Ada: AI-First Customer Service with CEO Mike Murchison

Full episode

Matt Turck [1:26] Mike, welcome. You are the co-founder and CEO of Ada, the leading AI-powered customer service automation platform. Ada was founded in 2016 in Toronto, Canada, and has grown very impressively. Today, the company has over 300 customers using the platform, including Meta, Verizon, and Shopify. And also, Ada has raised over $190 million in venture capital money, including a $130 million Series C in 2021, led by our friends at Spark. And I should add that I am a proud investor and board member in the company, having led the Series A in 2018.

Matt Turck [2:07] But hopefully, we can keep this conversation very unbiased and interesting for everyone who's interested in learning about AI, generative AI, and how to build generative AI companies in this new age. So I'd love to start with the founding story of Ada. Why did you choose that specific big problem at the time? And how did you go about starting the company?

Mike Murchison [2:34] Well, my co-founder David and I, as you know, were working on a completely different company. And I guess it would have been 2015. It was a B2C social search engine that kind of caught lightning in a bottle and encountered this customer service problem as a result quite early on in the company's growth. Namely, we couldn't figure out how to scale our customer service operations in line with our user growth. We had a freemium iOS app and accompanying web app, and we started to see our relationship with our customers change as our growth took off.

Mike Murchison [3:10] We went from treating our customers as real people, who we cared about and whose feedback we craved, to treating them as these anonymous numbers that we were trying to keep at bay. And I think, simply put, Matt, that just really bothered me. I'm a very product-oriented person. Part of what gets me out of bed in the morning every day is I love building powerful software that's delightfully easy to use. And I believe that that software is only built through incredible feedback that one gets from one's users and customers.

Mike Murchison [3:48] And I think that the quality of product, in many ways, is directly correlated to the extent to which customers feel ownership over it. There's this concept, I think, of bestowing ownership to your customers. If you talk to a customer who loves a product, they feel like it's theirs. And that feeling was eroding as our company was growing and as we were scaling our customer service. And so, simply put, I became really curious about that. And I picked up the phone and I asked a bunch of different VPs of customer experience and customer service how they were approaching their customer service operations as they were scaling.

Mike Murchison [4:29] And I asked them, hey, are you talking to your customers more or less as you grow? And all of them said, we're talking to our customers less. Customer service is a cost center, and I'm personally compensated to the extent to which I can reduce customer contact over time. In other words, our strategy is to increase the distance between ourselves and our customers the bigger we get. And my job is to reduce these costs. And when I heard that maybe the 12th time in a row, I became quite opinionated that these companies I was speaking with, who were some of the most innovative companies in the world, I really felt like they were operating on the wrong side of history.

Mike Murchison [5:16] At some point, I didn't know when, but at some point, businesses who figured out not how to avoid customer contact, but to crave it, those would be businesses that would confer themselves advantages in their respective markets. They would power a much better experience. And they would ultimately win. And it was when I felt the conviction that that was the case that, as you know, we made the tough decision to pivot that business into the customer service world. And I went back to those VPs and I asked them if I could join their team as a customer service agent.

Mike Murchison [5:41] And of the 12, seven of them said, sure, we'll hire you. And from a dingy office on the east end of Toronto, David and I worked for seven different customer service teams remotely at the same time, living and breathing customer service for what would become the first year of our company, Ada. And I'd say that that experience answering thousands and thousands of customer service inquiries inside Zendesk, Salesforce Service Cloud, and Genesys agent desktops at the time, I'd say that experience taught us three key things.

Mike Murchison [6:26] We learned, first of all, that 30% or more of the inquiries we were responding to manually were repetitive, mundane questions. Maybe more like 80% in some industries, some of the companies we were working with. It depended on what they were about. The second thing we learned is that the agent experience of responding to customer service inquiries inside this incumbent technology, this agent desktop, human-first technology, was incredibly negative. Nobody is getting out of bed in the morning going, I can't wait to spend more time in my agent desktop or my Salesforce Service Cloud.

Mike Murchison [7:02] That's not a piece of software, as robust as it is, and the software products that are like it, that people feel ownership over, that they feel a sense of delight or a sense of empowerment around. And the third thing we witnessed was that every one of our colleagues, they all wanted to offer a more modern customer experience through modern messaging channels. In other words, they wanted to turn on channels like WhatsApp and SMS and Facebook Messenger and Instagram DMs.

Mike Murchison [7:44] And that idea of communicating digitally through modern messaging infrastructure, modern messaging channels, was repeatedly rejected. It was a better experience, and it was repeatedly rejected because it was viewed as being antithetical to the customer service strategy of reducing the number of contacts that you have. And so it was very interesting to us to learn that this better experience that was universally regarded as a much better experience was being withheld despite the customer experience team advocating for it. And so it was on the back of these three learnings that we continued to fight the temptation to write a single line of code.

Mike Murchison [8:24] This whole exercise for us was fundamentally about experiencing customer service firsthand as agents and learning ourselves about truly the challenges of delivering an incredible customer service experience inside all the tools that were available at that time. It was on the back of those three learnings that we set out to be the most productive agents. And Matt, we literally became agent number one or two on the leaderboard of these seven companies. We just outworked each other. We pushed each other and we worked really hard.

Mike Murchison [8:52] And there's literally a leaderboard in these products. And David and I were literally agent number one or number two. And we saw some pretty amazing results as a result of our hard-earned efforts, which often required us waking up at three in the morning to answer our 100th "How do I reset my password?" email inquiry. We saw that churn went down in these businesses. It turns out if you respond to customers more quickly, they like your business more.

Mike Murchison [9:26] Big, big surprise. But it was great to see that validation that if you respond to people quickly, they're more retained. We saw that agent attrition rates fell, and that was a little bit more illuminating to us. And as many of our listeners familiar with customer service might know, agent attrition rates in customer service are exceedingly high. I mean, in the enterprise, customer service attrition rates are north of 40%. In the pandemic, they got as high as 80%.

Mike Murchison [9:58] In some industries, they're still hanging at 80%, meaning that 80% of your colleagues aren't with you after a 12-month period. And so it's very expensive for companies to train and onboard new employees when job satisfaction is so low and attrition rates are so high. What we found is that we could improve attrition rates, we could improve employee retention amongst our colleagues because they like their jobs more. We were handling all the complicated, very manual stuff.

Mike Murchison [10:21] And our colleagues could all handle the more complicated, interesting work that led to them being more retained. And then the third thing we saw was that the data that we were privy to, customer service conversation data, was just this powerful dataset that was underutilized. It contained sales opportunities, product insights, strategic info that we felt should have been informing company strategy, should have been informing different departments' decision-making, but instead was siloed in this one department that no one really thought of or cared much about.

Mike Murchison [10:59] And it was on the back of that very manual effort that we felt like we'd validated our software before it even existed. And so the problem for Ada, what became Ada, to solve was: how do we replicate this value we've created manually? How do we do that through software? And it was there that we took an ML approach because we had access to so much data, and we focused on making the ML techniques we were employing as easy to use as possible because all our colleagues were non-technical.

Mike Murchison [11:39] So Ada became about empowering the customer service organization to elevate their customer experience using AI. We wanted to give them control, put them at the steering wheel. And the big aha moment for us that really birthed our company was when we built the first version of Ada and we let it run inside the seven companies that we worked for, and we didn't get fired. And when that happened, we knew that we were onto something.

Mike Murchison [11:51] And the rest of the last six and a half years has been about bringing Ada to more companies and improving its capabilities.

Matt Turck [12:25] Great. I love that story so much. Thanks for telling it so well. Talk about living and breathing the problem that you're trying to solve. So fast forward to today. Obviously, it's been a few years with a lot of building, a lot of progress. What does the Ada platform look like in terms of capabilities, integration, overall design philosophy, different components—any way you want to take it?

Mike Murchison [13:00] Sure. Well, our product's mission hasn't changed. Its capabilities against that mission have changed significantly and continue to. Our product's vision is fundamentally to be the AI platform that helps businesses automatically resolve the greatest number of customer service inquiries across all channels, across all languages, with the least amount of effort. And so, to actually deliver that, there are several really key components to our software. The first is we're all about, as I mentioned earlier, low effort, ease of use. That translates into no-code builder environments, very easy-to-understand reporting.

Mike Murchison [13:32] It's deeply integrated into your business environment, and really the enablement of non-technical customer experience leaders to build with and learn from AI tooling. So that's the first component. The second component is that we integrate into all the data across your company that's most important for your customer experience. And so we have a whole set of knowledge integrations that make it really easy for you to build an intelligent bot that is grounded in the knowledge that your team has created.

Mike Murchison [14:16] That could be in a help center, it could be in internal documentation, could be in technical documentation, could be in a variety of PDFs you have stored in some server you don't know about. We unify all that knowledge really seamlessly and make it easy for you to automate conversations with it in an intelligent manner. Third, we have an ability to sit seamlessly on top of your existing agent software and hand off whenever your AI needs the help of a human agent. We hand off seamlessly into an agent who may be a call center agent who lives in Genesys or a support agent who lives in Zendesk, for example.

Mike Murchison [14:49] And so there's a lot of routing and agent desktop integration work we've done to make that experience seamless. And then I'd say finally, the last major component is we're really big on going beyond simply enabling you to automate an answer or a question for one of your customers to enabling you to take complex action on their behalf. And this is where Ada really starts to become quite magical for many of our customers and how they deploy us. It's pretty magical for an AI to be able to understand who you are, have perfect memory of your entire relationship with you, have access to all the systems that formerly only the most privileged manager could control on your behalf, and be able to do so and be at your command any time of day, in any language, in any channel that you need.

Mike Murchison [15:34] And that's what's resulting in an experience that I think is increasingly bringing us much closer to our vision, which is a world where every customer interaction is resolved by AI.

Matt Turck [15:36] What's an example of an action?

Mike Murchison [16:00] An action would be anything that formerly required a customer service agent to click around in a backend system to perform. So it could be an order refund, could be an account lookup, it could be a payment that's processed—anything that formerly required access to a separate system and is making some form of API call to execute.

Matt Turck [16:43] Great. So as you mentioned upfront, Ada has been an AI-native or ML-native company from the get-go. But obviously this was before the current hype and craziness and excitement around generative AI. So you clearly reacted to this very quickly because you announced, first of all, a partnership with OpenAI, I think, at the end of last year, and then launched a suite of generative AI products in April. But maybe take us behind the scenes. How does a company like you that has an AI product, has an AI/ML team, when a massive innovation like generative AI comes to the fore and becomes not just something that engineers are excited about, but that the whole world gets obsessed with?

Mike Murchison [17:29] Yeah, there's so much to unpack here, so I'm excited to get into it with you. So I think overall, I guess my overarching view is that models—we live in a very interesting moment in tech history, because I believe that model capabilities continue to outpace their application. And that's new, at least in my view, with regards to how builders of technology operate.

Mike Murchison [18:07] In other words, many of the foundation models are capable of doing far more than we realize still, literally the ones that anyone can access today. And therefore, I think it behooves the companies who are building with large language models, certainly those that want to be building at the frontier of AI, to organize themselves in a way where they can discover what those new capabilities actually are. And we've thought a lot about this inside Ada.

Mike Murchison [18:35] We, as you alluded to, have certainly benefited from the fact that we've been an AI-native company from day one. What does that mean? It means that the atomic unit of our product has always been an AI model, not a human. Our pricing and packaging has always been rooted in usage-based pricing. Now, as we'll get into, I'm sure, increasingly value-based usage pricing. But I think there are still learnings here that apply to any company, even SaaS businesses that are very human-centric and have business models and packaging connected to, say, selling seats.

Mike Murchison [19:14] I think that there are certainly lessons that still apply. A few things that I would encourage folks to think about. The first is that the actual skill set required to discover new model capabilities, new model use cases, I think has changed. I think it's actually no longer the case that you need a PhD in machine learning to understand and push the frontier of what an ML model is capable of. And I would actually say that I believe that the folks who are going to discover the newest capabilities are actually less likely to come from those backgrounds.

Mike Murchison [19:44] It's not to say that ML teams who have those backgrounds aren't valuable. They are, and they're probably even more valuable than before. But the number of folks who are involved in discovering new model capabilities is going to expand and needs to expand quite dramatically. A good example of this recently was we enable everyone inside Ada with their own AdaGPT, essentially a language model that we encourage everyone to default to interfacing with anytime you're trying to do any work at all, period.

Mike Murchison [20:26] And we try to set the expectation that large language models are advancing so quickly that you need to expect that the model is capable of doing far more than you realize. And if the model fails, you need to assume that it's your fault, not the model's fault. We're trying to teach everyone at Ada to be good AI managers and borrow many of the lessons from good people management that we've learned. I think great people managers know that if one of their reports doesn't perform, it's their fault.

Mike Murchison [21:01] It's not the report's fault. They didn't communicate effectively or they didn't orient their report properly. It wasn't an issue with the person. It was an issue with them, the manager. I think the same thing is true with AI for those who are trying to build the frontier. But an example, to bring this back, of this working really well from perhaps unexpected sources came just recently for us. We do a lot of data manipulation and data analysis with language models inside Ada.

Mike Murchison [21:35] And we were struggling. We had a couple teams who were collaborating, struggling to manipulate a large dataset. And there were a lot of data scientists working on this problem. There were a lot of ML folks. Our product team was involved, and it was actually a CSM from a non-technical background who created the prompt that actually solved the problem. What they did is they instructed the model to behave more like a mathematics teacher.

Matt Turck [21:37] Hmm.

Mike Murchison [21:53] And it was that simple change to the instructions of the model that required the perspective-taking of our customer success manager that no one else who was more technical than that person had thought of, that actually led to the unlock. And I think that's just a small example of why I think the skill set of higher empathy, better perspective-taking, is going to be rewarded in this shift of becoming more AI-native and pushing the frontier of model capabilities.

Matt Turck [22:09] Mm-hmm.

Mike Murchison [22:12] There's a lot more I could talk about in that.

Matt Turck [22:31] Yeah. No, super interesting. And then at a practical level, on the Ada platform, you use, I think, multiple models behind the scenes at the core. Is that some OpenAI, some homegrown? What's happening there at that level?

Mike Murchison [23:04] Sure. So Ada initially started seven years ago really as an NLU company. We had a core model that we were always improving as a customer of Ada, and it was, and continues to be, an intent classification model. And we employed large language models throughout our history initially to augment the folks who live inside the AI coaches or bot managers who live inside Ada to manage and improve their bots over time. And our language models, the generative AI, were being used to really enable them to be more productive, but not to generate answers in runtime for end customers like we do today.

Mike Murchison [23:55] And so today that has changed. Now, because large language models have crossed a fundamental quality chasm, as we see, the best experience, the most highly resolving experience that you can build with Ada that's really exceptional, actually has a large language model at its core that is generating an answer and executing a workflow in runtime, so that every generation, essentially every conversation between your brand and one of your customers, is unique. And underneath the hood, as you alluded to, there's a bunch of different models that are involved in that.

Mike Murchison [24:31] And we take away the complexity for our customers around what those are so they don't have to think about it. And what's changed on the MLOps side—or it hasn't actually changed, but it's been supercharged—is the infrastructure we've built around switching between different models efficiently and in a manner that optimizes for what we call our resolution rate. And so we're becoming very good at being able to essentially automatically leverage the model or combination of models that's going to maximize the quality of customer service conversation at any point in time.

Mike Murchison [25:01] And we think that's really important, particularly in an environment where the models themselves are evolving so quickly and the number of models and their capabilities are evolving so quickly.

Matt Turck [25:14] So part of the idea is to future-proof the company by creating an infrastructure that enables you to hot-swap whatever model capabilities become available. Is that, to play it back, correct?

Mike Murchison [25:23] That's right. I mean, I'd say it's more than hot-swap. It's more like optimize a combination or ensemble of models that's going to maximize the quality of customer service experience.

Matt Turck [25:24] Mm-hmm.

Mike Murchison [25:42] And to give you an example, there's one example of a very simple version of that. At the code level in Ada, we make two different large language model API calls. We make a fast call and we make a smart call. And they leverage two different models at any given point in time. And we optimize from a series of a bunch of different models behind the scenes around which actual one qualifies as the best smart model or best fast model.

Mike Murchison [26:04] At a point in time. But it's very interesting to think about how the principles that govern human cognition—anyone who's read...

Matt Turck [26:05] Thinking, Fast and Slow.

Mike Murchison [26:30] Thinking, Fast and Slow are increasingly showing up at the code level in how we build software. And I think it's through that paradigm that we think about: we always want to be using the smartest model, and we always want to have the fastest model too. Both are really important to us. And that's how we think about ensuring that our customers are always working with the best technology at all times. And as you put it, we're future-proofing our customer service AI.

Matt Turck [27:15] Fascinating. So thinking about the implications of all of this, not just for Ada, but for customer service in general today and in the future powered by AI, one consequence that I've seen you and the company talk about is what it may mean to have an AI-first strategy, including the concept of going beyond containment and thinking about success in terms of automated resolution. Can you talk to this in more detail?

Mike Murchison [27:40] Absolutely. This is, I think, one of the exciting things about some of the innovation out of Ada this year. As you know well, as a board member of Ada, we've spent a lot of our company's history annotating customer service conversations manually, and we've developed a deep understanding of what a good customer service conversation is and what a bad customer service conversation is. But starting this year, we've now trained models, specifically an AI model, to understand the quality of a customer service conversation better than a human.

Mike Murchison [28:17] And this to us represents a fundamental shift in the future of how customer service AI is going to be measured and the amount of value that businesses are going to be able to get out of customer service AI as a result. I'll walk through what I actually mean by that, but the general principle here is that many of us are familiar with the fact that AI can understand and process an MRI better than a human physician. This has now just happened with customer service AI, where AI understands a customer service conversation better than a human.

Mike Murchison [28:51] What this essentially means is that most of the customer service AI industry over the last 10 years has focused on a metric called containment or deflection for measuring its value. What that essentially refers to is whether or not the AI powered a conversation that involved a human agent or not. But it doesn't actually address or understand the quality of the conversation that was involved. Turns out a lot of conversations that were contained by the AI, by the chatbot, weren't actually that helpful.

Mike Murchison [29:26] I think many of our listeners have had the experience of speaking with a dumb chatbot and being highly frustrated and perhaps just simply closing the chat or hanging up the phone and leaving the experience. I think most people continue to have a mostly negative experience with chatbots. I think Ada's playing a key role in shifting that perception, but that has been, I think, most people's historical or common experience. What resolution or automated resolution does is it not only measures whether or not a human agent was involved, but it also measures whether or not the conversation was accurate, whether or not it was relevant, and whether or not it was safe.

Mike Murchison [30:13] And so we're giving our customers this added level of intelligence where we're zooming in on all the conversations that they're powering with their AI, with their customer base, and we're optimizing, via our resolution engine, all their AI towards having more of these high-quality conversations and fewer of those frustrating ones. And we're so big on this measure that not only is it the North Star measure of our entire company, but we're increasingly enabling our customers to price according to it. So we're actually only going to charge you for a conversation that's automatically resolved, that meets this really high-quality bar for what it means to actually have a positive experience with your AI.

Matt Turck [30:59] Great. So to play back, what you're saying is that AI has had perhaps a step-function improvement in terms of its quality of results, and therefore not just Ada, but the entire AI-powered customer service industry should have a much higher bar. And it's time for everybody to step up. And that's the end of shitty chatbots, and instead everyone should focus on chatbots that actually solve the problem. Otherwise, it doesn't count and it's a terrible experience.

Mike Murchison [31:30] That's right. And I actually think this will be true for all of software. Because of the quality of large language models' reasoning capabilities specifically, I think we're going to see a major shift, not only away from seat-based pricing towards usage-based pricing, but actually in a very particular value-based type of usage pricing, where software vendors are increasingly only going to charge for the value that they're actually creating. And so in our case, that means we're only going to charge you for a really valuable customer service conversation that we powered, and not for ones that weren't.

Mike Murchison [32:02] And we're going to see fewer and fewer proxy measures for value in our software industry. A seat is just a pretty weak proxy for the value that the software is creating. A conversation is an okay proxy, but a really valuable conversation is one. And so I think we'll see this—you know the data infrastructure layer better than I do as an investor in it—I think that we'll see this trend continue far outside of just customer service.

Matt Turck [32:34] Great. So also as part of this AI-first strategy, you sort of alluded to it a little bit in the context of employees at Ada, but there is this concept of onboarding the model like you would a new employee. So if I'm a customer and I'm trying to work with one of these AI solutions, whether homegrown or through a vendor like Ada, what does that mean?

Mike Murchison [32:56] Let me zoom out here for a second. What does it mean to be AI-native? I alluded to this earlier when we were chatting: being an AI-native company means that your humans are augmenting your AI model, not the other way around. But the reason for that is because I think that we're at this very profound moment in tech history where AI's impact is going to be as profound, maybe even more profound, than the internet's impact itself.

Mike Murchison [33:43] And the reason for that is because if the web made the cost of distribution essentially free for businesses, what AI is doing now is making the cost of cognition essentially free. And so because of that, AI-native applications really need to be treated as employees. A truly capable AI application really is no different than a labor source that was formerly connected to human cognition, but it just so happens to be AI now. And so the best applications, I believe, are going to be onboarded in a manner that is not that dissimilar from the way you formerly onboarded a human employee.

Mike Murchison [34:23] In our case, we make it easy for our customers to onboard their customer service AI. They teach that AI to read, much in the same way that you'd give a new employee the ability to read all your culture decks and onboarding materials and job descriptions and whatever enablement. We make it really easy for you to train your AI with all the content across your company. We then try to make it easy for you to enable your AI to actually do things.

Mike Murchison [34:57] So instead of just a new employee who showed up who knows the policies, you want them to be able to actually get access to everything so that they've got access to all applications that you use. You can actually do work. And so for us, that looks like all the integrations that enable your AI to take action on your behalf, create a ticket in Zendesk, process a payment with Stripe, you name it. And then third, and this is where things, I think, get most exciting for our customers, is we enable your organization to coach your AI to improve over time.

Mike Murchison [35:33] And it's here where I think there's a really interesting intersection, or perhaps a longer podcast for another time, around the relationship between modern HR practices and AI-native applications. Like, there's a whole new school of thought here around, like, how do you improve? How do you give your AI feedback? What does a one-on-one with your AI look like?

Matt Turck [35:34] Performance review.

Mike Murchison [35:55] And how do you do that most effectively? And what's the performance review? That's right. And increasingly, that is what Ada enables for our customers. It's essentially a one-on-one with your customer service AI, where you are coaching it to get better over time. And the structure of that, again, is not that dissimilar from what we all know very well, which is a one-on-one that you might have with a member of your team.

Matt Turck [36:06] So what does that mean in terms of customer service teams? How do you restructure your existing customer service team in the age of AI?

Mike Murchison [36:42] So our customers have been doing this now for six years, and we've learned a lot in partnership with them around what's worked well and what hasn't. And what has worked best is what we've packaged in what we call the ACX framework, the Automated Customer Experience framework. Essentially, this framework takes your organization from being an agent-first organization to an AI-first organization in three simple steps. First, we enable you to automate as much low-hanging fruit as possible. We call it 30 in 30: a 30% resolution rate in 30 days.

Mike Murchison [37:03] So we get you rapid value very quickly. We then enter an integration phase where we integrate with all your business systems, enabling you to go beyond simple answers to taking actions, like we mentioned. And then the final phase is elevate, and that enables you to turn your organization into one that's far more efficient, but also one that's revenue-generating, and that enables you to turn your contact center into one that's cross-selling and upselling customers and reinvesting savings in novel ways.

Mike Murchison [37:42] Typically, what this looks like is the formation of an ACX team that grows into an organization over time. And that team is initially staffed, we most often see, with the top-performing members of your existing customer support team. So the top two or three folks who have the highest empathy, who are perhaps the most savvy with regard to your product or service, and whom you want to coach and offer a new career path to, they get promoted into roles called ACX managers or AI coaches, and they live inside your platform all day.

Mike Murchison [38:22] The second part that we typically see is a promotion of folks into new advocacy roles. And so the AI savings are banked and often reinvested into new teams that are spun up that are tasked with offering a higher level of service to your customer base. IKEA did this recently in the form of reinvesting their AI savings into, I believe, design consultants that they retrained from former customer service agents who are calling customers to help consult them on how to better put together, combine different IKEA products, as an example.

Mike Murchison [38:54] So those are the two major changes that we typically see: the formation of the ACX org, which grows over time, and the reinvestment into advocacy-related positions. Yeah.

Matt Turck [39:12] And what should they focus on at different levels of sophistication in their ramp-up of the AI? I'm thinking, going from content creation to more generative replies and actions. I know you have a scale here.

Mike Murchison [39:40] Yeah, I think, so in general, the goal is to get to 100% automated resolution. And in order to get to 100% automated resolution—and, by the way, if you can get there, which I believe we'll have customers who get there within the next two years—we have customers today who are north of 70%. The biggest hurdles to overcome to actually get there are, interestingly, primarily cultural. You would think it would be technological, but it's actually cultural.

Mike Murchison [40:16] And it's the organizations who are able to put the most weight behind the importance of their customer experience strategy, the formation of their ACX team, and give them really top-down enablement and spotlighting. We often see highlighting your ACX organization's impact regularly at your town hall, for example, as a best practice. Those are the ones that get ahead. Those are the ones that overcome most of the downstream blockers that typically get in the way of folks getting more value out of their AI.

Mike Murchison [40:57] And interestingly, one of the consequences of that is we typically see the ACX team become almost like this AI hub inside their business that informs the rest of the company around how to get the most value out of AI in other areas of the customer experience, and even internally to the employee experience. So it's a really fast-track way to, I think, coach the rest of your company on how to work with AI and show real business impact quickly. The second blocker after cultural is often around the API strategy.

Mike Murchison [41:28] I think a lot of companies overlook the importance of their, call it their API surface area, how accessible their customer data is and how accessible their business systems are to their customer experience team. That's typically an area that we see folks overlook. And I think what we try to encourage is certainly far more conversations between chief customer officers and CIOs. I think it's the companies where you see really good alignment between the CIO and the CCO where those are the folks who are sort of most ahead when it comes to the quality of their AI.

Mike Murchison [41:43] Great.

Matt Turck [41:48] Any word on generative replies, generative actions? What do those mean?

Mike Murchison [42:25] So generative replies refers to our generative capabilities that allow your bot to be generating answers in runtime automatically in a manner that's grounded in your knowledge that you've integrated. And so this is resulting in customers seeing essentially a 25- to 30-percentage-point improvement in resolution rate. So it's a really major set of capabilities that we're really excited about that have had a big impact on the customer experience. Generative actions refers to essentially the same on the API side.

Mike Murchison [43:08] So instead of managing your API-driven workflows prescriptively, you're essentially just configuring your one API integration once. And your AI is deciding, essentially generating a workflow in runtime and executing it in runtime. And that's resulting in a much higher percentage of conversations where you're taking action, and that's elevating the quality of experience as a result. So that's been probably even more exciting than the generative reply side of things. But I think they're just good examples of how our customers are increasingly using Ada to coach their AI to improve and not to manage the prescriptive nature of everything their AI does.

Mike Murchison [43:43] There's a growing level of trust that they create between themselves and their AI as it becomes more and more capable, much like we've all experienced as passengers inside semi-autonomous vehicles. More and more time off the steering wheel.

Matt Turck [44:07] Great. As we get towards the end of this conversation, I'd love to touch upon voice, which is an important product launch that you guys did recently. What does that do? And how is that different, or not, compared to the existing sort of digital AI chatbot product?

Mike Murchison [44:32] Yeah, we're very excited about voice. Essentially, Ada Voice allows our customers to extend their AI into the telephone. And that's a really big deal because, for many of our customers, the majority of their customer service is still taking place over the phone. And with Ada now, you can manage your AI in one place and deploy it to any channel, now including traditional voice. And not only is that independently exciting, given the abysmal experience that most of us have had with traditional IVR trees and dumb voice AI, but it's also unlocking new experiences that we call hybrid experiences, where your voice AI is able to send you a text message or follow up with you on WhatsApp.

Mike Murchison [45:27] And it's really unlocking true omnichannel experiences. We've all heard the word omnichannel and have pitched omnichannel experiences for a decade. But I think AI, when deployed properly in CX, is actually delivering real omnichannel now. It's just beginning to, and that's what Ada Voice represents to us and our customers. So it's increasingly going to be a first-class citizen for us. I think there's a chance that Ada Voice could be as big, if not bigger, in the next two years than Ada Messaging.

Mike Murchison [45:38] And we're pretty excited about it.

Matt Turck [46:01] Okay, very cool. Well, that feels like a very good place to leave it and wrap it up. Thank you for joining us today. And it's, for me, really fun to hear you talk about the whole story and where we are now, because it gives a sense of just the unbelievable breadth of capabilities and the history there. I said I was not going to gush as a proud investor at the beginning of this conversation, so I should probably stop here.

Matt Turck [46:19] But what a journey. Where can people find you online, learn more about your work, learn more about Ada, resources, and all those good things?

Mike Murchison [46:32] Yeah, I think best to follow me on LinkedIn. I'm writing a lot more on LinkedIn these days, so feel free to follow me there. I'll be publishing more of our learnings building at the frontier of AI and CX.

Matt Turck [46:42] Okay, wonderful. Thank you, Mike, for joining us today. Really appreciate it.

Mike Murchison [47:12] Thanks, Matt. Great to see you. Thanks for joining us for The MAD Podcast. We're back here every Wednesday with new conversations with leaders in the machine learning, AI, and data space. And if you like this show, you can also find the video recording of not only this episode, but many, many more over on the Data Driven NYC YouTube channel. Thanks again, and catch you next week. Thanks for listening.