#233 Matt Price: How Crescendo is Disrupting Customer Service with Gen AI

0:00Companies are terrified of opening up a new channel which overwhelms their support teams and and their support teams are constrained by budgets and the budgets are constrained by how many humans they can. So, you know, customer service leaders have been in this challenge whereby they've been trying to move pieces around and optimize to try and get the best type of service they can. But often this translates to trying to squeeze the most possible out of a bot that's not very effective or trying to negotiate the rate of a BPO agent down to as small as possible. And what you end up with this is this very disjointed assembly of parts.

0:41And, you know, unfortunately, being in a generative AI bot and just dropping into that isn't going to change the picture at all unless you take a look at holistic service design and make it simpler. Go ahead and introduce yourself, how you got to Crescendo, what Crescendo is, how it got started, and then I'll start feeding in with questions. I'm Matt Price. I'm the CEO of Crescendo. Crescendo is a customer service company that was assembled by some founders who were really excited about the potential that generative AI can bring to customer service. And what the company does is actually provides all of the technology and all of the people that automatically handle all the peaks and valleys in support demand across all channels anytime you want 24 by 7.

1:32And what we've found is this integration of the technology and the people is the only way that you can ensure that you can deliver the best possible quality for customer service. So it's better than people by themselves and it's better than AI by itself. Yeah, and generative AI chatbots have really taken over all sorts of functions, but customer service, it's moving into customer service pretty quickly. How good are the chatbots, are your chatbots based on API calls to one of the big proprietary models, or have you taken an open source model and fine-tuned it? And how generally do you think chatbots are going to perform in this market?

2:32The short answer is yes, they can be excellent. The longer answer is that there needs to be a lot of care and attention and expertise to make sure that they can be excellent and live up to the quality expectations that people have. So there's been a big gap on moving from experiments to real-life production environments. We use all of the great work that's been done by the companies such as OpenAI and the other models. It does off linguistically and the ability to actually assemble information and deliver it and converse with an end user or a customer are second to none. and so what we do is we stand on the shoulders of those giants and we've delivered a huge amount of IP that makes them work in real life and integrates them into the people part of the equation and really just takes away a huge amount of the risk and complexity that exists if you do it yourself.

3:55Yeah. Uh, and until now with, with, uh, the generative AI, uh, chat bots, which are both voice and, and text, right. Uh, you've said in the past that, that, uh, that customer service was more about deflection on engagement. Uh, can you talk about that, about, about what still exists. I mean, I was on a call yesterday, a very frustrating call with a, actually not a call, a chat with a chat bot, some rule-based chat bot that was not getting me where I wanted to go. You know, the challenge is, is the reason why most customers are faced with this, is twofold. Firstly, they haven't been able to get the generative AI into production, and there are a number of reasons for that most equality.

5:03Secondly, there's a misalignment of incentives. The old chatbots are provided by technology companies who are making a lot of promises on how much that chatbot can stop people reaching the humans. And then on the other end of this are the BPOs who are trying to get as much human interaction and time as possible because that's how they get paid. We've discovered by blending the two, you take away this misalignment of incentives and you can really optimize for the very best customer experience. And so you don't try and make the AI do more than it should do or it can do at that point in time. So we actually constrain it quite significantly to really just handle the things that it can.

6:04But what we do then is make sure that when it does hand off, it's handing off to somebody who is at a level beyond what the AI has actually been doing. And this integrated experience and being able to tune this integrated experience because we can design the service as such means that you don't end up repeating yourself. You don't fall through cracks. You don't end up going to multiple layers. And the results are really encouraging and surprising. going back to your question around then on deflection versus deployment we are not trying to deflect people from stopping them from talking to humans there what we're trying to do is have the ai do the best possible job it can and if it can't then hand it hand it to a human yeah yeah in practice what we then start to see is because of that quality and when people get get excited about this and by focusing on quality companies start to deploy it in many many different places within their organization they move it from being something that's hidden on the website because they don't really want to talk to people to putting it front and center in their product and because the ai can actually resolve many of the inquiries they can do that for the same budget that it took to, or less than it took to deliver the old type of experience.

7:39So we've seen dramatic increases in the level of engagement for a lot lower cost per interaction for businesses. Yeah, and this is a point I don't think, I mean, I think a lot of people have felt this intuitively, but don't necessarily understand that it's really, it has been baked into these customer service platforms. is one that BPOs historically have been paid by headcount, which means that the more people they throw out a problem, the more they get paid, which isn't really aligned with the customer. I mean, with their client, not the customer that's calling for customer service. And that a lot of organizations deliberately in the past do deflect away customer inquiries.

8:40I mean, I was trying to get in touch with a company yesterday, and there's a Twitter link and a LinkedIn link. Neither the Twitter nor the LinkedIn can you message to. and there's no email, no phone number. You're just stuck with what they have on the webpage. And then for companies that do employ a BPO, oftentimes those calls are maddening or unsuccessful. And there's also all of the phone menus and hold music and recording, all of that sort of deflects customers away from the company. And how intentional do you think that has been? It's been very intentional because companies are terrified of opening up a new channel, which overwhelms their support teams.

9:51and their support teams are constrained by budgets and the budgets are constrained by how many humans they can. So, you know, customer service leaders have been in this challenge whereby they've been trying to move pieces around and optimize to try and get the best type of service they can. But often this translates to trying to squeeze the most possible out of a bot that's not very effective or trying to negotiate the rate of a BPO agent down to as small as possible. And what you end up with is this very disjointed assembly of parts. And unfortunately, being in a generative AI bot and just dropping into that isn't going to change the picture at all, unless you take a look at holistic service design and make it simpler.

10:45It isn't going to change it. So, you know, I come from a software, custom service software background. My co-founders come from a contact center background. And by fusing the two organizations together and really trying to think of this as an end-to-end product,

11:09we've found so many different places where we can optimize the experience and basically rebuild the experience from the bottom up. So customer service leaders can focus on being service designers in partnership with us rather than being systems integrators or procurement agents or people like that, these different sources. Yeah, and as you were saying, that then once you have a system like that in place that instead of deflecting customers, engages customers. You want it out front. You don't want to hide it. And that's our desire. That really is our mission is that we want to, the first experience that we create is something that companies are incredibly proud of and can deploy further.

12:05And that's what's happening. you know it's it's very exciting to see and it's very exciting to see the potential business impact of when you can lean into engaging with customers and have more conversations with customers what you uncover of as all of these hidden things that the customer does want to talk to you about but hasn't been able to. And there's been a lot of studies, but more customer engagement leads to better attention and leads to better sales. It's a classic and very simple model. Yeah, yeah. I'm sure there are studies, but it's also sort of common sense, right? The more you engage.

12:55And so in your solution, You have generative AI-driven chatbots, either voice or text. And then you have human agents under the same umbrella. How much training is there with the human agents to work with the AI? because I can see with a bolt-on solution that, you know, a company buys a chatbot and puts it on their website, and then you run into a problem and the chatbot has to hand off to a BPO that's not in the same organization. I can see that, yeah, there are problems with that. But to get around those problems, how much training do you give the human agents with the AI? I mean, are they familiar with AI, how it works?

14:01Do they see the conversation or hear the conversation when it's handed off? I'm curious about that. Yeah. Your first point, by the way, is really interesting. is that, and again, this is on service design, is that constantly recalibrating what the AI and skills of the AI are, and then handing that off to a human that has skills that are at the next level is something that's challenging in itself and something obviously with the fact that we have a very extensive human capital and the AI that we can merge, we're constantly redesigning that service and understanding that calibration. So that's really important.

14:52Then when you start to look at the AI to human interaction, there are some very basic things, like you suggest, is that making sure that when there is a transfer to the human agent, they understand and they can view the context. They have tools that we use using the same AI that can do language comprehension and response. So that's the core level. But what we found in some of our early product design and work, and when I say product design, it's not technology, the technology plus the human interaction, some very interesting things happen. So the team leads become accountable for the quality of the AI and the humans in interaction.

15:44So it's very much the incentivized closed loop of making sure that the AI quality is there and the human quality is there and constantly improve. But you can take it to the next level then and say, okay, what if the whole team was accountable for the quality of the service? Then you suddenly have all of the service agents being your quality control monitors and updaters as part of that loop. And you start to get this flywheel going that dramatically raises the quality of service and constantly innovates those interactions. And because we provide that service, those iterations happen very, very quickly.

16:25Whereas actually, if you're working with different segments, like a bot company who is then integrating to a CRM system, which is then going to BPO, those cycles can be weeks as you change processes. And so, which doesn't work in the rate of innovation that's happening with AI technology, right? Right. And you have not only use AI, not only in the conversational interface, you have AI analyzing the conversations on the back end, et cetera. Yeah, this is a reflection really of how the whole service stack, both from technology and treatment, gets redesigned and needs to be continually iterated. So the first thing that we did was we understood the idea that to get generative AI into production, it has to be super high quality.

17:25So we actually built our own technology to analyze and review every interaction, score it across a number of criteria, and then adjust very, very quickly and trap any errors and really score that quality. that that started from really just the generic system but then that gets used um for every customer that we bring live is that we're iterating and testing them such that the quality is is incredibly high what what became apparent then was that the same system was also good at assessing the human elements and the end-to-end so so our customers can get full visibility end-to-end of the quality of service, whether it's being delivered by the AI or by the human.

18:17It happens on every interaction. And we obviously build best practices around that. So it reinvents how quality assurance is done within organizations. But our customers don't really need to know that. What they do like is the dashboard that they get, which allows full transparency and scoring of every service interaction. and they do like the fact that because we only charge by outcomes, we will commit to if something dips below the certain quality of service that they would expect or we would expect to give them, which, by the way, is very high, then we'll credit them back for that interaction.

18:57So a simple piece of AI technology provides so many different opportunities for service improvement and bringing together the economic incentives between us and Acoly. Yeah, so that's interesting. You charge by outcome as opposed to by headcount, and that just makes a lot more sense. I mean, I was really surprised to hear that BPO is charged by headcount because what's the incentive for the VPO to be efficient? But in this case, by outcome, that aligns really well with your customer. And how has this affected the jobs in the call center? because historically those jobs were grueling and there was very high turnover because people don't want to stick around.

20:09It seems like this would make the job more interesting for the call center employee. When we first designed the service, We went and spent a week in a contact center, the whole team in the Philippines, and we sat next to agents. We looked at every role. And it's true the way that custom service contracts or outsourcing contracts has been structured have been in such a way that the drive towards metrics leads to behaviors and job roles which are not really very fulfilling for humans. So what you see is people copying and pasting chat answers from notepads into chats.

21:12And really, just the service agents really just performing very manual tasks. What's interesting then is if you observe agents where escalations have come through and TANs, or if they move on to certain things, Once you unlock the ability for them to actually use their depth of knowledge or their empathy or deeper skills, how much more fulfilling that those roles are.

21:46It's not unusual for a traditional contact center to have 50 % attrition of customer service agents every year, which is huge compared to most industries. very expensive for the businesses and that cost obviously gets passed on to the client overall. What we're finding is that with our approach whereby the AI is handling a lot of the repetitive work and in many cases, generative AI is very good at it. It can take context, earlier in conversation or historical context. It rephrases every time for the customer and it's very adaptable. And then hands off to a service agent who has the tooling and has more meaningful work.

22:36The retention goes up, the satisfaction goes up as well. So again, a lot of this is just rebalancing and continually rebalancing between the capabilities of the AI and human and making sure that the experience is integrated very well, not just from the customer, but for the human agents as well yeah uh but that's an important point because there's such a drumbeat uh around ai but it's that it's replacing or that it will replace humans but in this case uh there's a natural attrition in call centers because people don't stick around but once you implement implement this kind of a solution that's right uh people do stick around so uh So it's kind of the opposite of what one would or the media predicts.

23:33That's right. I mean, we're one of the fastest growing companies that delivers AI technology, and we expect to be hiring more people in the next 12 months. Yeah. Yeah. The training that goes into setting this up for a particular customer, let's say it's a bank, certainly the human agents are trained on whatever part of the banking service the customer service is focused on. But how do you train the chatbot? Are you using sort of traditional, I mean traditional, it's only been around for a few years, but, you know, retrieval augmented generation where you have a vector database, you load up with the customer's data, and then the chatbot answers from that?

24:44or are you just fine-tuning a model to talk about a particular industry? How do you handle that? Yeah, no, we take a customer's set of knowledge and we will use that in an ROG approach to then help provide the answers. And what we focus very carefully about, though, is making sure that knowledge is accurate and it's contained and the AI is trained in such a way that it will only use what you give it. And again, here are some of the challenges of making this work, because in order to do that, there's a lot of tooling. I think there's human expertise on how to craft that information. And a lot of the early attempts are, well, okay, we've got these fantastic LLMs.

25:43We're just going to train them on our knowledge bases, our whole knowledge base that exists.

25:50Almost certainly, in fact, in our case, every knowledge base contains inconsistencies and it contains hidden information, it contains old information. And so the curation of the information that goes into it is very important. So what we tend to do is very much contain the set of knowledge to begin with, and then we'll iterate and we'll grow that. And that gives us far and away the best quality. And the reason we can do that is, obviously, we don't mind about handing off to one of our humans to answer the question if the AI can't again, you know, the alignment. But there's a lot of work to do that now.

26:30And this is part of the transformation and why a lot of experiments within companies on generative AI for customer service get stalled. And we tend to pick them up. What you also have to do on that is then you have to have AI experts who are training, getting the right models. You have to redo how you do knowledge expertise and knowledge management. and change how you do workforce management within an organization. So there are so many jobs that need to be upgraded and changed in order to actually make this work. Obviously, that's part of what we can bake into the service, and it means people can go live very quickly.

27:16Yeah. Yeah, and I was looking at one of your press releases before the call, and you've grown very quickly. There was a recent acquisition.

27:34Would you call a BPO or a company with a lot of call centers that then you brought into the tent and you bring along with those call centers all the customers that they've been servicing. So have you seen an increase in satisfaction among the customers that you brought in through that acquisition? yeah the the interesting thing about doing i mean maybe to provide a bit of context on this so you know crescendo startups technology plus people company we we were growing fast and we were hiring um our own service agents and deploying and got to a point whereby from the delivery of service that we were looking at ways to accelerate that um coming from the technology side myself working with our partners on there, it blew my mind how well operated this company was, how optimized and how good it was at managing people and building models around it, to the extent that it would be very, very hard to replicate in-house for organizations.

28:54And then also how things could be organized to deliver. a lower cost service or a flexible cost of service. So the idea is you can put a bot on 24 by 7 and it can speak Chinese to somebody in the middle of the night. But what happens when you do a live transfer to a human? Typically, in order to stop a live transfer for a particular language, you need to employ 24 people just for that one dimension. The advantage of using an outsource or a BPO is that there is this pooling of resources or this ability to scale that that makes that scale scale happen. I was incredibly excited about, again, this next level of being able to provide a differentiated, seamless service and just wowed by the level of expertise that we can deliver that probably most mid-market customer service organizations wouldn't be able to do.

29:57and so going back to your question the um the customers that we're engaging with who are really just um had had a labor-only relationship with this business um the company's called partner hero that one of the in the top um and the reason we really wanted to merge with them is that they had very much focused on this high-end quality end of the market which would fit very well with our solution um i think what's interesting that we're seeing is

30:33the ability for us now to handle their requests in in a very different way so now we're coming up to holiday season now and many customers are asking many companies are asking oh can you add some more agents please add some more labor and we'll see it you know how about we talk about this in the capacity that you need and then how about we address some of we address that by committing that we'll handle your capacity but we'll do that with a combination of technology and people um oh and by the way at the same time um your cost per interaction will go down and at the same time by the way we're ready to switch on these new channels when you are in these new languages and shifts that.

31:23Immediately they get it. These are the service improvements that I've been trying to achieve as a customer service leader for a very long time. And now I have a bender or somebody I have a trusted relationship with that can help me make that happen. And I think it's obviously very exciting for us. And it's exciting for us to see how all of the people who are working, the operations managers who are working with these customers, eyes light up because they are now empowered to solve these problems for their customers too. Yeah. You mentioned Chinese. How many languages can you handle? uh 56 wow uh so the bot is uh your your ai is uh trained to to respond in 56 languages do you have ai human agents who can speak all those languages um there might be one or two um more uh um obscure ones that we uh that we're not servicing at the moment but i would say we're servicing most of those languages.

32:42And if there's demand for additional ones, then yes, we can do that. Which again, goes back to just the scalability and quality of an incredibly well-designed and well-run people operation. It really opens the doors to a lot of opportunity. Yeah. So the market, it seems to me, is virtually endless because not only do you have all of the bad customer services still using phone menus and understaffed call centers so you're on hold. And not only is that your market, but there's a lot of people that have never really done customer service who, if it's outcome-based pricing, presumably that's cheaper than the headcount-based pricing, now have this available to them.

33:46So how big do you think the market is? the um i mean the outsourced bpo market alone is is half a trillion uh well dollars the um i think there are a couple of other vectors there that we think that using this approach more companies will move from in-house to external um there because they won't think about it as offshoring they'll think about it as just using bpo or business process outsourcing using those exact words they are they are finding an expert partner who can actually deliver uh something that would be is harder and more expensive to deliver in-house i mean bpo has become synonymous actually with please find me some cheap off source labor you know and which is absolutely not the case so we think the market will grow through that and then the market will grow because the results that are being seen from increasing the number of interactions now we see companies who are turning on voice for the first time because it's an incredibly powerful way you know to interact the number of interactions um service interactions will grow exponentially as well.

35:10There's the current market, there's the expansion of the market by providing better surface, and then the growth of the number of interactions.

35:20It's very large.