The Truth Behind Automation Claims in Customer Support | Cresta CEO Ping Wu

0:00If you have a time machine to get people to see what's possible in 2001 and bring that back to 1995 or 1996, people will not be surprised what they see what they see. From my perspective, I do not see automation and the call center as separate thing, but rather one system. That's our belief that we build one AI platform that helping all the humans in the contact center, getting visibility, observability of everything that's happening in the contact center in terms of the data, in terms of workflows deployed on Cuesta.

0:56Welcome to Grit. I'm Juvin, partner at Kleiner Perkins, a show where we go beyond the highlight reel and explore the personal and professional challenges of building history-making companies. Today on the show, we have Ping Wu, CEO of Cresta. After nine years at Google building conversational AI, Ping is now bringing generative AI to the contact center so humans and agents can work together in real time. We talk about why customer service is so hard to automate and what the hybrid workforce looks like next. Enjoy the episode. What you guys are doing, I mean, you're kind of playing in many flanks right now.

1:27Like one flank is, you know, the core call center. And then, you know, you're moving into the customer support side. Like it's, you know, you go into like the Sierra, like you're really in the like eye of the AI hurricane right now. And the Sierra and the Decagons and you all. So you're kind of like playing in many flanks. One of the things that I was thinking about before this was like, you know, I don't think it's like going back to the engineering talent point. Yes. Like, do you feel how much do you think that whether it's engineering or any other talent at the company cares about what they are working on versus solving just a hard customer problem?

2:15I can be more specific about the question. Like, um, I have observed that a lot of, uh, folks today, especially in engineering need to go through this like metamorphosis of working first on the cool, sexy thing. You know what I mean? Like, uh, solving memory for LLMs, for example. Yeah. Uh, and then after a while realize, okay, like this is cool as a technical problem, but like it's not actually solving a real customer problem and then they kind of come back around to actually what i really want to do is solve a customer problem yeah and then apply interesting technology yeah in ways to solve that i wonder if you have a framing on that it's really i think folks that join us self-selected to be the ones that are really exciting about solving yeah hard technical problems that we have a lot, and as well as solving custom problems that can apply for real world, right?

3:17And that can drive ROIs and you can see the benefit. It's not just we have a hammer and looking for nails and that type of thing. And I think that, yeah, so once you have that group of people and then have that culture, so of course, I mean, people love to solve hard problems, but, you know, it's in the contacts of that can, you know, driving towards customer value and customer happiness. And in contact center as a whole, it's very easy to demonstrate because, you know, everything, almost everything's measured, efficiency and the call containment and the agent efficiency, attrition rate and customer satisfaction, the revenue per conversation, if it's a sales use case all these things are um you know contact center is being a to me it's that's why it's very interesting in the enterprise it's something that you know it's very measurable and a quantifiable in terms of impact yeah yeah you're um you're we're just talking about your uh folks that are on your board so i didn't know so sebastian reached out to me uh who's been on the show uh and um obviously like one of the godfathers of of ai and and basically started waymo so much yeah i think he started google x and why he yeah exactly um and then carl i'm uh for uh eschenbach now the ceo of workday yeah um i'm curious like what do you what do you ask them like when you like when you go into board meetings what are you digging from them like what are the types of insights that you want from them um i think depends on which board member um but i think through those two uh carl's no longer on the board he used to be on the board okay he left for work there as ceo doug liani for example um took his seat and france aquoya and for doug because he has seen so many different companies and especially around sas and you know he had been through so many cycles.

5:24So a lot of questions about company building and the failure patterns and the success patterns and for company building related questions. And Sebastian is allowed to get his advice on where he sees the trend from the technology perspective. And he's also very active in the startup world. So, you know, our conversation is mostly from the technology perspective and what does he see down the road. Yeah. But for Carl, he's, Carl has a very, very deep operation experience. So when he was on the board, a lot of discussion is around the golden market side and the numbers and how do you build a funnel and then how do you scale up through channel partners, stuff like that.

6:23The business is evolving into what I was saying earlier about competing on what is one of the core AI battlefields right now, which is the customer support use case, which is like if you are a company like United and your customer wants to change their flights with United, they're talking to a person that should probably be a AI agent. Yeah, United Airlines, for example, is one of our customers. And we have multi-years partnership with them. So we're very deep with them. So every 9 ,000 agents is deployed on Cuesta. So we have a system, AI system to help them on every aspect of ways through chat channels and voice channels to help them, for example, taking notes, automating workflows, and then also doing a lot of QA and insights and those products that deploy for United.

7:23Especially United have insight to action team that just spend their time, all their time, on Questa to servicing operation insights that you know why people call in, that you can potentially even eliminate a call by fixing the problems instead of having automation or argumentation. So from my perspective, I do not see automation and the call center as a separate thing, or AI agent and the traditional call center as a separate thing, but rather one system. That's our belief that we build one AI platform that helping humans and all the humans in the contact center and also getting visibility, observability of everything that's happening in the contact center in terms of the data, in terms of workflows.

8:10And then the same AI substrate is powering our AI agents to automate those conversations as well and for the type of conversation we want to automate. A lot of the marketing around a lot of these customer support companies talks about how they're automating whatever, 90 % of all support. but if you really dig in under the hood, it's like 90 % of the initial, earlier, easier use cases to solve. Like if the models do not advance anymore, like if these AI models were to remain static today, how much further do you think you could go? Yeah. So look, I think the way I think about it, so first of all, from the interaction perspective, right?

8:57And, you know, the way we think about the current interaction to contact centers across channels, chat, email, or call, right, it really falls into three different buckets. The first bucket are the conversation that shouldn't even be automated or augmented. They should not even exist. So it's a conversation that come in because you have failed processes or broken products or your bill statement is confusing. So the right way that AI to add value here is to really have 100 % observability into existing call center or contact center, only channel, and understand why people call in and then tie that, find the root cause.

9:35And if that's reflecting some operation deficiency, you should fix that. So that's what United Airlines is actually doing is that eliminate those calls from the root cause and then eliminate them altogether. And then the second bucket of the conversations are the one that neither party want to talk to each other. Like the customer do not really enjoy waiting in the line to talk to even a human. They just want to get it done, like password reset or tracking my shipment. If you can have an AI agent that can automate that interaction or even a website to automate that interaction, perfect. And the business doesn't really want to engage in the conversation either because it's low emotion value interactions.

10:16And then the third bucket, we think that, and by the way, for that, automation is the right solution. We should absolutely automate that. And for the third bucket are some of those in the real world, there are a lot of high emotion value conversations that people really call in in grief or in really stress out. Imagine calling an insurance company, your house got flooded, or I lost my blazer jacket on United Airlines closets on the plane. And in those situations, you probably want to have a human interaction that customer actually preferred that. And then either maybe due to the conversation nature is too complex today to automate, or just the customer, again, humans, want to get heard by another human.

11:02There are emotion value in there. So for those, we believe that the role AI should play is to really augmenting the expert humans and make them really do a good job and free up their hands so that they can be more emotional available to the customer. So that's how we think about holistically what the transformation should be. It's from the first principle of what the interaction should look like. either eliminate them because it's reflect broken process or product or automate them altogether because, you know, actually every party wants to, you know, fast resolution and then self-serve or they are tricky, you know, part of interactions.

11:41Either it's not either ready or, you know, by automation today or it is just high emotion value actually for business benefit of business to actually build a loyalty with a customer to provide the human expert service. And then so that's what we think about. Automation alone is not, you know, the right solution to transform CX, right? You see, you need different AI to play different roles in each different of those buckets. And then the other thing that you mentioned about what's the percentage, right? And a lot of people are actually very curious about that. Whenever they talk to me and they ask, how many humans do you think will still remain in contact center?

12:22Two years ago, there were people who said that within two years, there will be no one in the contact center. It should just AI take everything. And still today, there are companies that are saying we automate 90%. And the reality is it's uneven. Depends on three things, from my view. One is the complexity of the conversations or the complexity of the business. You and I start an e-commerce website on Shopify, and we buy toys and sell online. Very simple conversation. and very simple infrastructure, we can probably automate 100%, to be very frank. And then the other dimension is the complexity of your IT stack, right?

13:02And, you know, and sorry, so for the first one, the complexity of the conversation, on the other hand of the spectrum is really complex interactions, very complex business. If you're a business like some of the airlines touch hundreds of millions of people and you run routes globally and you also start new routes every year, all these combinations in the real world, there are so many ways this shit can happen. So the conversation can be very, very complicated. Some healthcare conversation lasts hours with very old demographics. So for those, I think even the model gets better, there are potential, it's not ready to get automated.

13:47it. The second dimension, sorry, is really about the IT infrastructure. So again, if our e-commerce site built on Shopify, it's everything is a modern API, you know, the shipment, the refund, the payments is all you can, you know, call. And these are the rails already ready for AI agent to roll. But on the other hand, if you, most of the Fortune 500 from as far as I understand in contact centers, agents, humans, have to deal with multiple systems, sometimes nine to ten different systems. A lot of those systems are homegrown, do not have APIs. So the rail is not there yet. So that's the other side of the spectrum.

14:30It's very complex IT infrastructure that's optimized for humans. Humans can still interact with all these different systems through graphic user interfaces, but it's very hard for AI to do the same thing. And then the third factor that determines the complexity of how much you can contain is the demographics of your customers, like the airline customer, for example, or healthcare insurance customers. So those customers, some just prefer humans. And if you're dealing with a young demographics, they're probably more open to digital channel interaction chatbot talking to chatbot versus a much older demographics they will potentially will really prefer in-person interactions so i would think that all these three dimensions determines um you know you know first of all you know how many you want to automate you can automate today and how many of your customer actually prefer what's the amount of conversation that should be automated hope that makes sense yeah it does make sense if i were to like double click on it, what do you want to do with the LLMs today that you can't?

15:43Like, where are you running into shortcomings that the models need to continue to advance across that lifecycle that you just described? Today, if you listen to the Android Capacity conversation about AI agents, and there are still a lot of areas that need to be improved around instruction following, especially complex instructions around multimodality that understand things that just not only text, right? And then around, you know, if there's no tools or APIs, can the model understand a screen and be able to kind of take action on top of that? And if that becomes possible, then, you know, the automation will become much, much easier, right?

16:29And then today, still, you have to do a lot of context engineering to bring the right information at particular interaction to feed into the model, right? And then that is really an art to determine what actually to bring from the knowledge base, assuming the knowledge base is there, by the way. So that's another thing. A lot of time for simpler businesses, the service procedure is already there. It's very clean and clear if it's simple. But for real complex businesses, a lot of those knowledge base, there's no clean knowledge base ready to be consumed by AI. They are usually scattered across multiple sources.

17:12And you need to first clean them up and then be able to kind of bring the right context to find the right piece of information and feeding to the model at the right time. And the voice is, of course, even harder. And then you introduce more ways for error to propagate. And then for humans that just fundamentally sometimes do not trust AI and want to just move on to other humans. So all these things that I think will get better over time with the model improvements. But, you know, there are still limitations on, for example, tribe knowledge, not in the knowledge base, the tools that, you know, is not available for AI to use in the environment that optimize for humans.

17:53So that's why our approach is really meeting the customer where they are. Automation is one product, not the only product, but rather the entire unified platform where the AI agents or agent assistant, they can also help humans when actually humans still need to take the call. And still majority of the contact center content happened today is still through humans. And how do you maximize productivity when the calls actually need to be handled by humans? And then another piece, a large piece of value in the contact center that people haven't really talked about is that if you run tens of thousands of human agent contact center, you probably have hundreds of people doing quality assurance.

18:36You know what those people do? They listen to the calls manually, and then they can only cover 1 % or 2 % of calls. Again, if you manage a lot of people, you need to know what people are really talking about. What are they doing? Are they following the playbook? look, are they following compliance requirements and all that. So there's quality assurance teams that usually do that. And then that do manually. It's not real time. It's low coverage. AI can automate way more than those folks. And then to really just reshape those workflows. So in terms of automation, so that's why it's multifaceted, that you can automate the conversation end-to-end.

19:14That's what AI agents do. And for the more complex ones or for the ones that with high emotion value, you can still bring a lot of value to the humans to make them. Maybe you can automate the first 20 % by handling authentication. All the after call works by automatically doing the data entry and then take away the summaries. And in the moments that bring knowledge answers directly to the human agents or, you know, automate the email draft or, you know, automate the workflows like data entry, that can also do a lot of savings. And then after a call, and there are so many other workflows that like QA that you can also automate.

19:54So the way we think about the full transformation of contact center is not just automation and at the AI agent level, but the entire, you know, contact center level. So you're saying you don't need any more advances in models? Oh, of course we need. Why? That can make a lot of the context engineering much, much easier, right? And then, like, you know, the other layers that you build around the models and to feed the right information. If you imagine you can have the entire telco knowledge base into the context And then the model can digest in real time and find a relevant piece in the ideal world and follow the instruction religiously.

20:38And if you can do that and then with the human interaction and then as the conversation gets longer and longer, that context also included. Right. Then, you know, the problem become much, much simpler. Right now, it's nowhere close to there. Yeah. Yeah, that makes sense. And a lot of the work that you all are doing is in big, big companies like the Uniteds of the world. My observation, spending a bunch of time with these types of CIOs in these companies, is like there is quite an insatiable demand to just like do AI. Yes. Which is abnormal for them to be on such a front foot for big companies.

21:21Yes. I wonder how long that lasts. like uh that music feels to me like it's going to stop at some point where they will focus on less projects and going deeper on those projects than letting a thousand flowers bloom across the entire organization right um yeah i wonder how you think about that yeah i think you know just a few years ago right there are a lot of different ai pilots and as you said they are pursuing many different directions. But I think as time goes by and people start to converge on use cases that will drive very concrete, undeniable RIs, coding is probably one of them, right? And then contact center, CX transformation is definitely the other one.

22:05It's just all the things I talk about, that there are automation that can bring very concrete value. There are AI that are helping humans and for sales for support can bring very concrete values. There are AI that, you know, automate the quality management at scale that can bring a lot of value. Those are concrete values. They're replacing existing spending and also increasing the span of control of your managers and replacing, you know, potentially your QA workforce. There are, you know, voice of customer like driving insights and to really do deep research in all your customer interactions that can give you the insights that you wasn't able to get before.

22:46So that's undeniable value. Some cases that may even replace survey spending because you can get everything your customer already telling you. I get it. I get it. I get it. I'm just saying, is the music going to stop? I think we care less about the music stop or not, but rather just focus on the thing that we think that will drive the long-term value. And then just like every technology before this, there are always bubbles that's, in my view, tied to human nature.

23:21And then, you know, but every time there are companies that emerge out of it, because people are usually right on the long term potential of those technology in aggregate. You know, you will probably drive a lot more GDP compared to today. and we just want hopefully we are one of those that can emerge out of it and just try to make sure that we are doing the things that have in the indeniable value and our customers love us and then we just keep our heads down building yeah the the music is definitely still playing in silicon valley around funding too like uh does like as you see you know like we're seeing some of our companies getting multiple rounds of funding done like in a year like insane i'm sure you're getting your door beat down by every vc like how do you weigh the trade-offs of okay i could capitalize the business more now take a good valuation like how do you do you have a framework work for that or a mental model?

24:27We just raised our serious D and$125 million this time last year. And so what if someone came to you with another$125 million? And we still have a lot of that in the bank, to be frank. So I think right now we're constantly evaluating opportunities, to be frank. But how do you evaluate them? Yeah, I think just multiple factors. One is how much value we can create with the existing funding equity value that we can create versus the cost of capital. So like dilution. Yeah, dilution at what valuation? Yeah. And also, I'm not sure driving high valuation is always the right thing to do. And I think you probably know a lot more than I do in those trade-off calculations.

25:15Yeah, I agree with you. I mean, I think in theory, it sounds good. But I think in practice, when you're an entrepreneur and is your last round public, the valuation? No, it wasn't public. I'll just make up a number. Let's just say your last round was$3 billion. Okay, I'm just making up numbers. And let's say someone comes to you now for a Series E, it's$6 billion. Logically, you'd say, okay, now I'm signing up for growth that is very, very difficult to achieve. it's going to take us a while to like fill into that valuation. Right. Right. Your salaries will naturally start to become more and more inflated because there's more money there and everybody kind of knows it.

26:01Right. And so you can say, all right, well, for this new engineer, we can pay an extra 50, 100K. For this AE, we can pay a little bit more because we have more money. Right. And so then your burn starts to go up. We're very careful about that. And I think one of our operating principles is lean. It used to be stay lean. We removed the stay. It's actually cross out stay. So stay got laid off. We don't need to stay. Just lean. So our philosophy has always been it doesn't matter how much money we have in the bank. But strategically, it may actually make a little difference if we want to accelerate things.

26:37Or whether we are cash flow positive or not, you always need to stay lean. It's not just from money burn perspective, but also I feel like it's just a good thing to have. Because once you're not lean, other trouble will come. Yeah. Okay, but like, it's a good example, right? So like, you want to be lean. Somebody comes in for a Series E, doubles your valuation. You take very little dilution to get. Yeah, it depends on how much you raise and whether the amount of you raise gives you enough cushion margin of safety to get to grow into the expectation. The more rounds you go and the closer you're to the public market, the more careful you need to be comparing with the public market, right?

27:27And you see where public market valuation is versus private market valuation. And frankly, private market valuation is set by preferred. And it's really a combination of debt and equity. And it's not pure public valuation. And that's one factor to consider. The other one is when you hire, if you're already pricing all the hard work and flawless execution for the next few years and bring that, mortgage that to today, is that fair for the people that are joining? And the smart people may think, oh, I may actually print my own stock versus joining your company. Because even if I execute perfectly over the next few years, that already pricing for today.

28:14So that's another consideration. Yeah, I think it's actually really well said. Like mortgaging your future towards perfect execution of what could be. Yeah, thank you. So because if you look at the public market and you can do the comparison and the data is already there and, you know, then the question is, can you at that AR with that growth probably higher? And how do you justify the valuation? And, you know, are you confident enough to grow into that? Yeah, but the inverse case of that is like, OK, at some point, capital won't be flowing this freely. Yes. and I will have to take another round.

28:54And timing that, it's an interesting dance, isn't it? Yeah, it is. So it's not just execution in terms of the product and execution in terms of the GTM, but also execution in terms of the capital raise. And that's why building great companies is very hard. Does it feel hard? You look at the numbers, how many companies getting to public, you know, 10 billion, 100 billion, and so on, right? It's very rare. And so that's why, you know, building an enduring company that can compound for many, many years is not easy at all, statistically. Yeah, but does it feel hard? For me, look, we're not anywhere close to those numbers I mentioned.

29:41But for me, I just feel pretty happy and um doing what i'm doing right now do you feel do you experience happiness on a day-to-day building the company like would you describe it as happiness oh yeah oh look if it's pain um there are sufferings i you know i mean probably every ceo will probably tell you and their their highs and their very lows and but like what's the alternative alternative i'm very happy and not taking challenges. And I don't have too many hobbies, and I don't think that will be relaxed and rest and anywhere. And I don't think I'll be happy or my mental state would not be good if I'm in those positions, especially with all these going.

30:34Knowing that we're in the middle of the paradigm shift, there's so much happening. I'm not in the arena. I don't think I would be happy. And I actually jokingly tell my team is that, look, I mean, I'm really happy to fly to anywhere to go to meet customers with you guys. And it's just like, you know, Golden Retriever, the ball is never the goal. It's never about a ball. I mean, if you lock the Golden Retriever in the room and with all the balls in the world, the Golden Retriever will get depressed. So I do feel very blessed to be in the middle of this paradigm shift and we have this role to play and I'm really enjoying it.

31:17Super interesting. When you joined the company, you led product and engineering. Then at some point, what happened? You become the CEO. Were you like, okay, yeah, I definitely want to do this. Did you want to stay, you grew up in the engineering org. Yeah. How did you weigh that? Was it a no-brainer? Were you like, yes? It's interesting because before I joined Cuesta, I spent 14 years at Google. And the last five years at Google, I was at Cloud AI, Cloud Artificial Intelligence team in the Google Cloud. They're building AI products for enterprise. So I started the Google Contact Center AI platform.