The split personality disorder plaguing many brands - Interview with Ping Wu of Cresta

0:00Ping Wu:So welcome to the next edition of the Punk CX podcast. With me today I have Ping Wu who is the CEO of Cresta. Now, Ping is one of the foremost experts in the field of customer experience AI. He co-founded and led Google's Contact Center AI and Vertex AI platforms. He was an engineering leader at Google and YouTube and joined Cresta in 2021 to build the future of AI-driven customer experiences. ping holds a phd in computer science from the university of california santa barbara and is here on the podcast with me today hey ping how are you doing and welcome great great

0:37Adrian Swinscoe:hi adrian great to meet you and thank you for having me here you're very welcome now i always

0:43Ping Wu:ask i don't sometimes i do a little bit of an intro to my guests but i always ask after i've done those intros if there's anything that i missed out from my intro that you'd like to add

0:54Adrian Swinscoe:Yeah, I think one thing is before I worked at Google Cloud and worked on contact center AI and Vertex AI, I actually spent eight years at Google mobile ads. I joined Google 2008 and spent 14 years there. So the first thing I did at Google was actually in mobile. So mobile ads at that time was also very early, just like AI in 2017. so yeah this was very fortunate to see that consumer paradigm shift towards mobile and was very fortunate to be part of that for seven years and that business grow over 1000x now nice yeah so but i think that ai opportunity and combined with cloud is is going to be even more than that

1:42Ping Wu:wow okay well we shall see shall we so in the run-up to setting up this kind of podcast we had a bit of a chat and you shared a number of things that i wanted to explore because i thought it was worth exploring and kind of i thought it'd be interesting for the audience and the first thing that you said is that you think that many brands are suffering from split personality disorder i thought wow that's quite a claim and i wanted to ask you to explain what you mean by that what are the implications for brands or organizations that want to deliver better outcomes for the customers?

2:21Adrian Swinscoe:Yeah, that's a great question. And the way we think about it, every one of us have experience interacting with brands and different businesses and the provide service to us. And then there are so many different touch points and each one of them feels like a different person almost. First of all, they rarely do they have the same contacts from previous interactions and then can anticipate the next interaction. And also at the same time, you know, think about during sales and custom service and marketing, these different stages of the custom journey, the personality feel very different, right? And we all have experience where during the sales cycle that just the business or the brand really want to speak to us.

3:06Adrian Swinscoe:And then once we actually sign up, become a customer, and it just, you know, completely turn into different personality almost. And it's very hard to get hold of them. And if you want to reach out to them, usually they will try to, you know, sometimes it's very hard to get to actual human to speak to. And it's not very easy sometimes even to find a phone number to call them in case you have an urgent service request. So from that perspective, it almost feels, I don't think it's anyone's fault, just the limitation of technology at a time that you don't, you know, suffer from the silos of the data.

3:43Adrian Swinscoe:and as well as just missing the context, right? And then on top of that, in each individual session, when you speak to a business, you are going through, you know, the context also get lost during different transfers. Either from now you have AI agents interfacing with human customers and I will speak to AI and then it got transferred into a human sometime. And, you know, the human will almost see me as a stranger. I have to start all over again. That is not a consistent experience, not as if that we are talking to the brand as if that's a relationship. We're talking about long-term relationship, but that does not feel that consistent experience.

4:27Adrian Swinscoe:So that's why we use the term split personality. But I do think that the AI power user journey and the future of CAX will solve that, will really turn that into a long-term relationship and really know you at the personal level.

4:43Ping Wu:You know, I think it's really interesting because I had that sort of experience just today, actually. I was having to call an insurance company to query something on a policy. And I was interacting with their self-service, sort of automated voice self-service system. And they asked me to tell me what your policy number is. So I told them what the policy number is. and then got to the end of that sort of stage and then i got passed on to a uh a human being to to to solve the kind of the problem and the first thing that they they asked me was what's your policy number and i was a bit like i've just told you that i mean that's been and it's been it's been it's not that i've told it to somebody else and they've forgotten to write it down or something but i've told it to a piece of technology that sucked that into its thing because they've confirmed it they've repeated it back to me they have that data in their system and it just hasn't been kind of like passed on so it it does feel like these little things as you like as you say it's almost you you feel like you're having one conversation with with a whole organization but you end up kind of it ends up being that you actually end up having multiple conversations and that's that just feels underwhelming let's say 100 yeah but i think the future is bright and i

6:05Adrian Swinscoe:I do think that with AI interacting with customers and also AI assisting the agents and our contacts will be maintained. And then that will really become like that relationship. You feel like consistent talking, speak to not a stranger anymore. Yeah. Just one business. I think that's the future that we're building towards.

6:24Ping Wu:And I think that's interesting because it leads me on to the next thing you said. because I mean, what you're talking about is how do we, how do we retain the conversational intelligence? If you like to kind of, to, to, so you can pass on context, you can use that kind of historically to understand kind of, is there any kind of patterns that can like show up? But previously you also said to me that you don't think that, well, we believe that CRM technology, which where it might traditionally or naturally can like sit in a historical context where this you know data would sit you didn't you think fundamentally it's not the right type of technology to develop that type of conversational conversational intelligence that needed to drive its better outcomes and i wanted to ask you this like going okay if why and what should we be using in its kind of place?

7:23Yeah.

7:24Adrian Swinscoe:I think it's useful to think about historical perspective, right? Why we have CRM. And a CRM is an application on top of a relational database that existed for multiple decades, right? Initially, it's used to coordinating between sales team and have a single source of truth, right? And then at that time, the relational database, you know, just get invented that that's a table, right? You have to store information there and then you have annotated with schemas. That's how a table look like. That's the data model that entire industry built around. And then the code itself, you know, cannot function without knowing the table, the schema, right?

8:06Adrian Swinscoe:And then that's the constraint at the time. Now, I think that's everything changes because of AI. And that doesn't mean that AI in the new paradigm, that model database or CRM should disappear. It will still work very well, and there are important information storing there. But what we are seeing is that that's not enough. Right. And first of all, the CRM data is very sparse. You only have some, again, structural information that you can capture, but that's not a full raw information that captured the entirety of the relationship. it's not really a full relationship, right? If I tell an airline that I always fly in November with my dad, my dad had Parkinson's, he need a wheelchair.

8:55Adrian Swinscoe:And that information, if the airline next time I call them or, you know, a year from now, they remember that, that will be a dry, dry, dry, putting moment experience for any customers, right? But CRM today, you cannot capture that. It's, you know, all this information is lost because, again, they're unstructured information as human conversation messy. It's not in today's relational databases because, you know, traditionally the code, the software cannot handle unstructured information like conversation very well. But now with the new paradigm, you can store all that historical context in its full raw form.

9:34Adrian Swinscoe:Or maybe you can compress a little bit with summarization and all that. And then you can leverage that with large language model power AI agent. Those agents are that new paradigm is what it brings is that it can make total sense of that unstructured information that previously the software cannot make sense of. So with that paradigm shift for the first time, now you can store all that and then be used for your future interactions. That is a very, very big thing. And I do think that instead of replacing the existing CRMs, the next generation of that customer relation context, what has its own storage, will capture all the historical interactions from how the business gets to know the user and all the way to the current interaction, almost like a Facebook feed.

10:26Adrian Swinscoe:It's like capturing all your interaction and then we'll make the future, all the interactions a lot more personalized. And I think that's a foundation piece that will lead to the future of really a custom relationship instead of the split personality disorder.

10:42Ping Wu:CRM will end up doing the job it's supposed to do, i.e. being able to help you manage relationships rather than just being a system of record. I mean, I think, do you not think that the next gen of CRMs might actually go, well, we can build our system to almost parse out the content, the sentiment, the intent, the context from this unstructured data, and then kind of populate it into almost add nuance. and flavor and you know and debt to kind of like to those kind of customer customer records but i guess on a an old historical architecture i kind of like it's a relational database that may actually be a bit hard from an architectural perspective right right so you know that's not replacing the

11:37Adrian Swinscoe:existing relational databases they're going to still have the information that's like the names or a number whatever the structure information but what i'm saying is on top of it now you can and store the historical unstructured interactions in its raw form, original form, like this by itself. And all the website interactions or the ads the users see and click and about how you get to the landing page, all that will be in the context. And then that will have its own different storage. And then that will be leveraged by AI agents or AI agents supporting humans or AI agent directly autonomously facing customer in real time.

12:14Adrian Swinscoe:and to make that interaction feel a continuation of the previous, all the previous conversations. I think that's the future we're building towards.

12:23Ping Wu:And you think that, I mean, people like a bit of a label. Oh, yeah. And does that mean you think that, you know, there's this whole system you're talking about, like a CDP type of system, which kind of captures all of that from all that sort of contextual data, the conversational data, the journey data from all the different places brings it all together? Or do you think are we talking about something completely different?

12:49Adrian Swinscoe:I think it's less about how we call them. But CDP, in my understanding, it's optimized for real-time access and those small data payload access to personalized ads or other interactions. And that's mostly still structuring nature. But what we're talking about is a long continuous window of all the interactions or even website, not necessary conversation interactions, but that capture its raw form. And then with the AI agent, they can, with the LM powered AI agent, they can process and make sense of all the past and then make the next interaction. Only selectively bring the bits of information that will make the current interaction personalized.

13:32Adrian Swinscoe:And forget about everything else. Just like real human. Real human with very long-term memory.

13:39Ping Wu:Yeah, no, no, absolutely. um okay that that you know that makes that that makes sense but it also kind of you know it it shows that there's uh for many organizations there's actually a lot of work to do to get ready for that type of environment i mean there's um i think my understanding is that many organizations sometimes fail in this regard because their data architecture or their cleanliness just isn't up to scratch.

14:06Adrian Swinscoe:Yeah. Surprisingly, I think counterintuitively, I do think that the type of data context I just mentioned is probably easier to build versus the previous data storage. Yeah. The reason being, you know, I think all these data interaction, you know, when it's first generated, it's messy. It's unstructured, right? You think about their original form is already unstructured. You do not need to do additional refinement. Yes. Right? So you can capture, as long as you capture it and store it, and then feed into your AI agent in real time, it can make sense. Of course, you may actually need to deal with some latency of AI agent, the limited context windows, or you have to maybe do some compaction and even do the summary, right?

14:49Adrian Swinscoe:But that is, to me, a simpler problem because you do not need to go up the ladder of structuredness. If you think about the data, its original form, it's almost like the energy, the level energy you need to take to take from its low, high entropy state to like lower entropy state where you have the structure form. It's tabular, it's very beautiful, but it's hard to maintain. So that process requires cognitive, requires computation, requires either human computation or machine computation. And it requires a lot of energy to maintain. So that's why it's hard. and it's brittle. But if you capture a raw form, it's not that complicated.

15:32Adrian Swinscoe:And if you use the AI agent or agent assistant or quest a product, we already capture that information for you and then we kind of store it away. So, yeah.

15:44Ping Wu:I mean, talking about that idea about using kind of agents and acting on all that sort of conversational intelligence, one of the final things that you talked to me about was that like many people you were saying that many people talk about taking an automation only approach to much of their customer support and you described it in a way that was funny because you said but you think that's a mistake and described taking an ai automation only approach like taking a shower with a raincoat on and i thought that was funny and i wanted to you know i wanted to get it on the kind of the podcast because i thought it was an interesting analogy but I wanted to help you understand what you mean by that and what should brands be thinking about instead.

16:26Oh, yeah.

16:26Adrian Swinscoe:So, you know, that's just one analogy we think about because we do think that, you know, contact center or custom experience transformation with AI, automation alone to not solve the whole problem, the entire problem, it's just one part of the overall solution, right? So for me, it's almost like you wash your face but with the wrinkle cover everything else. Right, I see. And you have a shiny face, but at the end of the day, it did not solve a lot of problems. Actually, it will cause a lot more problems if you just have a silo between the AI automation versus the rest of the humans. So I think for two reasons.

17:06Adrian Swinscoe:First is the reality is there's still a lot of humans. You know, you start with 100 % humans, right, you know, from five years ago. And then towards the future where you see more and more AI autonomously handle all these interactions and then also AI while helping humans. Right. And and, you know, the automation only approach is only automate part of the conversation. But, you know, it's a huge opportunity for AI to also assisting. It can also be AI agent that just work alongside the humans and take away the repetitive work and make them a lot more productive and be emotionally available to customers.

17:42Adrian Swinscoe:So for that, I think that's what a true solution should look like. It's not only automate, but also augment. And then even we have seen a lot of customers, what they told us is that we actually want to use AI to really understand the deeper reason why people call in the root cause so that we fix the problem. And then we fix the problem so it eliminates those unhappy interactions in the first place so that they do not even hit the automation layer. Right. So for me, the real transformation, like the real taking a shower, is have these three pillars. It's analyze and automate and also argument. That's at least the full power of AI.

18:25Adrian Swinscoe:And the second, we've been building AI solution since my contact center AI, CCAI days at Google. And one thing I learned is that in order to get to the best automation solution, you really want to tap into the human intelligence in the contact center. You know, I think the transition is gradual. You're going to improve the AI agent. AI agent is not going to be 100%, you know, on day one. No Fortune 500 company CIO will tell you that I'm going to unplug my call center tomorrow and it's not going to be all AI agent. It's going to be a gradual rollout and then it's a gradual improvement, right? It's very easy to launch something that maybe do some simple use cases like password reset or account status tracking or, you know, where's my shipment or that kind of queries.

19:12Adrian Swinscoe:But most of those, you know, things, you know, interactions are more complicated. And then there are a lot of long tail questions that, you know, humans actually handle today. And there's tons of tribe knowledge that are locked in humans' head that's not documented anywhere. So the path to get to the optimal potential of automation is like you cannot ignore the humans that already today that's there because, you know, there are so many intelligence that's been wasted almost. You know, every time, every day, thousands of humans, they do things. You need to watch them and to really understand what's going on.

19:49Adrian Swinscoe:And then to see where AI agents fail today and then get to the next percent better, you have to see where actually how humans get to resolution. and then you turn that feedback into AI agent. That's what I feel it's very much important. It's kind of missing is the continuous feedback group watching humans and then make AI better. So that's what I meant is automation alone is not the full solution, it's part of the solution. And then at Questa, we built this multi-pillared solution to transformation.

20:20Ping Wu:And I mean, that's interesting. I love the fact that you kind of said that you've actually got to understand the whole problem. And I think that's, there's a mistake that many people can make is they rush into just trying to either augment or automate without actually really understanding the, you know, and the problem. And so it's a bit like, we got a spanner. Now we're going to go find something to fix. Oh, yeah.

20:43Adrian Swinscoe:And because AI automation is not going to fix your problem. Like, you know, your CX, your call center is the symptoms of a lot of problems that you may actually have in your business. Either it can be a broken website, confusing policy, or UX problem on the application that caused people to get confused and then they need to call you. So if you can learn, get that insight out of those conversations, either it's AI-led or human-led, it doesn't matter. And then AI now is good enough that can really dive deeper into all these deeper issues and give you the insights and give you actionable insights that you can take to fix your business.

21:20Adrian Swinscoe:that's even better than automation that you just even have those interactions in the first place

21:25Ping Wu:custom happier i think the kind of the other thing i was going to say was that i really like that idea about you talked about tribe knowledge and i mean i wrote something i think last year and i it was alluding to the kind of the idea that i said that to be successful in this space there are two types of ai uh one is artificial intelligence and one is agent intelligence and you have to get both right to get that right. And because the scale of the tribe knowledge, I think is not something we necessarily can lay our hands on. But interestingly enough, there was some data I learned from another, I think, podcast some time ago.