner Novak:
And this is resolving like a human will say, "Okay, it looks like the AI is not quite getting there here." A human on the team is jumping in.
Mike Murchison:
Yes, the AI agent is involving human members of the team to resolve it. It's not fully autonomously resolving these issues. But Ada as a software is providing a window into why that's the case and is relying on the human manager to coach it to improve.
Turner Novak:
Oh, so Ada is actually saying like, "Hey, I know that I'm not getting this because of this." Or...
Mike Murchison:
Yeah, exactly. And so we provide as much as possible an understanding of the root issue that underpins why the AI agent isn't able to resolve the inquiry. And then we provide tools to the AI manager, the human manager, to address that root issue. And what's really amazing is that today we're pretty good at completing that feedback loop quite quickly. So the AI manager goes from diagnosis to instruction to improvement very quickly. What's super exciting to me is the depth at which we are now increasingly focused on resolving inquiries. So I'll give you an example. I met with a customer yesterday. They probably spend a quarter million dollars plus a year simply helping their customers reset passwords.
Turner Novak:
Whoa, okay. That's a big number.
Mike Murchison:
The scale which they operate at is such that it's an amazing amount of investment that's required to help their customers reset their passwords.
Turner Novak:
Why is that so expensive?
Mike Murchison:
Human labor is very expensive.
Turner Novak:
So it's basically the salary, the cost per hour spent of somebody just going in and being like...
Mike Murchison:
Someone responding to an email to help you reset your password, to pick... If it's on the phone, it's usually the most expensive channel. The average phone call cost in North America is north of $10 on average for a business. So extremely expensive. Now, there's a couple of different ways you might resolve that inquiry autonomously, right? You might say, "Hey Turner, I understand you're having trouble logging in. Follow these steps and you can reset your password by yourself."
Turner Novak:
That kind of exists, right?
Mike Murchison:
It exists, yeah.
Turner Novak:
That's pre AI, right? Yeah.
Mike Murchison:
Exactly. And that's a pretty surface level resolution. Maybe a slightly deeper resolution might be, "Turner, I've automatically reset your password for you, check your inbox." Better experience, right?
Turner Novak:
I've gotten that before.
Mike Murchison:
But the root resolution is actually probably we've understood that there's a bug on your iOS app that's creating a negative or a broken login experience. Increasingly, Ada's focus is on understanding that that's the root issue. And the solution is actually to push an update to your iOS app that eliminates the need for these issues resolving in the first place. And so that is where we are taking... When we think of ourselves as a resolution company that's going to make customer service extraordinary for everyone, it's not just about reactively addressing the surface level resolution. Increasingly, our job is to understand the root resolution and to autonomously resolve that. Such that you go from customer inquiry to production solution fully autonomously.
And I think that really is starting to provide a window in our eyes when we think about the vision for the role of customer service in the future. That is an example of a truly AI-native organization. A company that has quite a large feedback loop from customer input to improve product or service. The time at which it takes a business to do that today is extremely long and extremely expensive. But with the right AI and the right AI partner, our conviction is that that feedback will be far shorter and far faster ahead.
Turner Novak:
Because you might have in a traditional management practice entirely human business, you might have a customer support person, a manager, you might have maybe the engineering team, product is involved in this, design is involved in this of like, "Oh, we got to redesign the way that the password box works."
Mike Murchison:
We have to organize seven meetings to figure out if this is worthy of reprioritizing our existing roadmap. And by the way, our roadmap, our next sprint plan isn't for another six weeks. So you can just see it just cascades and how expensive and slow it is to just ship an update to your iOS app. And so I think that's a very good example of just how AI is really changing how businesses learn from their customers and how they act on feedback ahead. And that sort of scratches it, part of our impact on the world. We think about our company as today the average business is focused on actually reducing the amount of time they spend with their customers. That's how customer service works. Literally, customer service is a cost center. Almost every customer's experience leader is literally personally compensated to the extent to which they can reduce the number of contacts they have with their customer.
Turner Novak:
Wow, that sounds bad. That's not a good thing, I feel like. Good for the business.
Mike Murchison:
That's how we started Ada. We experienced that problem ourselves and we woke up and we were like, "Why are we doing this?" But AI, when properly deployed, changes the game. And the companies that work with Ada actually talk to their customers more than they ever have before. It's a counterintuitive thing for many people, but AI can bring you closer to your customers. And the companies that win ahead are absolutely going to be far closer to their customers as a result of deploying AI in a customer facing capacity, not further away. And they'll have more customer data than they ever have before to learn from and to improve their products and services from.
Turner Novak:
Because when I think about just generally how AI is discussed sort of publicly, especially, in a customer service arena, it's a lot about reducing the cost, humans, you don't have to talk to them as much. The AI will kind of automate and handle all this. I think we've seen some of those, the crazy headlines of whatever company automated whatever with just pure AI software. It's very almost counterintuitive to what you would think. How does that play out? So when you say talk to your customers more? Is that the AI-powered, AI-native software is giving you more touch points? Or is it actually the people on your team are talking more? Or is it kind of somewhere in the middle?
Mike Murchison:
Well, I think the two begin to blur pretty quickly. When I say that it's the total number of conversations a business is having with its customers, those conversations are mediated by an AI agent. But by virtue of there being far more conversations happening, and by virtue of every conversation being annotated, which we'll talk about more in a second maybe, this conversation data is now far more usable. And so there are far more people who are acting on customer data than ever before. And there's far more empathy for the customer inside large organizations than there was before. And I know that's completely transformative, it really is. It wasn't that long ago and still is the case for most businesses, you call them and they say, "This call may or may not be recorded for training purposes."
Turner Novak:
Oh, yeah. Does it actually get... Do they do anything with it? I feel like I get that all the time.
Mike Murchison:
Yeah, they do, but what's interesting is that it's almost always a sample of conversations that are manually transcribed by a human who is being paid 20 bucks an hour.
Turner Novak:
Is this a judge, they're gauging how well the person's doing in the call manually?
Mike Murchison:
They do it. They do take a sample. That's how they do QA, quality assurance for the phone experience. And just by virtue of how expensive it is to annotate and judge these calls or these transcripts, only a small percentage of them are actually annotated. AI has completely transformed that. Now, and this is true of Ada today, 100% of the conversations we power are annotated at a level that exceeds that of an average human annotator. So it's 100% coverage, you can learn from every conversation. You know the topic of the conversation, you know what you could have done better from the conversation, you know anything you really want to annotate on the conversation. What was the sentiment of this experience? What was the tone of it? How should we handle this differently? And so the level of granularity that you can learn from your customers has really forever changed. We've gone from starting with a sample to generalizing, to now being able to really slice the data and extract as many insights for whatever customer cohort you want in perpetuity when you're AI-powered.
Turner Novak:
What is maybe the most extreme or interesting or fascinating sort of example that has kind of come out of this? You pull out insane insight that somebody got from just being able to do this.
Mike Murchison:
So we had a customer recently who learned that they should build a new product. They're a large e-commerce company and they are missing a SKU. And so they learned from their annotated data that Ada provided that they're missing, in this case, it was a pair of sunglasses. They learned from trends in their customer conversation data that if they were to ship a pair of sunglasses, that that would probably perform quite well.
Turner Novak:
Just from customers bringing it up like, "I wish I could have gotten sunglasses in this order with my whatever else."
Mike Murchison:
Sort of, yes, they inferred that pretty clearly because there was an adjacent product that they sold that was negatively reviewed, or there were parts of it that were clear negatively reviewed. And then there were sort of requests, implicit requests for a pair of sunglasses that they didn't offer that would've paired well with some of their other products. And so they saw this, and because every conversation is annotated, we were able to extract the insight. They ship the new SKU and it's performing super well. And I think those examples are popping up all the time now. And it's a good example also of how customer service is not a cost center ahead. It's a revenue generator, right? If you learn from your conversations and those conversations inform your product strategy, your service strategy, your overall business direction, this is something that is definitely something you want more of, not less of.
Turner Novak:
Yeah. And intuitively that makes sense because if you prescribe to the YC mantra of talk to your customers, build stuff that they want, you're going to get it from the front line of communicating with the customers because it's going to tell you what they're thinking.
Mike Murchison:
Exactly. That's what I found so counterintuitive. Part of the founding story of Ada was-
Turner Novak:
Well, yeah, why don't we get into that right now? Because I know we were going to talk a little bit about just how it all got started. You're probably one of the most expert customer service people, just because of time spent in the trenches. So yeah, I guess how did it all get started? We're going back to the early days.
Mike Murchison:
Well, much like you were alluding to, David and I, we were obsessed and continue to be obsessed with our customer feedback. And one of the guiding principles that led us, David and I, to start our company, it has and continues to be this notion of bestowing ownership of our product to our customers. We want our customers to feel like our product is theirs. And how do you foster that? You foster it by listening deeply and acting on feedback.
Now, with our previous company, we started a social search engine called Volley, and it was growing quite quickly and it encountered this customer service problem. Simply put, we couldn't figure out how to scale our customer service operations in line with our user growth. And this experience of craving customer feedback really started to become challenged by our scale. And we started to face this perennial trade-off that every business used to face, which is how do I trade efficiency for quality as I grow in my customer service? I can't keep quality if I want efficiency.
And so Ada, like every other company, this is before we were Ada, our company at the time, we witnessed our wait times go up in our customer service operations and our number of customer complaints grow. And simply put, we woke up one day and we went, "This is crazy. Why is it that we are no longer treating our customers as people? But we're treating them as anonymous numbers that we're trying to keep at bay?"
I mean, these are the same people whom we're trying to bestow ownership of our product to. And so I picked up the phone and I cold called as many VPs of customer experience that I get my hands on. And I asked them, "Are you talking to your customers more or less as you grow?" And everyone said, "We absolutely are focused on talking to our customers less.
Customer service is a cost center. And I'm personally compensated the extent to which I can reduce customer contact every year." It's like, well, no wonder customer service is so bad around the world.
Turner Novak:
Everyone's doing their job of talking to them less.
Mike Murchison:
That's right. And that's why in North America, the average person will waste 43 days of their life, 43 days of their life, on hold. That's how bad customer service is. That's how long these wait times are, and it's the incentives all add up.
Turner Novak:
What is that? Like half a percentage point of your entire life you're waiting on hold on the phone? Something like that. It may be a little lower than that, but that's still a pretty big amount if you add it all up.
Mike Murchison:
It is, I think it's more than a month.
Turner Novak:
I literally only do customer service related things if I have something else that I have to be like, I'm working on this project, I'm going to be on hold with, insert massive business that we've all heard of and we hate their customer service, and I'm just going to wait for 20 minutes while I get this other thing done.
Mike Murchison:
And we all do that. And it's shocking to me that we tolerate it like we do. The bar is so low and our expectations are far too low.
So for us, we became very curious about that. We became quite opinionated that at some point this surely has to change. That the best businesses in the world were going to reverse the relationship with their customers at scale. That the best businesses in the world will find a way to crave customer input, not avoid it.
And we set out to learn firsthand. And we went back to those 12 VPs of customer experience that we had spoken with, and we asked if we could join their teams.
Turner Novak:
What?
Mike Murchison:
Seven of them said, "Yes, we'll hire you." And we joined seven different customer service organizations as remote customer service reps. And we worked around the clock inside their legacy ticketing systems to provide customer support.
And this as an aside was definitely a crazy time in my life personally, because most people in my life really were questioning what I was doing with my time.
Turner Novak:
You went from probably startup founder, maybe there's some glam from the outside to working seven customer service jobs?
Mike Murchison:
And it was stressful. It was really stressful. We were around the clock. We were waking up at three in the morning to answer yet another, "How do I reset my password inquiry?" And David and I just lived and breathed it and we learned a ton.
We learned that 30% of the inquiries we were handling, minimum, absolutely minimum, were just completely repetitive and mundane. We learned that half our colleagues, at least in some cases, 80% of our colleagues, would leave in the next 12 months. That's the attrition rates in customer service continue to be that high.
Turner Novak:
This is about half the team will turn over in a year.
Mike Murchison:
Yeah, in some vertical it's literally 80% will turn over in a year. And that's in part because the role can be so repetitive and mundane.
Turner Novak:
I did actually do a customer service role like internship. It was in freight tracking. I tracked freight shipments, and I guess it was more of an outbound customer. It was sort of customer service. It was like internal customer service.
It was a lot of, I'd call people, I'd make sure all the shipments that we had for our different customers. It was a 3PL. It was third party logistics company, but all we did was track the things for you. It was a really interesting company.
Mike Murchison:
Who were you calling? Were you calling colleagues or were you calling customers?
Turner Novak:
I was calling, I guess it wasn't really customer service, like customer customers. I was calling the people that ran these different brokerage, like logistic company warehouses. So if you've ever seen Arkansas's Best, YRC, Saia, any of these trucks that deliver things, I would call the people that were in the office or in the warehouse and be asking about the shipments.
And then we would relay that to customers. Sometimes we'd actually have to talk to them. By the way, as a kid in college, 20, I think it was 21st time ever talking on the phone, cold calling people. You learn a lot of skills getting outside your comfort zone.
Mike Murchison:
For sure. I mean, absolutely, I can imagine how... Were you trained before you did that extensively or were you just thrown in the fire?
Turner Novak:
I'm pretty sure I got trained for a week maybe to shadow someone. It's a lot of data input. You just do the same things over and over again. I would just get a list of things I had to check on. It was usually we had, it was some a problem shipment, or there was a delay of some kind.
So, it was maybe to your point earlier about up-leveling the customer service. I was problem solving a little bit. But again, it wasn't a glamorous job, and I lasted a semester and I got a different internship.
Mike Murchison:
I mean, I'm sure in my experience you develop a lot of empathy for your colleagues and for the role. I mean, you probably learned a ton about that business by virtue of speaking with customers and colleagues all day.
Turner Novak:
Oh yeah. I learned a ton about the trucking industry from it. Because I was just constantly talking to the people that work in the offices of all those businesses.
Mike Murchison:
That's exact same thing that we learned is that our colleagues were true experts of the customer experience and of the business actually as a whole by virtue of doing what you were doing, talking to customers all day, but they weren't respected.
They were viewed at being at the bottom of the hierarchy, weren't empowered to really share their feedback. They weren't actually empowered with credible data to justify their qualitative opinions. And that's in part because the software that was supporting them wasn't maximizing their productivity.
It was designed to sell more seats to the business. It was a misalignment of incentives and it wasn't annotating every conversation. It wasn't extracting insights that-
Turner Novak:
Even though it was software, it could have easily done that, in theory.
Mike Murchison:
It could have done some of it for sure. So we did that job for about a year. We became amongst the most productive agents on each one of the teams that we work for. As you may recall, there's usually a leaderboard internally of agent productivity, human agent productivity.
Turner Novak:
Everyone knows who the best people are internally on the team.
Mike Murchison:
And we were literally number one or two, that was our job. That was our mission. And we saw a couple of interesting things happen as a result of that. One, we saw that customer loyalty improved. We were specifically focused on trying to handle as much of the repetitive and mundane stuff so our colleagues could focus on higher value customer conversations.
And because we were responding to the mundane stuff so much faster than the company had historically, customer satisfaction in all our early employers all improved, and the customer loyalty that they experienced improved. Not exactly a big surprise, but if you respond to people faster, they like your business more.
Turner Novak:
Yeah, responsiveness is so important. It's such a key differentiator that is super simple in theory, but actually very difficult, especially as you get bigger in whatever capacity you're serving people in.
Mike Murchison:
It really matters. It's a sign of respect. Secondly, saw that our team retention improved, team attrition reduced. And that's because our colleagues were all handling much more intellectually stimulating problems. These are much more creative problem solving role, much less of a repetitive workflow based role.
And then third, we saw that the data that we were privy to, the conversations we were having with our customers contained so many valuable insights and sales opportunities that we just felt completely convinced that there was an insane amount of value being left on the table, just siloed in this department that few people cared about in an application that most of the business wasn't thinking too much about on a day-to-day basis.
And that's really when we set out to start Ada. We felt that we'd proven the value of better customer service manually on the business, and as an applied AI company today, every new hire, every team member I meet with, I remind us that we have manual routes. The way we build software, the way we solve problems is we understand the underlying manual process deeply first and then we seek to augment it with software and specifically with AI.
Turner Novak:
So is this what you were doing back... When was this by the way?
Mike Murchison:
This was nine years ago, eight years ago, I guess. Very early days of it.
Turner Novak:
So this is like 2017, 2016, depending on?
Mike Murchison:
2016, 2017, that's right.
Turner Novak:
So, were you starting to build some software?
Mike Murchison:
We were. Once we saw the impact of our manual efforts, we said, "How do we replicate this with more businesses, and how do we give ourselves more leverage?" Well, we're going to learn how to do that and because we have so much data, we're going to take an AI first approach.
Turner Novak:
Was AI around back then, 2016? Was it a thing?
Mike Murchison:
Yeah, there weren't large language models.
Turner Novak:
I'm being a little sarcastic here, but you guys are pretty early in it.
Mike Murchison:
Yeah, but I think a lot of people, I mean, I know you're joking, but a lot of folks, I think colloquially when they say it was AI around what they mean is were language models that are as capable as... Really, what they're saying is, "Is ChatGPT around? Was GPT 3.5 around?"
And the answer is of course no. But there were highly performant NLU models, models that were very capable at understanding human intent. There were not capable models that were capable of generating thoughtful responses and cogent responses. But on the classification side of things, classification was something that models existed that could perform quite well.
Turner Novak:
So, you were basically using that to ingest the data and figure out what they wanted, but then you had to manually still do the things or you had to build software that maybe wasn't quite large language models that were applying, but it was maybe some sort of other form of automation.
Mike Murchison:
Yeah, it was more workflow-based. It was AI to understand and then workflow to execute. And that massively improved our productivity, and we knew it worked because we didn't get fired from our jobs.
Turner Novak:
That's a good bar.
Mike Murchison:
That was the perfect A/B test. Our managers were like, "This is great. These responses are as good or better than for these use cases than we're doing today and they're far faster. Let's keep going."
Turner Novak:
So did they know that you were building a company?
Mike Murchison:
No. That's why it was the perfect test. It was the perfect test. Because I think there's so much risk when you're building anything, especially if something's hot today. The biggest risk I think in building an AI company and acquiring early customers is that people are working with you for the wrong reasons.
They're working with you because you're the cool AI startup person and not because you're solving a real problem. And we were very specific. We were very intentional about really serving as employees and not exposing how we were getting our work done and just ensuring that our colleagues and our managers were focused on the work we were doing, the work to be done. And that continues to be the case today.
What does Ada do in our mission to make customer service extraordinary for everyone? Our mission is to resolve customer service conversations just like a human. That's what a human's being paid to do. We're being paid to do the same thing. And so that grounding has been and continues to be very, very important to us.
Turner Novak:
So then you probably made a decision at some point you were like, This is working in some capacity." How did you decide, "We were doing this other thing. It was called Volley." At what point did you decide, "We're going all in on Ada."?
Mike Murchison:
It was a gradual process, I should say. There was a very clear moment where David and I had made the decision, "Ada's the future. We're creating so much value here, and we really care about this problem. We really care about up leveling the quality of customer experience for everyone."
But there was a specific moment where I had to tell that, explain the decision to Boris, who is our mutual friend who introduced us, who continues to be on our board today at Ada. Boris was the main investor in Volley, and David and I made the decision, we're going to go into this customer service world. We're going to focus on it.
This is before we'd written any software, by the way. So the decision actually to do what became Ada predated any software development. We just knew we wanted to be in the customer service space and we had started to be employees for companies, and we wanted to just immerse ourselves in learning how to do that and not be building Volley at the same time.
And I booked a lunch with Boris, he was in town in Toronto, and I sat down with him at a lunch spot on Spina Avenue and I was so nervous.
Turner Novak:
What were you nervous about?
Mike Murchison:
Well, this is a younger version of myself, first time founder perspective. I felt like we had failed. And so I was meeting with our single investor, we raised our entire pre-seed round from and I, my goal of the meeting was to tell him that Volley has not worked and we are going to explore the customer service world. We have no idea what's going to happen in it. And if you want your money back, you can have your money back, because we had some cash left in the bank.
Turner Novak:
This actually a sidebar. Were you getting paid from all these different companies you were working at?
Mike Murchison:
We were. We were being paid, but not much.
Turner Novak:
But you had seven jobs, so you were making at least a decent amount or...
Mike Murchison:
Yeah, we were being paid. The relationship we had with most of these businesses, one, it was on a performance basis, so we said... We were being paid, but it was directly connected to how helpful we actually were being, how many issues we were resolving.
So we had some cash flow, so that was helpful. But we probably burned through definitely more... Maybe I think we raised a $500,000 pre-seed round, and we burned through at that point in Volley's history, it was like year two of Volley, like we were down to maybe a 100K left in cash in the bank or less.
Turner Novak:
And you were just going to give them all the money back and just say...
Mike Murchison:
Well, I wasn't sure. It was an unusual situation for me at the time, and I think there was more context around how to handle this situation now, I think for founders maybe. But at the time, I was just super nervous and I wrote out literally what I wanted to say to Boris.
I literally wrote it on a piece of paper and brought the piece of paper and my intention was to read it. And so we sat down at lunch and I'm so nervous, the waiter comes by, Boris orders lunch, and the waiter asked me, do I want anything to eat? I'm like, "No, I'm good."
Turner Novak:
That's probably weird for him. He's like, "what's going on?"
Mike Murchison:
I know, man. The whole thing is weird. It's so weird. It Really speaks to Boris and I think what it means to be founder friendly. BC I take out the piece of paper, I start reading it, Boris goes, "Whoa, whoa, whoa, whoa." Puts this hand on a piece of push it aside and goes, "What ideas do you have? We got this."
I start to share my excitement about the customer service world and we're learning about it and the problems of scaling customer service at Volley. And he immediately went into, "That sounds awesome. Keep going." He made a number of intros to folks literally in that meeting, some of whom became early Ada customers weeks later.
And it was just such a great example, I think, for me, a very seminal moment for me as a founder where it really... It was a good example of I think the level of support I now hope to show other founders who I'm an investor in.
I think it's very easy to be an investor when things are going great. I think your true colors are seen when things are not, and so that was the transition point, to answer your question. It was a lunch with Boris where I was super nervous and we made the call, and he was encouraging and we went back to work as agents for the next few months after that.
Turner Novak:
I think there was also a point where my friend, Fahd, who works at Shopify, he said, I don't know if this is customer convo, but he said, you really started, when you were contemplating, all right, what are we going to do here? He said there was a walk that you guys went on. What happened there?
Mike Murchison:
Yeah. Well, Fahd became an investor in Ada too, as you know. Well, it was a very confusing time. We felt like we had failed with Volley. We had some ideas on customer service.
Shopify was growing very quickly, and I think Fahd was at Shopify at that time, and they were very interested in hiring founders and it continues to be a focus of Shopify's hiring. But we were definitely faced with a decision of like, hey, do we go and try to join what is clearly an incredible success story and since become even a greater success story?
Turner Novak:
Had they just gone public around that time 2017-ish?
Mike Murchison:
I believe so, yeah. Definitely, I remember talking with Fahd at length about, and Satish too, who went on to run a product for and then partnerships for Shopify about joining forces and growing as a product person within Shopify. So definitely lots of existential, deep questions about where are we going and what do we do next, and is this a transition point? Do we call it? Do we throw in the towel?
And I think Fahd was quite helpful as were a number of other friends in helping illuminate some of the early insights, helping me reflect on some, and David reflecting some of the early insights we had in the customer service world. And were quite encouraging in us continuing to learn more.
And I'm really grateful for that because sometimes you need folks to say, "Keep going. You can always join another company." But building something massive requires taking advantage of an opportunity that is usually fleeting. And so if you have any inkling that that opportunity might be in front of you, don't wait.
Turner Novak:
I think from what Fahd told me too, basically the entire team that wanted to hire you ended up investing in what became Ada, and then you guys hit the ground running.
Mike Murchison:
It was great. And so much respect and admiration for folks at Shopify and continue to learn so much from those guys. So, it did all work out.
Turner Novak:
So then what would you share, because I know today we talked about in the beginning, it's basically you built a AI agent, verticalized for customer service. We want to super simple with the buzzwords, but you didn't really go into this saying, I guess, LLMs weren't even a thing back then.
You built this from the ground up, first principles, very manual. What advice would you give to someone today that's thinking, "Oh, I want to build an AI agent for AX."? I mean, it's a popular category, a popular way to approach some of these different industries building software for them. How would you advise somebody to just think through that process and go through it today, if you were starting from scratch today?
Mike Murchison:
I would say understand the work that's being automated, and if you can understand the work that's being automated better than anyone else, you're in the best position to build the best software that automates that problem. And I really think about what is a startup's advantage. I think a startup's advantage boils down to the depth of problem understanding.
Your big companies are so distracted by so many different problems, but what can a small team do better than anyone? They can understand the problem better than anyone, by living it, by being obsessed with it. And I think that's absolutely core to building incredible AI native software because AI native software, unlike the previous generation of software, isn't designed to essentially give access for humans to collaborate, digitize physical workplaces, which is essentially what I think SAS 1.0 is about.
This next generation of software, of which I really believe Ada is a leader in and will continue to be, is about doing work. It's about the software itself providing value that doesn't require human collaboration or input. It may learn and improve from human input, but the previous generation of software was assisted by AI. This next generation of software is assisted by humans. It's a complete inverse.
Turner Novak:
Can you give an example of software assisted by AI?
Mike Murchison:
Most traditional SaaS that has a co-pilot or any sort of agent assist type experience where a human is the atomic unit of the product and there are models that are helping them understand data better or helping them slightly improve their productivity, but the whole workflow is designed for a human. That's what I mean by the former. The latter is the atomic unit of the software is an AI model or a group of AI models, and the AI models are seeking input from humans and increasingly we'll seek input from other AI agents to improve over time. That, in my mind quite clearly is the future of software and is going to be a prerequisite to up-leveling the quality of experience and customer service, for sure.
Turner Novak:
Yeah, I think an example you mentioned before the customer service is potentially upselling or helping somebody generate more revenue by solving adjacent problems. Do you first see any, I don't know, selective ways to make sure that you're not stepping on toes or fumbling the transfer between these different functions? I don't know if you understand. I'm not really sure how to ask this question.
Mike Murchison:
I know what you're getting at. I think pretty clearly in our world, AI is going to unify the customer experience. I think we're going to find increasingly that there have been relatively arbitrary handoffs in your customer service experience as a consumer.
Turner Novak:
It's literally like, hey, let me transfer you on the phone to the other department.
Mike Murchison:
Oh, I need to bring in a product expert. Oh, I need to talk to my manager. This really disjointed, oh, you have to call in. You call in and they'd have no idea who you are. It's a completely different person, completely different record.
Turner Novak:
You have to re-verify your social number or your phone number each time.
Mike Murchison:
All that is going disappear. When we mean we're making customer service extraordinary for everyone, part of that is a single unified experience you have with a brand, works across channels, has perfect memory. It's instant, it's effortless, it speaks any language. It can take action on your behalf. No manager required, no product specialist and handoff required. You truly are speaking with an expert who knows you and that person or that entity is capable of serving you in a way that no human or group of humans can serve you today.
Turner Novak:
With all this stuff changing, if I was starting a company today, we've been talking maybe a little bit about building the software but using the software. If I'm thinking about how all these different functions are blending together, how do you think that's going to change how we're building companies or managing and running companies today? Have you seen anything with customers or do you have any hypotheses on how things are evolving?
Mike Murchison:
I have a lot of thoughts on this. Absolutely, I think the world has forever changed in this capacity. There's a couple of things that are becoming increasingly clear to me, and this comes from some learnings around how we use models inside Ada and the expectations we have of our team to be regularly applying AI to augment their own day-to work and collaboration. I think the first insight that I've had is that I think it's clear to me that what the internet did is it put a premium on our ability to think critically and to scrutinize information. I think you and I saw that when we grew up, our whole education system was pretty focused in part on teaching us critical thinking. We learned about what a primary source was, we learned what a secondary source was. Because the internet proliferated information, we had to get good at understanding what quality information is and how to form an opinion from varying sources.
That's the impact of the internet and the web I think on, let's call it human cognition or the core skills that are required to operate in this new context. AI introduces a new set of skills that I think are going to be rewarded ahead. The core skills I believe that are critical to taking advantage of this technology are clear communication and the ability to delegate work. What I'm noticing is that the folks who are able to clearly communicate what they want and know that they can delegate this expectation to a non-human assistant, those are the folks who are increasingly pushing the limits of what AI can do and are augmenting their productivity more than other folks. Today, to your point about Excel, that's not yet a baseline expectation of companies. In the same way that when Excel came out, it was like a differentiator for you as an employee to say, "I know how to use Excel, hire me."
Turner Novak:
Yeah, I can Google.
Mike Murchison:
I can Google. We're in that phase right now of AI. If you can incredibly show that I can augment myself with ChatGPT, with this group of models, with these other AI tools, I think you do stand out today, but very soon that will become a baseline expectation and it will not be impressive. You will be at an extreme disadvantage if you don't know how to use Excel. I think that's the skill to really foster and to hone is clear communication, clear delegation, folks are doing that well are succeeding. There's a couple of things that are underneath that, that I've noticed. One is that the folks who are really good at using AI, because they apply this management lens to it, they tend to blame themselves when AI doesn't perform. In the same way that a great human manager when their team doesn't do what they expect, they look inward first. They don't go, "Turner, you're not capable." They go, "Oh, I probably didn't give you clear enough instructions." They're like, "I didn't set you up for success."
That's the default position, and the folks who have that default position in their relationship with AI, they tend to self-improve far faster and they tend to therefore push the limits and exceed the capabilities, or push the capabilities of what the models are capable of. We're living in this time, which I think is really important to ground ourselves and remember is that these models, most people, the vast, vast majority of people are barely scratching the surface of the existing models' capabilities. There's just so much latent potential to access in these models that very few people are accessing. It's also the case that the model capabilities are improving at a rate that is historically unprecedented. Again, I think that's the skill set that's being rewarded as a result.
Turner Novak:
You got me a little scared here because I'm definitely, I've been in the camp of the AI sucks. It's just not good enough.
Mike Murchison:
That's the natural position for us all to take. I think so many of us, myself included, our default position is to be threatened. Whether we're aware of it or not is to be threatened by AI and to assume that, and I think that's because so much of our sense of self is connected to the work we do and that's especially true in 2025. Our identity is very, in some cases, inextricably linked with our work. At least in North America. I think AI is very threatening in that way and that's one of the reasons why it can be so existential. It takes work to approach AI with low ego and an expectation that it can do your job better than you can, but those who do that I think are going to be the ones that truly realize its benefits.
Turner Novak:
It's really to reflect on it, you basically need to treat AI as an employee, something that is doing work for you and you are managing it. I mean the communication and the delegation is exactly what you're doing if you're a manager. It's like, I have this task I need you to do. Communicating how to do it, get it done and give it back to me. That's basically what you do all day if you're a manager really, at the end of the day.
Mike Murchison:
That's right, and that's why very interestingly, that's why so many people who are managers are great users of ai. It's also why some of the folks who like, yes, AI adoption is highest and OpenAI just released their data a couple of days ago about number one user group of ChatGPT are students.
Turner Novak:
Oh, I didn't see this data.
Mike Murchison:
That's true. It's also the case I strongly suspect that most students are barely scratching the true potential of the models they're accessing, and most of them are not even using the frontier, like the true frontier models. They're using the free version of ChatGPT. It's the human managers who have the skill set of clear delegation, clear communication who are doing that better. Interestingly, that's an example I think of perhaps it might be relatively new, but I think of my dad in this scenario. My dad is 70 years old and is an exceptional user of language models. I learned new things about how he's using perplexity and ChatGPT and Claude and other tools. The reason I do is because he is someone who's had the experience of managing many people over the course of his career and his usage, his learning curve is so much lower I've noticed than folks who are much earlier in their career actually.
Turner Novak:
Oh, really? It's interesting, the thing that it's fascinating with the perplexity, you'll give it a string and then it'll be like, give me all this whatever was announced today or when you see the ad, like things from today, things that were posted today that you add to the end of the string and it's always right. It's like just knowing what little tweak to make to the prompt that you're putting in significantly improves the output if you do it the right way.
Mike Murchison:
It's true. I think there's something that we've noticed in our business too, which is that there's a tendency to hold AI to a higher standard than you hold your human team. I think that's an important thing to pay attention to as well. For example, Ada is way more consistent in how we respond to customers than a thousand-person contact center, human contact center would be. Yet, it's often the case that the expectation around consistency is so much higher than what the human team has ever been able to perform and is able to be capable of. I think the other analogy where we see this all the time is an autonomous driving. It's like, we know that it's better. It's better, per mile, it's safer, but the first crash that happens, it's held to a different standard and that's because there's so many unknown unknowns that people feel that the technology and the context is different. I'm not saying that, that's irrational by any means, but I just think it's interesting and it speaks to the psychology of new technology.
Turner Novak:
You said something really interesting earlier that I don't want to forget about because I don't want forget to ask you this. You talked about people not taking full advantage of the models and using the frontier models to the extent that they can be used. It brings back, we've probably all seen those posts on LinkedIn or Twitter where people are like, "99% of people are not using chat TVD correctly. Here are seven things." Or they're like, "You're only using 1% of your brain." To bring that analogy to what you just said, what are some ways that you're seeing people almost low-hanging fruit of using the models a little bit better or even the top 1% of people, what are they doing with some of the models that most people maybe don't know about or haven't taken advantage of yet?
Mike Murchison:
I think a bunch of things. I think in the same way that 10 years ago, the expectation was that you don't ask for information if you haven't already Googled it. If you're talking to a friend or a colleague and they're like, "Where's this restaurant?" Or whatever, and you're like, "I don't know, Google it." If they haven't Googled that, it's just bizarre. Increasingly, I think it's the case if you don't show up with a thoughtful decision, you haven't used AI and that's just increasingly not acceptable.
Turner Novak:
Because you can basically type in any of these products, these consumer-grade AI products, just like, tell me about this thing. Help me form an opinion on this.
Mike Murchison:
Here's all the context and what I have. Here's my recent quarterly performance. Here's our strategy. Here's my four ideas. Tell me what the optimal solution is and why. Reasoning models are so capable now of making exceptionally clear and thoughtful decisions that if you don't show up in your conversation with your colleague, your manager, your report, with that as a starting place, I just don't think you're doing your job. That's I think a very clear example of how language models are changing work, just a baseline expectation that you're editing a thoughtful decision. You're not making them on the fly yourself nearly as often.
Another area where I think just basic low-hanging fruit is if not recording, if you're not transcribing your meetings, you're not utilizing, you're barely scratching the surface of what AI can do for you. Meeting transcription is just like the root of so much productivity on lock from AI that I think it's just an important starting place that every company have a workflow where they're transcribing all digital meetings and increasingly in-person meetings I think are really important things. Think about how much time is wasted in drafting a follow-up email with a customer today. In a 10,000-person company, it would be shocking the number of hours that are expended simply on folks spending an hour reviewing their manual notes, drafting a thoughtful response like that. There's so much low-hanging fruit that is catalyzed by meeting transcription.
Turner Novak:
This would be a way to take advantage would be after we get off this call, me and you, my email on behalf of something that's transcribing this, my email automatically drafts up an email. Mike, it was great hanging out today. Loved hearing the founding story of ADA and the tips about how to use LMs better, really going to change my life. Thank you so much, and have that ready to go and send it.
Mike Murchison:
It should be automatically drafted for you, and so that's a one-click send. Absolutely, I think that's a really good example. By the way, you're assuming you're selling me something, your Salesforce record should be automatically updated. A member of your team inside the studio should be tagged to send me some swag afterwards or whatever your workflow is. These whole processes that take many days and many hours of human collaboration are going to disappear, increasingly are disappearing.
Turner Novak:
It's not like these people don't have jobs anymore, it's just that you are helping them do their job better or more efficiently or faster.
Mike Murchison:
That's right. I think we'll see if you're a customer-facing person, we'll see quotas go up because you're able to close more deals, you're able to speak with more customers. We'll see the number if you're a post-sales person in an enterprise software company, number of accounts you manage will increase. Yeah, we'll just see productivity grow and I think overall we'll see revenue per employee at the top line of companies grow a lot. That ratio will continue to widen as we've seen the foundation model layer and the infrastructure layer. I think what's the company Nvidia has just absolutely insane revenue per FTE. I'm not sure what it is at this point, but it's many millions of dollars per headcount.
Turner Novak:
I've never actually seen it, actually, now that I think about it. I feel like some of them can be a little skewed, like Amazon I think is pretty low because they have million warehouse employees.
Mike Murchison:
That speaks to the opportunity, why are they invested so much in automation and their warehouse? Well, in part because I think they probably have an opinion around what their revenue per FTE should be in the next five years.
Turner Novak:
Yeah, true. You hit on something interesting, like using AI to close deals faster, do sales faster. What have you found in enterprise AI, like selling AI-powered software to large businesses? Anything that you're seeing or just general advice out there for founders trying to do it?
Mike Murchison:
There's a premium right now I'm being highly empathic around, the buyer's experience today, like high empathy for what it's like to actually buy enterprise software. If you're pick a department head, if you're a CFO buying financial software, if you're a VP of customer experience or CCO buying customer experience software, and it is just unbelievable. There's a couple of things that are happening. One, there's board level pressure that didn't exist before to apply AI, so you have the board and your CEO breathing down your neck saying, "Turner, what are you doing with ai? I saw that Klarna just automated equivalent of quoting 700 employees. I see that companies like Ada are automating 50% plus of companies like monday.com's customer service in their earnings report. What are you doing?" You're getting that pressure that didn't exist two years ago, nearly to the same extent.
You are feeling this existential pressure like we spoke about and we're like, "What's my job if AI is here? If this model can respond to our customers as well as I can, what is the future of my job? That feels uncomfortable and uncertain and I don't understand this technology well and what does it mean for me?" Then third, you're being bombarded with inbound from a bunch of companies you've never heard of who are saying, "Turner, buy my software." By the way, when you hop onto a call with one of those companies, they show you a demo and you have no idea if it actually is real.
Turner Novak:
What's going on there? I run into some of the similar things as well.
Mike Murchison:
It's true on the investment side, too.
Turner Novak:
Yeah, that's right. You got a lot of really impressive demos that are literally on video. It's not even the actual software being used.
Mike Murchison:
Yeah. Well, I think what's going on there is that it's relatively easy to give a great AI demo and it continues to be, it's a very different thing to deploy AI in production at scale.
Turner Novak:
Why are the demos so easy compared to actually doing it in production?
Mike Murchison:
I think the cost of software development is falling quite precipitously. That's one reason, and there are more people who are therefore trying to start software companies. I think it's both those factors. I also think that we're at the beginning of a completely new technological wave. AI is impressive, it's magical, and by virtue of it just being AI that for most people who see it, they are impressed. It's very impressive to pick a category, to listen to a voice AI agent that sounds like a human.
Turner Novak:
The voice specifically has been really interesting to me, and I think one interesting realization was there's a lot of maybe older folks who just, maybe they never really did the food delivery thing. They've never really been on their phone, like using a mobile phone, whatever, like ordering from DoorDash, but if they can pick up their phone and call in a takeout order and place an order for delivery or your groceries or something, that might actually increase the adoption of some software. You basically created a new interface for people to use it.
Mike Murchison:
Totally. Yeah, completely. I think absolutely. We see this with it with Ada voice. It very much changes our customer's relationship with the telephony channel. It used to be the case that because voice was the most expensive channel to service that you were focused on really reducing the number of phone calls, bringing people to digital channels. Now with voice AI, that strategy is very much being challenged. Why not drive-up voice adoption? Especially as you're mentioning, when I have a whole demographic of my customers, core customer segment prefers speaking on the phone. I think that, that's a good example of how AI-powered customer service improves things. It makes it more accessible for everyone. They can pick the channel that they want.
Turner Novak:
Yeah, that's fair. I think too, just generally speaking on enterprise sales, you had to learn how to build a sales team for Ada from scratch. I don't know if you'd ever specifically done that before. What was your general process? Any mistakes that you made? How did you structure the team, all that stuff?
Mike Murchison:
So many. I mean, there's a few things that I've learned and continue to be really important to us at Ada. The first is that we really and seriously view ourselves as a partner to our customers, like our offering is both a technology platform and a partnership. These two things have to work together for us to power a single experience for you as the business that dramatically improves your customer experience. It's like, the aspiration here is in the same way that what makes an Apple product magical is the tight integration of the hardware and the software that only they can do. That is what our aspiration is with the Ada experience.
When we partner with you, it's because our platform and our partnership work together seamlessly. So a lot of the learnings have been under how to get our platform and partnership to work together seamlessly. One way we've learned to do that, and again, the aspiration is for us to get you promoted. So we align our interests, our product success with you as the software buyer's career success, and that has been very helpful and continues to be something that we're really proud of. We want to make sure that when we work with you, Turner, as the chief customer officer of Peel Studio, let's say, that your career will be accelerated. You will be the COO as well at year one of our relationship. So that alignment has been really important.
And then I think, finally, the last learning has been to really be unbiased and check our thinking in how the buying experience should evolve. I think there's a lot of... Especially in enterprise sales, there's a tendency to be quite prescriptive and to assume that enterprise sales, enterprise sales is how the world works and there can be a lack of empathy for the actual experience of purchasing software. And so we are always iterating and improving how Ada's actually purchased, and we ground a lot of that aspiration in restaurant metaphors, actually. We think a lot about what is the experience of dining at an amazing restaurant. The server is a guide. The server is giving you a menu that's quite prescriptive, actually, but what we think you should probably purchase.
Turner Novak:
Yep. They're giving you the recommendations, their favorite dishes.
Mike Murchison:
But they're also attuned to the overall experience that you want to have and what's most important to you. And it's about so much more than just selling you food. It's about an overall experience that's being fostered and really orchestrated by a really talented group of people, some of whom you see directly and some of whom are back of house who play a core role in giving you the experience that you don't have a direct relationship with.
Turner Novak:
It's like the whole, "Give my kind words to the chef," or whatever, like, "Great food."
Mike Murchison:
And it's an amazing experience when the chef comes out and goes, "What did you think of the meal? I made this especially for you." It's like, "Wow. I got to meet the chef."
Turner Novak:
Yeah, that's like the CTO or the head of engineering coming out or head of product.
Mike Murchison:
Exactly. Exactly.
Turner Novak:
Or the CEO. The CEO comes to the meeting.
Mike Murchison:
That's right. Like, "This is my favorite meal. I made it for my daughter on her wedding. What do you think of it?" You can push it, but there's so much to learn from these experiences. Again, it brings us full circle. Right? Do it manually first, understand what works in the physical world, then apply technology to scale it.
Turner Novak:
And speaking of scale, we haven't really talked much about the actual Ada business, how much traction you guys have. I don't know. What are you willing to just share about where you're at today? Over the past eight years, you've gone from manually being customer service agents to now what's it at today, beginning of 2025.
Mike Murchison:
Well, we announced last week that we've now powered more than 5.5 billion customer service interactions for businesses globally. We work in many different countries at this point. We power conversations in more than 50 different languages. We now have customers who are autonomously resolving more than 83%... 84% as of this morning is the new high watermark of customer service interactions across all channels. And we are doubling our conversation volume every six months. And so it has been a very exciting trajectory. We've been very fortunate to have raised a considerable amount of capital and now be very capital efficient. We've raised about $200 million from investors including Bessemer, Accel, FirstMark, Version 1, Boris. So it's been an amazing journey so far and we are very early in what is, I think, the biggest software category of all time. More than $500 billion is invested every year in human customer service agent labor. If you include outbound, like the role you played in part, it's probably closer to about 750 billion globally.
And over the next few years, that spend is going to flood towards AI, and our objective is to make customer service extraordinary for everyone. And as part of doing that, we expect that a very material percentage of that spend will be diverted towards the Ada partnership. We are an incredibly compute-intensive business, and because we're an AI native company that is doing reasoning and runtime to autonomously resolve customer service increase, we are developing very close relationships with the cloud providers who are running the best-in-class language models. And so I think it's been very interesting for us to both learn just how capable Ada is increasingly for our customers. And we think by the end of this year, on average, Ada will be more capable than a human customer service agent. And the result of that capability is that Ada's extremely compute-intensive and I think that's allowing us to get creative with our customers, particularly our enterprise brands who have purchased a lot of compute from the large cloud partners and they can increasingly utilize that compute expenditure to partner with Ada.
Turner Novak:
How does that work exactly? So you might have a customer, let's say, a Fortune 50 company that books 10 million or 100 million... I'm actually not familiar with what that scale might look like, but big number, with Azure or with AWS that they commit to spending on a go-forward basis, and then they have to use it?
Mike Murchison:
That's right. So they've pre-purchased $50 million in compute from Azure, AWS or GCP like you mentioned. And by virtue of Ada being on the Azure marketplace, which we're now close partners with Azure, we're also on the Amazon Marketplace, they can use their compute agreement to purchase Ada. And so we are helping them utilize their credits effectively. And because Ada is so compute-intensive, it tends to be a really good investment for businesses to put their Mac, if in the case of Microsoft, to work with Ada.
Turner Novak:
Yeah, it seems like maybe a common thread of some of the prior conversations I've had in the podcast is, in some way, AI is still underhyped. And based on everything we've just talked about over the past hour, hour and a half, I feel like maybe you feel the same way. I don't know.
Mike Murchison:
I do. I do think AI is underhyped. I think that there are some circles on X where that's not the case, but on average, I think it absolutely is underhyped. I think the average person has been fatigued by AI development and they've habituated to just AI being here and ChatGPT being a thing they've tried or maybe use weekly for basic stuff and they haven't fully appreciated what it actually means to have non-human intelligence that is essentially free in our day-to-day and, by the way, is getting smarter every single day. And so the implications of that are completely profound. We know what a few of them are at Ada and we're helping shape a handful of them, but there are many more that lie around the corner that we don't know. And that's what makes it such an exciting time to be building.
Turner Novak:
What's the biggest inhibitor to somebody you're probably taking advantage of AI?
Mike Murchison:
For sure, it's ego. Whether you label it ego or not, but it's a self-consciousness of fear that the models can do what you identify with.
Turner Novak:
You might be worried that they're better than you and you get replaced or you just don't think it could be good enough so you wouldn't...
Mike Murchison:
Both. Both those things. Right? I think there's a tendency whether, again, we're conscious of it or not, to use something and to actually hope that it doesn't do as good a job. Because if it were to do as good a job, that would force some pretty existential potentially, at least at minimum, some deep thinking about what our role is and what we're here to do and what we're about. And I think it definitely is the folks who suppress that tendency to be defensive who are taking the greatest advantage of AI today. And in my view, it's just a matter of time. We're all going to have to do this. And so why not approach it with creativity, and curiosity, and open-mindedness right now? I think the advantages of that are really, really, really tremendous.
Turner Novak:
Yeah. I mean, it's just like software in general. If you were an accountant and you said, "I don't use the computer to be an accountant," someone would be like, "I'm not using you to do my... You're going to do this manually on paper? What?"
Mike Murchison:
That's right. Yeah. And I think a lot of the old way of working will continue to persist for sure, but it will increasingly be people still ride horses, of course, but it's a hobby mostly. And I think many of the things we do today will persist and be enjoyable for some people to do, but they won't be the most productive and pervasive way of getting the work done or of creating something new.
Turner Novak:
Yes. You talked a little bit about maybe how the world might change or economic impacts, and maybe shorter term, it's harder to... Maybe it's harder or easier to predict. Maybe longer term is also harder or easier to predict. But how have you thought through that? Just how you think the economy or... I don't know, labor markets, humanity will change? Maybe we're getting a little philosophical here, but what all of you thought through just as AI continues to just get better, cheaper, faster, et cetera? What's going to happen?
Mike Murchison:
I think there's going to be an absolute just explosion in creativity and productivity. Both. And that will be fueled by AI adoption and it's the economies that are the most AI-literate the soonest that'll bear the greatest and most immediate benefits from this. But it will be ubiquitous over time regardless of where you are. And I think what is both uncomfortable and exciting about it is that we don't know what these new jobs are going to be. Ada has played a role in creating some of them, the AI management role, but they're in the same way that when...
Pick a technology. The combustion engine was invented. No one could forecast or foresee that the motel industry would be created as a result, which is downstream of the road and industry being created as a function of the combustion engine that produced the automobile. It's impossible to know what lies ahead and just the amount what these new occupations and roles are going to be. We just don't know what they are. But I think that the impact that AI has will be far greater than that of the combustion engine. And as a result, there will be many new industries that are created downstream in the years ahead.
Turner Novak:
And you've also said something interesting, I can't remember, I think it might've been when we were talking a couple of weeks ago, it might not have been live, but you mentioned that you think that large companies are probably going to benefit the most from AI, which I would just naturally think the opposite because we just talked about, "Oh, you can be more productive, you can get more units of work done." We've probably all seen those videos, but there's going to be a billion-dollar company with one employee, but you actually think larger companies are going to benefit more. Can you talk through that? Because that's a little non-intuitive.
Mike Murchison:
Yeah. I think it depends on what timescale we're talking about benefiting. I think what I'll share is that at least I'm opinionated that big companies, in the short term, can benefit far more than I think most folks appreciate from AI. And one of the core reasons for that is because of how structured large companies are compared to small companies. Large companies document things, large companies have clear processes, large companies are set up to onboard a new employee in a repeatable fashion.
And when you treat AI like an employee, which is how you should be treating it if you want to maximize your return from your AI investment, it's much easier for AI to navigate an environment that you've built systems and processes around. And so that is pretty interesting because that's perhaps a counter narrative to how technology has been adopted historically, where the bigger the company, the slower moving they are, the less innovative they are, the harder it is for them to apply new things. There are other factors that make AI adoption harder to the bigger the business, but structurally, there are some advantages that big businesses have and I think that's very exciting if you work in a big company and if you're a leader who cares about AI transformation.
Turner Novak:
Yeah, because bigger companies, to your point, have more rules, more structure. And that is what LLMs are very good at, just give it rules to follow and it will do exactly what you want it to do if it's possible.
Mike Murchison:
That's right. I got structure, I got knowledge, I've got processes, and I've got a lot of people who can help me.
Turner Novak:
One other thing with running Ada, what are some of the big unanswered questions around AI or maybe things that you're the most interested in or following, big changes maybe we're expecting? A little bit of an open-ended question, but depending on where it takes you.
Mike Murchison:
I mean, there's so many. I'll share maybe one key one that's very topical for us right now. We're extremely interested and excited about multimodal language models.
Turner Novak:
So what does that mean?
Mike Murchison:
It means that for a model to be able to input any modality and output any modality, meaning... A simple example of this might be an experience whereby Ada can listen to you on the phone and send you a video that is automatically generated to your phone... Text you a video that helps you navigate or troubleshoot an assembly issue you're facing as an example. And so increasingly we now have the experiences where we're doing a lot of voice-to-text where you can talk to Ada's voice AI, Ada, depending on the inquiry you're facing, can say, "Hey, Turner, I'm happy to send you a text message with this so you have it for later." We'll start to blend these channels. But truly multimodal omnichannel experiences are something that is going to be enabled by advancing what language models are capable of inputting and outputting across modalities. And that's going to unlock a whole new level of creativity in the customer experience that we're pretty pumped about. That's one area that we're looking a lot at and are investing in.
Turner Novak:
Is that possible yet or will be soon, you think?
Mike Murchison:
There's some demos, like we said before.
Turner Novak:
It's okay. Not quite there yet, but hopefully soon.
Mike Murchison:
Not quite there yet, but it's advancing quickly. And if you squint, you can see it. The other area that, I mean, many folks are thinking about, and Ada's no exception, is the ability for models to operate computers autonomously. And so there's a lot of data and there's a lot of systems that don't have APIs, there's no application programming interface for them, but there is a GUI, there is a graphical user interface. And AI, as we've seen probably with... You've probably played with Anthropic's computer use, and you've played with know ChatGPT's Operator, AI can increasingly navigate human-computer interfaces. And that is very exciting because it means that, ahead, Ada will be able to help customers in a way that formerly required a lot of manager approval or a lot of navigation in the backend that was only a group of humans could perform, and therefore resulted in a very slow and often disjointed customer experience.
Turner Novak:
Yes. That might even be going back to the earlier password reset example. It's not your password reset, you got to update your app. It can actually update the app for you in theory.
Mike Murchison:
That's right. Yeah, exactly. But, I mean, that I would even say isn't the deepest resolution there though. The deepest resolution is the app is fixed.
Turner Novak:
The app is fixed. Yeah. I feel like we could keep going for another hour. I know we have a couple of topics we didn't get to, so maybe we'll do another recording sometime, but this is a lot of fun. Thanks for coming on.
Mike Murchison:
Thanks for having me, Turner. Really enjoyed it. Yeah, we should talk... I know we wanted to talk Canada, US. Ada's half Canadian, half American, especially in light of the Trump tariffs is a very interesting topic for us to explore. So maybe another time, we'll talk more deeply about that.
Turner Novak:
Yeah. Plus we're coming off the gold medal game last night. I feel like, for some Canadians, that was a big moment.
Mike Murchison:
We defended our sovereignty.
Turner Novak:
Another year of freedom.
Mike Murchison:
That's right.
Turner Novak:
Awesome. Well, this is a lot of fun. Thanks for doing it.
Mike Murchison:
Thanks, Turner.
Stream the full episode on Apple, Spotify, or YouTube.
Find transcripts of all other episodes here.