The ActivateCX Podcast
Join Frank Rogers on The ActivateCX™ Podcast, your resource for demystifying, clarifying, and providing guidance around AI, CXM, and the modern Cloud Contact Center.
In this Podcast series, Frank interviews Thought Leaders, Unpacks critical AI & CX technology, and addresses the leading Experience topics of the day.
#cx #customerexperience #ai #ex #cxm #contactcenter #salesstrategy
The ActivateCX Podcast
Use these 3 Tech Tools to make Money with AI
Learn more the 3 AI Solutions to improve business Profitability and Sustainability:
https://activateCX.arroyo360.com/ai
----
Want to improve customer experience, learn how AI contact center solutions can help? In this video, we interview Kolin Koehl (CPO at ObserveAI)! Together Frank and Kolin dive into the world of the 3 main AI tools for improving customer experience as a means to boosting your sales and customer support performance.
Kolin also addresses common concerns about bringing AI into your CX plan. At the end of the day, learn how to keep your customers and become sustainable against competition with this compelling value proposition?
Contact Arroyo360 to get clarity on approaching AI in a way that makes sense for you. Don't forget to subscribe for more tech updates!
CHAPTERS
00:00 Intro
00:10 A Day in the Life of an AI Product Leader
05:29 What is the Winning AI for CX Product Set
07:18 Getting to a Shared AI Understanding
10:19 How far do you go with AI in CX?
11:52 Knowing that your AI is working!
13:04 How to ensure the AI behaves nicely!
14:34 Agentic AI, the new Kid on the block!
16:08 Adopting, Adapting, and Getting in the Groove
17:50 In Security we Trust!!!
20:01 Is your C-Suite on Board with AI?
22:18 Thanks Kolin!!
Hey, Colin, welcome to the show. Thanks for having me, Frank. Excited to be here. Colin, you're a product leader for an AI company and one especially that is hyper focused on CX and just right off the bat, I'm just curious, how do you split your time between the governance and the oversight of just production on a daily basis and what you're thinking in terms of the future, that R& D mindset? Yeah, good question. I've been really blessed at Observe to have I think one of the best performing AI and engineering teams I've ever been a part of in my career. So it makes this decision of how do I split and manage my time a bit easier. But I'd say, as, at the VP level, I kind of spend my, my time in thirds. So about a third of my time is spent really tactically with that team, the product team that's leading these AI products with the engineers, about a third of my time is spent, cross functionally of making sure go to market arm is aligned and just with the rest of the org. And then I try to spend as much time with customers as possible too. So about a third of that time is outwardly facing. And, in the meetings, helping with even prospects that I've never heard of observed before sitting in on those calls and all the way through implementations, right. I try to sit on and be a reference point for those customers. So they know there's a product touch point there. And it helps me learn as the tip of the spear, like what are our current customers. What are their problems and pain points and how are they evolving? And same with the prospect world. So I can build better roadmaps. That's amazing. I like the idea of ultimately that the customer being one of these aspects of your week and your day that is influencing the product path. Do you see a lot of that being formulated based upon those conversations that you're seeing with clients? Yeah, I mean, one of our core tenets that observe is customer first, right? So it's really easy to just say, I'm going to be on this customer call and really sit and listen to their problems in their contact center. And our customers drive our roadmap, they drive our innovation. And in a lot of ways, we're pulled into the market everywhere we go, because we follow our customers and what they're looking for and the problems they have in the contact center. So let me ask you a quick question just as a follow up there is that when you're talking about the contact center in particular, how does, how do you think that AI is doing currently right now in terms of replicating the nuances of? Fundamentally, just human empathy, intuition, adaptation to the conversation. Do you see that there's a widening gap, between businesses and their customers, or do you think that's actually helping to promote closing the gap? I think that if you look at how AI has performed, let's say for the past year, the gap is definitely closing and it's improving, but the goal was never to, replace that human touch that white glove touch your experience are going to have when you come when you call a contact center or you need help for something where you're trying to, be sold something, right? I think the idea is that a eyes are really fantastic and handling a lot of these transactional parts of the conversation. A lot of these transactional use cases. Where a customer is going to call in because they need something, they need it quickly and they just want the answer and they want to go on their way. And it's less about, again, replacing that human empathy or replacing that human touch, but more about, having to make sure that the humans are spending time on the interactions that are most problematic and the ones that are most impactful to your org. And I think the most exciting thing for me as an AI leader in the space too, is this is the worst it will ever be. Like the AI, the models, right. So it's only going to get better from here. Yeah, that's great. I mean, I think that there's this, sometimes this dystopian view of the AI versus humans. And, and do you see it a little bit differently? Do you see this as being something where this hybrid workforce of digital agents and human agents, and when I say agent, I just want to clarify because that's used an awful lot. And I think it has a nomenclature that. that plays into a contact center view like the customer support. But I also think just really sales and support people, just all the customer facing people. Do you see the AI and those individuals really more or less kind of working together in the future? Yeah, I definitely do. I mean, I think there's a nice, bridge where the three layers can play together and the three layers being first, there's AI agents that can actually automate some some level of interactions and pass along a lot of context to human agents when they need to escalate conversations to human customer support team. And when you do take a call,, let's say it's a very complex call, like a fraud call that comes up and gets escalated in. Right. How do you provide guidance as in co pilot that customer support person through that interaction, because it might be a very complex process, the SOP might be 10 pages and it might be buried somewhere in the back of, the company SharePoint. So how do you pull that front and center, how do you help them just follow here are the steps, make sure they cover everything from a compliance and governance standpoint. So that they're giving the best service they can in a really just a backseat co pilot type type role. And then lastly, it's after the interaction ends, right? How do you then go and measure that experience and understand not only did the customer support rep or the agent perform well, but how was the customer's experience? Did they have a good experience with the brand? How likely are they to return? And how likely are they to call again because of the same issue? So those are the things we try to look at holistically from that perspective. Yeah, that's great. The industry has morphed really over the last three years. And when we talk about, CX and AI coming together, I see three primary products that sit out there. There's the conversational AI, which is your voice bots and chat bots. Then there's agent assist, which is really meant as an internal function to support that customer facing person. And then really this conversational intelligence that gets you that deep analytics around all the engagement. Fundamentally, it's just this triumphant of technologies that are advancing. As, a product leader, do you see anything on the forefront coming out that you see a fourth product or a fifth product that is starting to emerge as as a complement to these technologies? That's a really good question. At the moment I think what we're going to start seeing is agentic AI and AI agents really take More of a role in each of those three areas that you described before you just described there is the traditional chatbot voicebot world right of if this keyword is said you say this phrase and if this is said or this dtmf tone is picked up say this and do this or call this function ai agents can Really provide a natural conversational way to handle those interactions. Now, any agents are now starting to really own interactions, and I think you'll see the same technology deployed and the assist products and on the back end for measurement afterwards and analytics. products. It's yeah. How do you have the same similar concept? The agent, AI agents, helping that agent navigate the conversation. I think AI agents are going to be the forefront of contact center here for the next few years because of that. So we'll see really deep improvements across those three product areas with AI agents being the backbone. I couldn't agree more. I see. Across all of our clients, that there's a broader acceptance to the AI agent being a strong component of the stack. And, but yet there's still, a mystery around how the product works. And we spent a lot of our time on education . I can understand that because AI is somewhat of a. pervasive term out there, but still people don't know how the knee bone is is fundamentally connected to the shin bone with it. So how do you work with customers? What do you think is the best way for an AI company to be transparent around how their tech works? How the algorithms function , where there are rough edges, where there are things that, are damped down in terms of exposure. How do you have that conversation? Yeah, education is a huge part of this journey right now. Two years ago, ChatGPT was basically born, right? And this whole, wave kind of kicked off, and, even a year and a half into that, you're still explaining to some people,, they're learning about this for the first time, and how to use these tools. And then, you're going to say, hey, use these tools in the contact center, right? So the customer education part is a huge journey, and it's a huge part, part of our, initiative now, is how do we educate contact center leaders so they understand. AI, generative AI, agentic AI, and how it can impact their their support center and their contact center. But in terms of how we like to view it is with full transparency and with full control. So all of the, AI agent products or gen AI products, even traditional AI products we ship. We try to ship as many levers of control into the UI for the customer to navigate and manage. A good example of this is, historically we have a, we have a feature called moments, right? So did something happen on a call and you might put in keywords to detect those moments? Did you say the disclosure at the beginning of the call? Right? Check these keys. These keywords were said that can go in reporting or be used in a QA form. And then you start to use gen AI for those same type of moments. And now you can handle really advanced moments. Did the agent show empathy on the call? Empathy is not easy to capture with keywords, but it's easy to capture in a description and you can describe what empathy means for your brand. And we give the user all of those controls where they're putting in their entire description, even SOPs to support what their description of that is. So then when they go to reporting, they can see it, they can calibrate it, and they can tune it. And we try to give them all those capabilities so it's truly Owned by them and controlled by them. And if anything, they're leveraging us just for support to configure it if needed, when they get there. Yeah, I think that's great. I like the way you explain that because there is just so much of a high level of tuning and configuration that goes on when you deploy an AI. So it's not something that is akin to having, chat GBT and you're just dropping in a prompt and letting it do its thing. You really can damp down the parameters. Let's talk a little bit, about conversational AI, because I think that's the one where we're trying to drive containment and deflection away from a human being, and we want to resolve the issue or support potentially a sale in that particular engagement, what do you think is, the sweet spot for companies in terms of that balance, because there's some concerns out there that we could become over reliant and that would drive a particular risk to the organization. I think one way to view that is when you configure a conversational AI agent, at least in Observe, right, is everything is configurable via natural language prompts, but guardrails are a huge part of that journey too. So how do you configure and manage your guardrails of areas of the conversation you do not want the agent to engage in? So if you're doing a sale, let's say it's a health insurance sale or a car insurance sale and you definitely don't want to be having the, the agent engage about accident history or medical history, providing that in the guardrail so it's not brought up during that sales process is, is, is key. And if it is brought up to escalate it to a human agent to finish that, that process or that that intake form. So we try to leverage and highlight how those can be used to help manage the conversation. And the other thing I would say is when you set the goals for the agent is making them simple and measurable to so the agent doesn't over index on them and try to perform too hard too hard there. So if you have those two things in balance. Very, transparent, simple goals and really good guardrails. You end up with an AI agent that kind of does what you need. And then with the measurement after the fact, you can measure exactly how that call performed. We give you a report on what happened and what you could do to improve it. So you can tune it as well if you think it's, overselling or being way too passive and it's transferring too many calls. Those things are tunable there. In any given day when you're running, your customer facing operations, how do you get a sense that the AI , is working, that it's doing its job the right way, and that the experience to the customer is being felt in a way that you really want it to be? You shouldn't get that from from looking at the reports every day that are generated. But when you first create the agent to one of the nice benefits of observe is we have all the previous interactions as well. So when you make an agent, you can simulate it on. Hey, how did this? How would this agent perform across all my other calls about car insurance intake, right? You can actually simulate it and say, Okay, it handled these 80 percent or 90 percent of the way containment. All right. So that's how you actually go to deploy it into your queue or into your contact center. You have a good understanding of how performance is going to be. And then when you do deploy it, you have a quick measurement loop with the observe post analytics platform to measure that. And you can measure it every day and you can trigger alerts in different workflows. You understand how it's performing. If there's something that's happening or the calls have changed, you can then make those modifications accordingly. So, in those conversations, ultimately there is an outcome that we're working towards. And I think that from a consumer perspective, there's perhaps a view of does the AI work to drive or manipulate the consumer behavior. But I think as you kind of unpack a little bit in the configuration, would you say that that's pretty much dependent upon that organization, maybe what their intent is and that it's not really an AI that's driving manipulation? It would be the company and how they, configure the application, to drive that engagement. Yeah, that's fair. And true to say, it's also true that we have guardrails baked into the system level to prevent things trying to manipulate an end customer to do something they're not working to do. Right. So there are system level guardrails we put in place that when our companies or our customers configure their own agents, guardrails. So there's a baked in safety layer that happens there. And I think that, helps prevent and manage a lot of that too. Another piece is just standard best practices that we try to provide, disclosure, right? You want to disclose. It's an AI agent contacting you or a virtual assistant contacting you. So you're not trying to necessarily trick a customer into thinking it's a human and then three minutes into the call realizing it's not, that's just not a good experience for anyone. I think it's okay to be upfront with those things. Yeah, exactly. We, I think we've all been on a chat bot before where we say, are you a human? And it tries to kind of. Bogart its way through that without being, you know, fully clear When we think of agentic AI coming into the picture, how do you think that that works in terms of driving a very adaptable experience to the AI working with the consumer? That's my favorite part of the agentic AI though is how adaptable it is to the conversation. I mean, before in the deterministic, flow chart world, it was really hard if the call went off the rails to like, you bring it back. Right. You were kind of like, no, no, I can do billing requests or I can do FAQ. And, if you want something else, no. And if you've already went down this tree, you can never go back to the top. And now it's just, it's flat. It's conversational. You can, halfway through the conversation, say, actually, I want to check my bill balance and it'll go use that tool and solve the problem for you. So I've been really impressed by the fact that instead of spending, six months building an intent library for every bot you're going to deploy, you build an agent in a week, because you're describing the goal and the prompt and the LLNs are so powerful now they can just, they can have the conversation and they can have it really naturally. And that was kind of the light bulb moment that went off for us a year ago was doing one of these internal demos. And just, it was almost funny actually hearing it have a conversation, place the order. everyone wanted to call it and test it. Cause it was like, this can't be real. There's nobody set up, nobody set up a list of keywords for this, right? And that's what gets me the most excited is that what you described there is how we just handle those conversations naturally. Yeah, that is absolutely amazing. I agree with that from from a challenge perspective, an organization adopting AI. What do you think the biggest challenges are in terms of going through that experience? And how do you think that you fundamentally overcome that? Because once you enter into the fray with AI, especially in the customer facing side of the business, You have started not only a new customer journey, but you've started a new journey for yourself as well. That is not going to end how do you see people, overcoming challenges to getting into that flow, I just encourage them to just, adapt AI in the contact center to move slowly, it's okay. Pick and choose exactly the problem areas that you want to bite off. Cause every contact center is plagued with, with problems that they have, right. Whether it's too high HT, whether they think that, too many, too many calls are escaping their current call deflection pages, whatever the challenge is, really think about prioritizing what those problems are, try to assign a dollar value to them, and then you can go work with someone , like observe to really figure out, , how are we going to impact these problems for you and let's measure them along the way to show success and we can work down those lists with you and we can even bring new ideas that you probably didn't think of before because we have a huge library of 350 plus customers that we've seen. AI impact their contact centers. And we can give you those playbooks on how they can impact yours as well. No, that's great. With our clients, we really look at the lowest common denominator being the user story. It almost has a big arrow over the top of it that says, start here. And it allows you to drive out a form of experimentation, right? Because the company's learning you're building out a new experience and it's definitely the starting point that's going to evolve and it allows you to kind of minimize risk as well, because you're not trying to make it so incredibly pervasive. One of the elements of AI that I think is really a big subject is anything that has to do with security, data, privacy, all of those types of things. How do you think companies should be addressing that as AI becomes a larger component of their forward facing engagement with customers? I think those are, quarterstones of anybody is offering that they should be right in the market today. If they're not, then there's a fundamental problem with the partner you're choosing. If security and trust and, enterprise graded aren't the things they lead with, then, consider having other conversations and when it comes to AI, what I mean by that is , are they when you when you pick a partner and you're trying to build an AI solution like to us for assisting an agent on a call or even automating an interaction or measuring the interaction, what do they do with the data? Where does the data go? Right? Does it go into the platform? And is it staying there? Is it a single point in and out? Is is your data going to open AI? Is your and then and then coming back later. Are you comfortable with that? Is your security team comfortable for that? So the thing you really want to push for here is transparency and understanding the full Information architecture and finding out where the data is is going where it's residing and the thing with observe that we like to claim is is for the post call interaction side. We do use our own llms to handle All those interactions and enrichments. So the data, when it goes into the observe ecosystem, it stays there. And when you use something that's more complex, AI product, right, where it does need higher grade LLMs or higher performance, LLMs. You choose which LLM your security team is comfortable with. Do you want to use OpenAI? Great! We have integrations with them. Do you prefer Google Gemini? Do you want to run it on Azure? Or do you want to run on one of our LLMs inside of our Amazon account? All of those are available to you. We provide that transparency and flexibility so that your teams can choose the right tool for the job based on their comfort level. And again, keeping everything, secure and enterprise grade, because these are real conversations with your customers and they have to be secure so that you can be protected as a company, too. Yes, indeed. There's an aspect with Arroyo360, and observe that there's always that entry point into working with a customer. Like you said, there's an educational component to the consultation that we provide. If you look at businesses at large, and if you were to, give a high level advice to the corporate part of the business, maybe it's at the C suite inside the organization, what would you advise them about in terms of how they are going to incorporate AI into their customer service model. And do you think maybe the sales function of the business that organization should also be leaning into that and thinking more across a broader part of the journey and not just customer support? I think we've seen across the entire board, whether it's sales or customer support, a large need for agentic AI and a large need for AI in the contact center because of the large cost component that comes in with this. But I think more importantly, it's about the service component. And I think I would say or advise is like. Think about your customers, they're contacting you for a reason, you know, because they, something went wrong, they need support, they need help. And think about that experience. And I think if that is the first thing you consider and think about choosing, going down the AI route always ends up, always makes sense because you end up with use cases that should be, and could be automated. I'll give you a great example of this. My car insurance last month went up 12. I thought it's a little, little odd. I don't know why that happened. Right. It took 45 minutes on the phone to find out why. And all it was, was that my plan reset and it's the new year's plan. And this is just a new price, but that's a question I could have been answered and, three or four minutes. And, but I was on hold for so long that I was transferred. And this is a great example of just had there been a simple automation of, yes, this is my account number. Let's pull the diff. Here's, here's what happened, I would've gotten what I need and I would've been very happy instead of. Very frustrated waiting on the phone for 45 minutes to have that experience. Yeah, that's brutal. That is absolutely brutal. And I, and you look at the time value of money just for yourself as well, just going through that experience, it was worth a heck of a lot more than 12. Hey, Colin thanks for being on the show. This was a lot of fun. Thanks, Frank. I really enjoyed it.
