Transcription
Welcome everyone. This is Mike Meyer, CEO and founder of Quick. I'm here today with Bill O'Neal and Dave Schlub Mill from IDC, and um, we're going to talk about some of the learnings and observations that came out of a recent round table that we did with CIOs from large brands.
So to kick things off, um, I'm going to ask my uh guests to introduce themselves. Dave, do you want to go first?
Sure, happy to do that, Mike. Uh, so my name is Dave Schlub. I'm a research VP uh with AI and automation at IDC, and I've been with IDC since 2012 and researching AI for since 2013. So uh, happy to, you know, be on this with you and uh looking forward to our discussion today.
Awesome. Thank you for joining. And uh alongside me is my co-founder, Bill. Go ahead and introduce yourself.
Thanks, Mike. I'm Bill O'Neal, the co-founder of Quick, also SVP of product of engineering. Um, been with Quick of course from the beginning. Mike and I've done this from the beginning. Prior to this, uh I worked at RightNow Technologies and later was acquired by Oracle. I've been in the space for um over 15 years and with Quick for 10 and definitely in the AI space for um, so since the beginning. So anyway, great to be here and um glad to follow up the round table. A lot of great discussion on that.
Thanks, Bill. And then I'll complete my introduction. Um, prior to starting this business, and Quick has been existence for about nine years, I was CTO for a company called RightNow Technologies, um and we built an ancient experience that was purchased by Oracle. Um, so I've spent a long time working in the the contact center and customer service uh software space, um and it's been a several years of rapid transition uh in the past couple years, so it's pretty exciting uh what we talked about the CIO Round Table. Um, so should we actually talk about what what happened in the CIO round table?
Sounds good. Good.
So we had a had a great conversation with a number of technical leaders at Enterprise Brands about agentic AI, and you know, uh I think, you know, it's been it was a really terrific discussion, uh, you know, but before we do that, maybe, you know, Bill and and Mike you can talk a little bit about uh Quick and and you know the what you folks are doing around agentic AI. You know, first and then we can kind of get into the rest of the discussion.
Perfect.
Well, um, just to introduce Quick, um we are a platform to help customers and businesses engage, and those Communications can happen inside of uh AI, so an automated conversation between an agentic AI agent and a consumer, or it could also happen uh because we realize that AI doesn't resolve every issue, could also happen between a human agent and the consumer inside of a UI digital messaging Channel like WhatsApp or Apple Messages for Business. And uh the Quick system automates those uh interactions with AI as well as has a complete AI powered contact center system that allows agents to be assisted by AI as well. So that that's what we do.
Um, uh in terms of uh the space that we're in, it's very interesting because uh I think we're going to talk about it in a second, um the agentic term has gotten very confusing in the space all of a sudden. I just want to kind of want to pile on what Mike was saying. AI's been I think hijacked for sales and marketing, um but we're losing track of the promise of that word. And if you look at agentic, the root of the word is agency, and so this is AI that's demonstrating true agency. Now we're not talking about unsupervised self-learning; that's a whole dangerous category, but we are talking about keeping people in the loop, but also allowing this AI and all this great technology around this to be leveraged in really creative and new ways so that it can solve problems on its own within, you know, within boundaries and within constraints. It can understand what goals need to be, um the high-level goals that a consumer might want, and it can actually have agency to go figure out how to how to solve those goals with the human together. So agentic really is more than just like RAG, which is question-answer stuff; it is really about talking to an AI that empathetically understands the consumer and goes and helps them get things done. So when consumers are interacting with their brand, they're not coming to ask questions alone; they're coming to get things done. And so agentic AI is kind of like a partner with your consumers to help them interact with your brand and get things done autonomously. So really passionate about this topic and really want to make sure we're all clear on what agentic AI really means. And I think from our discussion, you know, during the round table, it it really did seem like there was a lot of uh misunderstanding or confusion about what agentic AI actually is and you know how it works. You know, maybe, you know, Mike, Bill, you guys can, you know, provide us with, you know, your definition and and how you think of agentic AI and how it helps organizations.
I'm going to give a uh, there's I think some pretty technical definitions out there. I'm going to give like a layperson's uh definition, and I think the difference between generative and agentic AI is um really around. Bill mentioned the term agency a few moments ago, um but it's really around the ability to reason. And so the difference being, I think about generative as probably more so question answering, like like go and look up and an answer or tell me what my account balance is, and those things typically get done in self-service contexts, but a lot of inquiries to a contact center involve a human agent. And so a human agent um has ability to reason, plot of course, um react to if something is different than what was expected, take some adjustments, and then uh complete a multi-step workflow. And so when we talk about agentic AI, the simplest explanation is it's what your first-line customer service agents would do. Um, they're able to understand a whole bunch of different issues, respond to those issues, react to unexpected events in the process, and complete a multi-step workflow, whereas, you know, traditional uh gen of AI, and Bill mentioned the term uh retrieval augmented generation or RAG, that's more of kind of like a single an question and answer session. And so um really the promise of what we're working on is much more around that first-line customer service agent use case instead of just like information retrieval.
Yeah, no, that makes sense. And you know, you mentioned agency, Bill, and and I think the other aspect is, you know, I think organizations are looking for some level of autonomy as well, you know. So the question, you know, is, you know, rather than just having something that, you know, answers a question or, you know, provides a single response, you actually wanted to do something for you uh and and try to solve some kind of problem. And I think, you know, some of that, you know, some of that is what resulted in, you know, I would say a misunderstanding of the market about what agentic AI actually is and how it works. Um, you know, how are you folks seeing when you talk to Enterprises, you know, what are they looking for in a vendor, you know, in terms of agentic AI Solutions? How are you seeing that play out in in your market?
Yeah, it's a great question. So it either comes from um a lot of in the beginning when we first saw like OpenAI, a lot of excitement and a lot of unrealistic expectations what as technology can do. That's one side of it. On the other side is complete paranoia that this thing's going to take over the world and ruin your business. So what we need to do is really understand the buyer mindset of both sides of this and the reality is in the middle, right? And help them understand like the security aspects of it, the privacy aspects of AI to keep the people who are really concerned about what AI is going to do to their brand, and rightfully so, and then also the ones on the other side who really think this thing's going to replace all humans; that's not right either. So it's definitely understanding the the mindset where they're at in this journey and helping them align. And of course, underneath all of this is we need to understand the business goals. Early adopters are fantastic; I love working with them, but sometimes technology can get we get wrapped up in the technology and we lose sight of the business value. So whenever I work with customers, first thing you want to understand is like what is the business value we're going to deliver here, and what's some realistic expectations, and then what's some kind of constraints around security and compliance and that, and then once we get that vision, we can start to roll out typically incrementally um agentic AI in their in the Enterprise.
And you know, if if we talk about, you know, maybe switching gears a little bit, we spent a lot of time during the Roundtable talking about build versus buy, and I I think a number of the the folks that we spoke to are actually, you know, doing their own proof of concepts; they're actually out there, you know, attempting to work with the various Frameworks that are out there and and playing around with that. Um, you know, how do you see that that buy versus build? I mean, one of the things that IDC uh research has been telling us is that there's a shortage of AI engineering talent, and you know, we that really is causing a lot of issues for organizations because, you know, they don't have the engineering talent inhouse to actually develop these kinds of solutions. How do you see that playing out? Do you see, you know, people buying off-the-shelf, you know, agents or or is it going to be more of a, you know, Bill-type mentality or maybe kind of maybe a combination of both, you know, where people start with pre-built templates and then customize those things? How how how are you seeing that evolve in in your in your area?
I think first I I comment that, you know, build versus buy, it's not a binary choice. Um, there's a degrees in between where um in some situations you need, because of your business requirements and the uniqueness of your business, you may need to build that particular use case, but in a lot of situations um a business process such as customer experience, customer experience is fundamentally very similar across many different businesses. The um actual information that's provided is of course business-specific, but the process of delivering customer experience is the same across a lot of businesses. And one of the things that came out in the round table was an observation that was like strategically we need to build AI when it differentiates our business, but for common use cases like CX, it may not make sense for us to build internally because we need to to partner and leverage the um the investment that the world has made or a vendor has made in a particular solution. Um, so that was one one interesting thing. I think the the other um interesting thing as I think about build versus buy is the alternatives are often times raised again very much at the opposite ends of the spectrum. So on the one end of the spectrum you have um I'm going to pay a vendor to build this for me. Um, that'll probably mean I'll get a black box and I'll be dependent upon the vendor to deliver that uh solution. Um, I won't be able to see inside; I will be limited in how much I can control or maintain that myself going forward. And so like everything I just raised there, um control, um being able to understand what's going on, those are oftentimes the the concerns of about uh having someone build software for you. Um, but then on the other side of the spectrum, you know, the um build uh function is oftentimes phrased as we're going to go start with a foundational model; we're going to get an OpenAI license; we're going to put together a an open-source chain of LangChain and and um we're going to hire a bunch of people, and we're going to kind of like recreate the um technology from the the bottom of the stack to the top of the stack. And um from our perspective, um we respect people's desire to to build because the control and not being relying on a vendor and being able to see what's inside the black box, those are all bid business reasons. And so for that reason, we've built a platform that we just referred to as really a clear box. And so the tools that we would build AI for a company, and if they wish us to build for them, we're happy to; if they want to build themselves, we're happy to enable them to build themselves. Um, we're going to give them a clear box where we'll be using the same tools that they would use to to build. And so we have a um a spin on this whole situation that we say that says actually you should buy to build. And so in other words, you should buy a much more advanced platform and starting at the lowest levels, uh but then you can build, and in fact we'll teach you how to build uh alongside of us. And so that's part of our um secret sauce and our differentiation is that uh buy to build.
Yeah, no, that makes a lot of sense. That makes a lot of sense. Uh, and and the other aspect is um I think we noted a few of the participants talking about how they are using partners to actually help them provide expertise and things like that. Is that something you're seeing as well, that that the organizations you're working with are leaning on, you know, vendors like yourself to, you know, provide that level of expertise and and to help, you know, get started in the first place?
This world is these this world this AI world is very new, and um so as a result, there's not a lot of people who've done it before. Very little few people have done it before. Um, you know, we have that that um advantage because we've worked with a bunch of different brands, and so we've done it a few times, but we have we don't have five years of experience; we have 18 months worth of experience building on top of generative AI. Um, so I think that what you're seeing is um it's almost like the analogy is like if I'm going to climb Mount Everest, I need a guide. Like building AI is is complicated; it's risky. Um, I wouldn't think about taking a a complicated mountaineering expedition on my own; why would I try to build AI on my own? And so um but after I've done a few expeditions, um I'll understand how to climb and the what the elements might entail, and I can overcome those things. And so I think that at this point in time, the one of the things that the themes that we saw in the the webinar is the or the the round table is that people don't have the resources internally; they don't necessarily have the expertise; they're not going to be able to uh hire the expertise, and so they're going to bring um vendors in to help bring raise their level of knowledge, um because they want to be able to do it themselves over time, uh but initially they need that Sherpa to help figure out how to do the the first implementation.
Yeah, well said, Mike. And then I just want to pile on that as well: like we're all resource-constrained, no matter what business you're in. So you have this these innovation dollars; where are you going to spend those dollars? You going to spend them building up Frameworks, orchestration? I mean, there's a learning curve to this. By bringing in the right partner that's going to shave that learning curve down, there's two concerns here: one is vendor locking; we can talk about that in a second because we're really against vendor locking; we make your assets that you bring that you can take them out of Quick and go put them somewhere else. The point of this is like Quick better deliver value for you, and then back to those assets, what they look like: knowledge and APIs. So we need the knowledge so we can answer questions, and then we need APIs for the agency to go get things done. So when we work with companies, we really want, like Mike said, be the Sherpa, guide them and say, "Listen, you have this IT dollars; go build better APIs; go build the foundations that are specific to your business so that we can build an an exceptional agentic experience." So that's really how we try to guide customers. And then finally, it's totally a gradient as well. We have some customers who almost pair-program with us; they have their own AI engineers; they sit with us; we develop it together. And other ones, like we hand off completely, and they just, you know, anytime they want something changed, we change it for them. And of course, there's everything in the middle as well.
I I think the other interesting thing just is um nobody has figured this out. Like we had um, you know, the largest brands on the the round table with us, and there there were varying degrees of strategy; some were more side more on the build, some were more on the um. One of the important uh factors that somebody raised is it's not just the initial implementation, but the reality is these things are going to change very rapidly over time, and so the resources allocated to uh if you're going to build internally, the resources allocated to build probably are going to be on the project for the time uh going into the future because um new models are coming out; prior generations are being deprecated; reasoning is is changing. Like this stuff is not like build it and then put it on the shelf and it's like the project's going to be done. Um, it's going to be pretty heavily uh a pretty heavy ongoing maintenance concern as well. And so this whole build versus buy equation, it was really interesting; some folks were worried about maintenance; some people hadn't thought about it; some people were leaning more towards builds; some people were were leaning more towards buy. There's no clear-cut strategy at this point in time, and I don't think we every brand will ever have the the same strategy, uh but everybody's still plotting the course; it's an early stage in this journey.
Yeah, no, great points. Great points. And I think another aspect is um I think we noted a few of the participants talking about how they are using partners to actually help them provide expertise and things like that. Is that something you're seeing as well—that the organizations you're working with are leaning on, you know, vendors like yourself to, you know, provide that level of expertise and and to help, you know, get started in the first place?
Yeah, I mean, I think the we certainly have, and we work with clients in regulated industries. Spirit Airlines, for instance, is a as a client, and the DOT actually has a fair amount of Regulation related to the the um information that's provided uh from an airline. Um, we worked with some insurance companies. Um, so I think in general the concerns are raised are how are you going to use my data, and can you provide information accurately? Does the AI work well? Does it not hallucinate? And um, you know, we've worked through um both of those issues, um and we routinely um spend time with a brand during an uh initial vendor valuation to explain that we're not going to actually use your data for any other purpose than to serve you, and um we have techniques where we can use specialized AI to measure the performance of the main large language model and provide uh accurate information. So um it's, you know, the the um technology involved here at making AI be more accurate is getting better and better, and and at this point in time we can provide uh solutions that are far more accurate and trustworthy than a human agent um in most circumstances. So um that's the measure that we should be uh comparing it to; like brands are already accepting the fact that humans may fail, and there might be some liability associated with something that a human agent says. Um, if we do better than with the AI than the human agent, it decreased the overall risk, and that's really I think how people should think about the risk of AI.
Yeah, that's really well said, Mike. And then also everyone knows about hallucinations; that's a really common thing, but also bias in large language models is overlooked sometimes; it's also a real concern. So we have tools that will help you monitor bias as well as look at hallucination detection. Basically, we look at the output of the large language model; we look at the evidence we gave it from your knowledge base, for example, and we had to get a confidence score that how accurate is that response from the large language model? High confidence, we can return it back; low confidence, we keep it. But these are things that you're going to have to build out as well. And then also, like to the security and compliance things comes down on some other things. We have some customers who are insistent that we use their own large language model; they might have fine-tuned it; they might be bringing it themselves. So any vendor you work with should be LLM-agnostic; you can just swap out these large language models as needed. And then also, like Mike said, sometimes you just have to really understand the customer. Like when I work with Spirit Airlines, there are some topics that are really sensitive; you can imagine some topics you would say to an airline need to be escalated to a human agent, especially around security. Um, and there's also some things that are under the Department of Transportation where we can't return a generative response; it has to be a canned response. So just when you get in that nuance, you really have to have a vendor who understands the compliances and the that you're under, and so we can work together to navigate those complexities.
Great points. Great points. And you know, I think one of the outcomes of this round table and and and in this discussion, you know, is that we saw lots of lots of interest in agentic AI. You know, I I think we had lots of participants of the round table, and I think everybody is, you know, wants to make use of this, and it's just really a question of, you know, how do I find the right partner? How do I develop the right use case? How do I put that in process? And how do I measure return on investment once I do all of that? You know, so I mean, you know, at the at the end of the day, there there's a lot of potential value; I think people are trying to figure out, you know, how can I make that value work for my organization?
Yeah, and it's still early. Like if I looked at the web development, we can all remember writing HTML; a couple years down the road, JavaScript came out and unlocked the browser completely. Like I I'm still waiting for that those inflection points coming along in agentic AI that's really going to propel us not just evolutionary but revolutionary in this thing in a safe way, but we have yet to see those coming, and they will come. Um, hopefully Quick will be the very first one, and I think we're very early on this journey. I mean, I I think we've we've all been around for a while. Like if you think back to like we thought that the um the internet and PC personal computing was like changing the world, um but look how far it's come since like 1990. Um, and so we're we're very early. I'm I'm looking forward to, you know, in the future looking back on today and be like, "Oh my gosh, remember when we used to do this?" Like it's the world's going to change a lot. There is, and there's I know we're running short on time, but one thing I really want to drive home here is this: agentic is the first time that brands this responsibility with agentic. This is the key point; it has flipped on its head. Before, the consumers had to navigate your IVR; they had to download your mobile app; they had to go to your webpage; they had to interact with your chatbot. It was always on the consumer to figure out how to talk to your brand. With agentic AI, truly now it is your brand's responsibility to interact with the consumer in the way the consumer understands. If we can find deliver this promise, we have the tools to do it now; we just got to go forward and…
Do it. So that's a really important point. When people start talking about genic, that's the one takeaway I think people should take. Terrific. I think with that, it's a wrap. Thanks very much for your time. Thank you.