In this episode
Kevin Lee, Chief Technology Officer and Key Pursuits Leader at NiCE, explains why the bigger risk in an enterprise AI program is not a pilot that fails but a pilot that succeeds and never reaches production. Recorded at Ai4 2026, the conversation traces the specific gap between a controlled demo and enterprise-scale deployment: compliance rigor, use-case selection driven by interaction data rather than executive intuition, governance proportionate to actual risk, and the cost discipline of matching the model to the job. Lee argues that the organizations moving fastest are the ones treating regulatory constraints as a design input at the onset, buying platform breadth instead of point solutions, and measuring maturity by how quickly they can expand use cases — not by how many pilots they have running.
Key takeaways
- The pilot trap is a scale problem, not a technology problem. Something that works for 10 customers or 10 interactions is an entirely different engineering and compliance proposition at 10 million.
- Partner selection at the onset determines whether a pilot can ever ship. If the solution can’t meet enterprise scale and compliance rigor, the pilot stalls and goes back in the closet regardless of how good the demo was.
- Low-hanging fruit doesn’t reliably yield high ROI. The obvious use cases executives name first — order status, claim status — are the ones least likely to pay back the resources invested.
- Mine historical interaction data to choose use cases instead of guessing. Brands already hold the record of which interactions are worth automating; the answer is in the data, not in a brainstorm.
- Compliance is a design input, not an obstacle. Organizations getting the best results name CFPB, HIPAA, and PII requirements up front and let those constraints drive partner selection, rather than shying away from regulated workloads.
- A single hallucination is unacceptable in a regulated industry. Ten questions cannot produce twelve different answers; guardrail layers like NiCE’s Guardian AI exist to make outputs predictable, not merely plausible.
- Not every use case needs generative AI. Deterministic NLU answers “where is my order” definitively and cheaply — there is no reason to burn expensive frontier-model tokens on a package-status lookup.
- Mature buyers have stopped buying point solutions. Assembling 15 bolt-on tools is not a scale equation; portfolio depth from a single partner is what lets programs expand.
- Expansion velocity is the maturity metric. The difference between advanced and stalled organizations is how quickly they can turn on the next use case, not efficiency gains on the first one.
- Chat pilots cover a small slice of real volume. A chat deployment might represent 8–10% of interactions; enterprise-scale voice is where the rest lives, and a single platform turns that on without running a second pilot.
- Automation work compounds into agent augmentation. The same integrations, SOPs, and knowledge built for agentic capabilities can be surfaced to human agents to reduce cognitive load in the moment.
- Token economics will look materially different by 2027. Model availability and run cost are becoming the front-and-center variable, and partners will be judged on how fast they operationalize those savings.
Chapters
- 0:00 — The real risk isn’t the pilot failing, it’s that it succeeds
- 1:50 — Kevin Lee’s role at NiCE: CTO and Key Pursuits Leader
- 2:12 — What the NiCE platform does across voice, chat, SMS, and WhatsApp
- 3:41 — The Million Dollar Pilot Trap and the strategic missteps behind it
- 5:06 — How to make a pilot scalable from the start
- 5:45 — Why low-hanging fruit often isn’t worth the squeeze
- 6:51 — The industry-wide refocus on outcomes over activity
- 8:01 — Getting governance proportionate in regulated industries
- 9:53 — No black boxes: why determinism is the requirement
- 11:24 — The first questions to ask when a client wants agents
- 11:58 — Deterministic AI vs. frontier models, and the cost of guessing
- 13:14 — NiCE Labs and choosing the right tool for the job
- 14:21 — What separates mature buyers: portfolio, not point solutions
- 15:54 — Measuring success by expansion velocity
- 17:45 — The compounding effect of a single platform
- 18:31 — Chat is 8–10% of volume; voice is where scale lives
- 20:00 — Reusing automation investments to augment human agents
- 21:03 — What changes by Ai4 2027: token costs and model availability
- 22:21 — How Kevin Lee stays agile
Why a successful pilot is the more dangerous outcome
A pilot that fails is cheap information. A pilot that succeeds creates an obligation the organization may not be able to meet. Lee describes teams pressed by leadership to “go do something with AI,” building a genuinely impressive demo, and then hitting the wall of enterprise scale — the rigor and compliance required to move from ten interactions to ten million. The demo isn’t the hard part. The absorption is.
Choosing use cases from data, not from the top of your head
The use cases executives volunteer first are usually the wrong ones. “Where is my order” and “what’s the status of my claim” surface immediately because they’re familiar, but Lee’s point is that familiarity and ROI are unrelated variables. Brands already own the historical interaction record that answers which conversations are actually worth automating and which will pay back the investment. Using it removes the guesswork from the single decision that most determines whether a pilot survives.
Why deterministic AI still wins a large share of the work
Matching the model to the job is a cost decision as much as an accuracy one. For a status lookup, deterministic NLU returns the same definitive answer every time at low cost — Lee’s framing is that there’s no case for spending a hundred expensive frontier-model tokens to tell someone where their UPS delivery is. Frontier models earn their cost on multi-threaded, emotionally complex conversations. NiCE Labs exists internally to evaluate released models and map them to the use cases where they actually pay.
Governance proportionate to risk, not applied uniformly
In payments, healthcare, and financial services, the compliance requirements are fixed and known. Lee’s observation is that the organizations getting the best results treat that as an early input — naming CFPB, HIPAA, and PII obligations before selecting a partner — rather than something to route around later. Agents are less forgiving of undocumented workarounds than previous systems were; they expose the seams. The standard he sets is bluntly stated: ask ten questions, get one predictable answer, because a single hallucination is unacceptable in a regulated context.
The compounding math of a single platform
The clearest illustration in the conversation is channel coverage. A chat pilot is comparatively easy to stand up — text-based, contained — but represents roughly 8–10% of interaction volume. The rest is voice. On a single platform, turning that same experience on for voice at scale means provisioning numbers and handling parallel call volume, not running a second pilot, sourcing another vendor, or integrating another stack. The same principle carries into agent augmentation: integrations built for automated agents (Salesforce, Marketo, backend systems and SOPs) can be surfaced to human agents to reduce cognitive load, so earlier work keeps paying.
Expansion velocity as the real maturity signal
Efficiency is the metric organizations reach for first. Lee’s answer for what actually distinguishes advanced customers is pace: how quickly they can address the next opportunity after the first one shipped. Organizations still swirling around step A are running pilots; organizations that chose the right partner and settled compliance early are counting how many use cases they can add and how fast.
FAQ
What is the “million dollar pilot trap” in enterprise AI? It’s the pattern where an organization builds a successful AI pilot that can’t be scaled into production. The pilot works at small volume but fails the enterprise-scale, compliance, and rigor requirements needed to serve millions of interactions, so it stalls indefinitely.
How should companies choose which AI use cases to automate first? By mining their historical interaction data rather than picking obvious use cases from intuition. Kevin Lee notes that low-hanging fruit like order or claim status looks appealing but often doesn’t deliver meaningful ROI, and the data reveals which interactions actually justify the investment.
When should you use deterministic AI instead of a generative model? For definitive, repeatable lookups — order status, claim status — deterministic NLU gives the same correct answer every time at far lower cost. Frontier models are worth their expense on multi-threaded, complex, emotive customer conversations.
How should AI agent governance work in regulated industries? Compliance requirements like HIPAA, PII handling, and CFPB obligations should be named at the onset and used to drive partner selection. In regulated contexts the system cannot behave like a black box — outputs must be predictable, and guardrail layers exist to guarantee that consistency.
What distinguishes mature enterprise AI buyers from less mature ones? Mature buyers stop assembling point solutions and buy portfolio depth from a single partner, and they measure themselves on expansion velocity — how fast they can bring the next use case live — rather than on efficiency gains from a first deployment.
Why does a chat pilot understate an AI program’s real scope? Chat typically represents only about 8–10% of interaction volume. Most volume sits in enterprise-scale voice, so a program that only proves itself in chat has not yet been tested against the channel where the business actually operates.
About Kevin Lee
Kevin Lee serves as Chief Technology Officer and Key Pursuits Leader at NiCE, where he leads the company’s technology vision and its most strategic customer engagements by aligning platform capabilities, AI strategy, and architectural vision to deliver differentiated outcomes.
Kevin joined NiCE in 2021 and has held multiple leadership roles across the company’s digital and go-to-market organizations. He founded and scaled the Customer Service Automation team, transforming it into a cornerstone of NiCE’s growth strategy. He has played a pivotal role in many of NiCE’s most competitive and highimpact pursuits, shaping both technical and commercial strategy across global opportunities. Prior to his current role, Kevin served as Global Head of Digital Sales & Strategy, where he led the growth of NiCE’s digital portfolio and helped customers transition into the self-service era. With more than 15 years of experience in cloud and software sales, Kevin brings a comprehensive approach to account strategy, technical storytelling, and executive engagement.
Kevin Lee on LinkedIn: https://www.linkedin.com/in/thekevinlee/
———- Resources ———-
Coca-Cola FEMSA, NiCE: https://www.nice.com
The Agile Brand podcast is brought to you by TEKsystems. Learn more here: https://aglbrnd.co/r/2868abd8085a9703
We’re proud to be a media partner for #MAICON26 – Oct. 13-15! Learn how AI can power your marketing and business and help you grow smarter. Use code AGILE150 to save! https://aglbrnd.co/r/7fe458ced0f04658
Reach your customers with Reddit. Spend $500 in ad spend, get $500 back in ad credit! Learn more: https://advertalize.com/r/491818c79fb1873f
Chaser is the only Slack-native project management platform that helps teams turn messages into tracked tasks, automate follow-ups, and maintain team-wide visibility, without adopting another tool. Now integrated with Claude and other GenAI tools. Learn more at trychaser.com and use code AGILEBRAND for a 3-month free trial (normal trial is 14 days).
The most influential minds in software, AI, and engineering leadership will be at WeAreDevelopers World Congress North America, September 23-25 in San Jose. Learn more: https://aglbrnd.co/r/60a7299222a7bcf1
Start building your own apps with Replit and get $20 off. Learn more: https://aglbrnd.co/r/93531742a7625a20
Enjoyed the show? Tell us more at and give us a rating so others can find the show at: https://aglbrnd.co/r/faaed112fc9887f3
Connect with Greg on LinkedIn: https://www.linkedin.com/in/gregkihlstrom
Don’t miss a thing: get the latest episodes, sign up for our newsletter and more: https://aglbrnd.co/r/35ded3ccfb6716ba
Check out The Agile Brand Guide website with articles, insights, and Martechipedia, the wiki for marketing technology: https://www.agilebrandguide.com
The Agile Brand is produced by Missing Link—a Latina-owned strategy-driven, creatively fueled production co-op. From ideation to creation, they craft human connections through intelligent, engaging and informative content.
Transcript
[00:00:00] Greg Kihlström: Hi, I’m Greg Kihlström, host of The Agile Brand, and here’s a question for you. What if the real risk in your AI program isn’t that the pilot fails, it’s that it succeeds and you still can’t get it into production? Agility isn’t how fast you can launch a pilot, it’s whether the organization can absorb what works and still behave like one company while it does. Today we’re going to talk about moving enterprise AI from theory to practice in customer experience. Specifically, we’re gonna cover what separates an AI pilot that scales from one that quietly stalls, how to govern AI agents proportionately, enough accountability to be defensible, not so much that nothing ships, and how to tell a program that’s creating business value from one that’s simply handling volume. To help me discuss this topic, I’d like to welcome Kevin Lee, CTO at Nice. Kevin, welcome to the show.
[00:01:31] Kevin Lee: Thanks for having me.
[00:01:32] Greg Kihlström: Yeah, looking forward to diving in. Definitely a timely topic, and, the right place to do it, at AI Four here.
[00:01:38] Kevin Lee: That’s right. This is amazing. There are so many people here.
[00:01:40] Greg Kihlström: I know, I know. It’s, it’s, it’s, pretty, pretty overwhelming. [laughs]
[00:01:44] Kevin Lee: It is.
[00:01:45] Greg Kihlström: So yeah, yeah. before we dive in though, why don’t, you give a little background on yourself and your role at Nice?
[00:01:50] Kevin Lee: So Kevin Lee, CTO and Key Pursuits Leader. I’m at the intersection of product and technology and the brands that we partner with, helping them understand our technologies and how to extract the most efficiency out of it.
[00:02:04] Greg Kihlström: Great. Great. So for those listening that maybe are not as familiar with Nice, can you give a little background? You know, what, what does, what does the platform do? Who do you serve?
[00:02:12] Kevin Lee: Sure, sure. Yeah, so Nice is, the world’s leading, CX, software organization. So we provide the, the solutions that help facilitate all of the interactions between a consumer and the brands that they choose to, to do business with. So you can think about any time that you’ve ever reached out to your bank, your cellphone company, your utility provider, insurance, et cetera, whether that be a voice call or a digital interaction, whether that’s web chat, SMS, or, something like WhatsApp.
[00:02:44] Kevin Lee: We facilitate all of that, and then, when a consumer reaches out, some of those interactions are self-served through technologies that we provide, automated through virtual agents with agentic capabilities, and in some cases, escalated to a human agent that is augmented with the capabilities that we provide them, taking a lot of, the cognitive load off of those agents so that they can focus on the matter at hand with the consumer.
[00:03:11] Greg Kihlström: Yeah. Love it. So yeah, let’s, let’s dive in. We’re probably gonna touch on a little bit of, of all of that, here. But wanna start with, so you had a, a solo session, at AI Four here, called The Million Dollar Pilot Trap, and I think a lot of people are probably feeling, things related to this, which is, um… Well, may- maybe explain a little bit, you know, what, what is the trap and what are some of the common strategic missteps that cause it?
[00:03:41] Kevin Lee: So what we see is that a lot of organizations are being, pressed by their leadership to implement AI-
[00:03:48] Kevin Lee: … and go do something with AI, and that’s, where they fall into this pilot trap, right? Let’s pilot fast, let’s get out there, let’s go do something. But what we’re finding is that a lot of organizations are running up into this roadblock, which is enterprise scale, right? They, they tried something, they put together something, and it was really cool. It was a great demo, but ultimately, taking it from 10 customers or 10 interactions to 10 million is an entirely different thing. The scale, the rigor, the compliance that goes behind that, that level of interactions is what, sets apart organizations like ourselves at Nice and these point solution providers that are running into those pilot issues.
[00:04:35] Greg Kihlström: Yeah. So, I mean, to your point, it’s, a pilot can be highly controlled, it can be relegated to a, a subset of customers a- and things like that.
[00:04:46] Greg Kihlström: you know, what, um… I guess, what should companies be thinking about? I mean, ’cause there’s still value in, in doing the pilots.
[00:04:55] Kevin Lee: Absolutely.
[00:04:55] Greg Kihlström: I mean, you gotta, you gotta learn and, t- in order to innovate. But, you know, what should companies be thinking about to make these pilots more, you know, more scalable from the start?
[00:05:06] Kevin Lee: Sure. so there’s two things that we would say that organizations should think about at the onset. One is a bit more discretion-
[00:05:14] Kevin Lee: … at, on choosing the right part. Right? Because y- you will hit that point where you say, “Hey, I wanna take this to 10 million interactions, 10 million customers.”
[00:05:23] Kevin Lee: But if it doesn’t meet that compliance rigor that, and that enterprise scale that’s needed, then it’s just going to stall, and they’re just-
[00:05:33] Greg Kihlström: Yeah
[00:05:33] Kevin Lee: … gonna go back, into a closet and, and it’s not gonna see the light of day. secondly, the, what we’re seeing a lot of is organizations will pick the proverbial low-hanging fruit.
[00:05:45] Kevin Lee: Turns out the low-hanging fruit doesn’t always yield high ROI.
[00:05:49] Kevin Lee: And it, it’s not necessarily the wor- the juice is not worth the squeeze, as the saying goes.
[00:05:54] Kevin Lee: Right? So utilizing their data, the, historical interactions that these brands have, mining that to understand- Which interactions are actually worth automating? Which ones will pay back the returns on that pilot and the resources that were invested? So those two things we see need a, a little bit more focus or and a little bit more attention. and then once you choose the right partner and then you’ve identified the right use cases to then go apply agentic capabilities and augmented capabilities, that’s where we see organizations being able to take off and, reach escape velocity much faster.
[00:06:31] Greg Kihlström: I mean, that, yeah, low-hanging fruit, that phrase is used so much.
[00:06:35] Kevin Lee: That’s it.
[00:06:44] Greg Kihlström: … it might be easy, it might be quick, it might be cheap, but value has to be considered in there too, right?
[00:06:50] Kevin Lee: That’s, that’s absolutely it.
[00:06:51] Kevin Lee: And you know what’s interesting is there’s this new move or, or refocus toward outcomes.
[00:06:57] Kevin Lee: And I’m seeing this across AI writ large, regardless of the domain and the vertical, whether you’re inside of CX or if you’re inside of ERP and CRM.
[00:07:06] Kevin Lee: Right? Driving an outcome is the focus now. It’s not just, you know, doing something.
[00:07:12] Kevin Lee: And, that’s where you come back to the, the low-hanging fruit. Is the low-hanging fruit going to give you the juice that’s worth the squeeze?
[00:07:20] Greg Kihlström: Yeah. When, I mean, it’s, it’s good, right? outcomes are, you know, it’s kind of why we’re in business.
[00:07:26] Kevin Lee: That’s true. Exactly.
[00:07:27] Greg Kihlström: So it, it seems like the right focus, and yet, you know, that, that also brings, you know, something that I know you’ve identified as well, you know, the- it just brings greater attention to the need for governance. And I know that’s another challenge that a lot of organizations are running into. Where do you see companies getting, you know, getting the proportion of, you know, innovation, governance, all, all of those things wrong, you know, maybe applying the same review weight to a low-risk change as a high-risk one, you know, things like that. You know, where, where do you see things going off, off the track?
[00:08:01] Kevin Lee: So that goes back to one of the points I made earlier about enter- enterprise scale and organizations that can, deliver the compliance rigor-
[00:08:11] Kevin Lee: … that’s needed. Because it turns out, to no surprise, that when you’re in a highly regulated industry like a, pay or a healthcare-
[00:08:21] Kevin Lee: … FinServ, that compliance, those requirements are always gonna be there. So instead of shying away from it, the organizations that are yielding the best result are the ones that lean into it, and go, “Okay. There is this organization called the CFPB. There is these organizations or these compliance standards like, HIPAA or PII-
[00:08:42] Greg Kihlström: PII
[00:08:43] Kevin Lee: … and others. Let’s think about them now, and that’s going to drive us, to the right types of partners that can a- address those and partner with us.” And so that when we do get past the pilot, we again can escape that, that-
[00:08:58] Kevin Lee: … orbit, that, that velocity that we need to get i- into outer orbit.
[00:09:03] Greg Kihlström: Yeah. Yeah. Well, and I think maybe along those lines, too, another part of this is, you know, people have a tendency to maybe absorb some of the complexities or some of the workarounds or things that-
[00:09:16] Greg Kihlström: … a lot of this, this governance takes on a different approach when we’re talking about agents, right? I mean, they’re a little less forgiving in some ways. They kind of expose some of the seams that, again, people might have worked around stuff and, and gotten it-
[00:09:33] Greg Kihlström: … in compliance, but-
[00:09:35] Greg Kihlström: … in sort of an ad hoc way more times than they might want to admit. So, you know, what are you seeing? How does an organization kind of take stock when there’s a lot of things sort of happening that are, let’s say, less documented than they maybe should be?
[00:09:53] Kevin Lee: s- again, going back into those highly regulated organizations and industries-
[00:09:59] Kevin Lee: … you cannot have that black box when you ask 10 questions and you get 12 different answers.
[00:10:04] Kevin Lee: Right? Right. It can’t be a magic eight ball. I ask 10 questions, I wanna know the exact answer that I’m gonna get back every single time. And, that’s the difference between organizations that have harnessed the models appropriately, are putting the governance in, the guardrails in that can guarantee that you’re gonna have the right outcomes every single time. things like Guardian AI on our side are what actually ensure that the answers coming out are what we expect, right? A-
[00:10:36] Kevin Lee: … a single hallucination is unacceptable for highly regulated organizations, and that’s why we take those extra precautions and steps.
[00:10:46] Greg Kihlström: Talking a little bit more about, you know, re- return on investment and things like that, you know, I, I know when we talk about AI agents, there’s a lot of… And when we talk about AI, that’s a huge umbrella. Even when we talk about AI agents, that also covers a lot of ground, you know, everything from rules-based workflows to a system that reasons and takes action on its own. when a client comes to you wanting agents, you know, what’s the, what are some of the first questions that you ask to make sure that not only are they asking, you know, what they’re asking for, but that the, you’re finding the right solution for them?
[00:11:24] Kevin Lee: I’ll, I’ll go back to what we talked about earlier, starting with the data, right?
[00:11:27] Kevin Lee: Now, why guess at it? We-
[00:11:30] Kevin Lee: … we, we need to drive toward outcomes. We don’t need to guess at what might be there because, kind of the, just executive human nature is to say, “Hey, the obvious use cases, the ones that come up to the, to the top of my head are, where is my order? What’s the status-
[00:11:50] Kevin Lee: … of my claim?” Those don’t deliver the ROI that we’re looking for. And it turns out you don’t need generative AI for that.
[00:11:58] Kevin Lee: You don’t need a frontier model that is hugely expensive to operate. Versus having something that’s deterministic in some basis. A deterministic AI, standard NLU, where you ask a question, “Where is my order? What is my status?” And it tells you very definitively every single time.
[00:12:18] Kevin Lee: And, that’s why using the data to identify which are the ones that we should use one type of technology versus another, which will then drive the results and the ROI we’re looking for. Because I, I, I’m not necessarily looking to burn 100 tokens-
[00:12:34] Kevin Lee: … at $5 apiece, right?
[00:12:37] Kevin Lee: To give you the status of your, your UPS delivery.
[00:12:40] Greg Kihlström: Right. Right. Well, yeah, and I mean, to me this feels like it’s mature, a maturing practice and capability, right? I mean, you know, I think three years ago everyone was like, “Yeah, let’s throw everything in,” and-
[00:12:54] Greg Kihlström: … it’s magic almost or something like that, and, and, you know, quickly realize that, to your point, you don’t need to generate… You shouldn’t generate everything, and you don’t need to generate everything. So it’s, you know, how much education is there in, in what you’re saying? You know, it’s, it’s essentially picking the right tool for the job, right?
[00:13:14] Kevin Lee: Sure. That’s e- exactly it. So y- leveraging the data helps us-
[00:13:18] Kevin Lee: … identify where do we focus. And then beyond that, we actually have what is called Nice Labs. It’s our organization internally that helps, actually evaluates all the models that are coming out, and the, the practical application for the right tool set and use case. So when someone says, “Hey, I want to adapt Where’s My Order?” Hey, deterministic AI, probably your best bet, low cost, electrons, that’s it.
[00:13:46] Greg Kihlström: Right. Right. [laughs]
[00:13:47] Kevin Lee: Right? But then someone says, “Hey, I want to address a multi-threaded, complex, very emotive conversation topic-
[00:13:55] Kevin Lee: … with our customer,” okay, maybe we’re looking at a frontier model.
[00:13:59] Kevin Lee: And Nice Labs helps these partner brands navigate those conversations, because ultimately we are not a, a, a model purveyor. We’re not building models-
[00:14:11] Kevin Lee: … right? We are here to provide the, the, the IP of our unique harness and the framework of CX1, our platform, to apply to that solution.
[00:14:21] Greg Kihlström: Yeah. What, what… Based on what you’re seeing with maybe some of your more advanced customers, you know, what’s the capability that kinda sets them apart from some of those that are still, you know, kinda climbing the, the capability ladder, so to speak?
[00:14:37] Kevin Lee: I think our more mature-
[00:14:42] Kevin Lee: … clients are not looking for point solutions.
[00:14:48] Kevin Lee: They’re looking for enterprise scale, and what they have come to find, age, wisdom, wh- whatever quality you wanna apply there-
[00:14:56] Greg Kihlström: Right. [laughs]
[00:14:57] Kevin Lee: … is that bringing 15 different solution sets and bolt-on solutions and, and bringing one for this and another for that-
[00:15:07] Kevin Lee: … is not a, a scale equation, and that finding the right partner that has the most complete portfolio with the right depth of wherewithal and knowledge and capabilities, that’s the winning equation, and that’s what we do at Nice, right? We have, everything inside the portfolio to address the customer experience equation, whether that’s on one side of the equation being employee experience, and on the other side being customer experience, right? We bring all of that into CX and, you know, digital channels, voice channels, silicon-based agentic agents, or augmenting our, our human agents as well-
[00:15:48] Kevin Lee: … all inside of one portfolio with the compliance and rigor and scale that we’ve been talking about.
[00:15:54] Greg Kihlström: Do you see, that they’re also looking for… You know, is there a difference in the measurements that they’re, they’re looking for, you know, beyond… Certainly there’s efficiency to be gained but, you know, are the more mature customers also looking for other types of measurements of success?
[00:16:12] Kevin Lee: So what I would say there is their ability to expand out at velocity-
[00:16:21] Kevin Lee: … and pace. That’s the, that’s the difference, between organizations that are swirling around the drain when it comes with pilots versus-
[00:16:31] Kevin Lee: … okay, we chose the right partner, right? We addressed compliancy, and now what they’re looking at is, how quickly can I expand the use cases and address these other, opportunities? If that’s what they’re focused on versus, okay, well, we still haven’t gotten past step A. And, it, it’s that maturity again, right? Looking at whizzbang and shiny versus hardened, capable, scaled.
[00:17:01] Greg Kihlström: Yeah. There is just so much pressure to prove, but there’s also so much pressure to just get started, so it’s like, is that… I know, I know we talked about this a little bit earlier but I, you know, I just, I talk with a lot of companies myself and they’re, they’re struggling to do both of those th- You know, to get started and yet as soon as they get started they’re, they’re tasked with proving the results. Like, what, you know, what does a leader kind of, what does a leader do in that case of, okay, I don’t know, like, do we start another pilot? Do we, do we double down on, on existing, existing things? Like, where, where should their heads be at?
[00:17:45] Kevin Lee: I, I’ll a- I’ll address it from a different lens.
[00:17:51] Kevin Lee: The thing that is helping our folks accomplish those other, these accelerated capabilities is the compou- pounding effects that we have with being a single platform. I’ll just give you a, a kind of a simple example.
[00:18:06] Kevin Lee: Okay. You, you’ve chosen, you’re a, OpenAI versus Anthropic shop, no problem. We’ve chosen-
[00:18:13] Kevin Lee: … the model. we wanna attack a certain use case, and we’re going to deploy it inside of a pilot for chat. Typically one of the easier ones, right? It’s text-based channels, ASCII characters, a- and we can do inside a web chat, no problem.
[00:18:31] Kevin Lee: The moment you say, “Okay, pilot to production,” I’m getting 8%, 10% maybe of volume inside of chat. Where is the other volume coming from? Enterprise-scale voice.
[00:18:46] Kevin Lee: And to be able to now turn on that same experience inside a voice at scale, that’s where the compounding effects of having a single platform are- become very acutely evident.
[00:18:58] Kevin Lee: Because now, guess what? You want a phone number in Atlanta, Georgia, New York, LA, great. We can bring up all those phone numbers for you, enterprise scale, toll-free numbers. we can take two calls or 2,000 calls in parallel. We have the scale, the ability to bring in that volume, and now s- speech enable those capabilities-
[00:19:24] Kevin Lee: … right now. We don’t have to go run another pilot. We don’t have to go find other technology stacks. We don’t have to go integrate them. We don’t have to worry about, latency, and SIP integrations, and jitter, and loss scores, and things that no practitioner and executive leader cares about.
[00:19:43] Greg Kihlström: Quite right. [laughs]
[00:19:44] Kevin Lee: They care that it works-
[00:19:45] Kevin Lee: … and that my customers can call it and get the answers to, where is my order, or whatever use case that they wanted. And so they were able to go from pilot in, let’s say, maybe an obscure channel, to at scale mainstream volume.
[00:20:00] Kevin Lee: And so that’s just like one of the examples of single platform, right? And then now you can imagine that there are scenarios where, okay, we’ve automated what we can, but where is our, our highest costs? Contact centers with human capital, with our human agents. How do we augment them? Well, all those things that we built, that we, that we automated, those integrations into Salesforce, Marketo, et cetera, et cetera-
[00:20:26] Greg Kihlström: Mm.
[00:20:26] Kevin Lee: Great, we can now leverage it to augment our human agents because the same knowledge, the same SOPs, the same backend systems that we integrated into our agentic capabilities, we can now avail to agents.
[00:20:38] Greg Kihlström: Mm. Yeah.
[00:20:39] Kevin Lee: And in the moments that matters when that consumer is talking to that agent, right? We can bring it into their viewport and take the cognitive load off. So you can start to see the compounding effects of having that, that platform that lets an organization focus on the value and not on the architecture.
[00:20:58] Greg Kihlström: Yeah. Yeah.
[00:20:59] Kevin Lee: And so that’s where there’s this drastic difference.
[00:21:03] Greg Kihlström: Yeah. Love it. Well, Kevin, thanks so much for joining. I’ve got two last questions for you as we wrap up here. So we’re here at AI4 2026. If we were having this conversation at AI4 2027-
[00:21:16] Kevin Lee: Ooh
[00:21:20] Kevin Lee: that is a great question.
[00:21:23] Greg Kihlström: [laughs]
[00:21:23] Kevin Lee: And if I knew that answer, uh-
[00:21:25] Greg Kihlström: Right, right. [laughs]
[00:21:26] Kevin Lee: … I’d play the lottery, right?
[00:21:27] Greg Kihlström: That’s right. Exactly it.
[00:21:29] Kevin Lee: You know, it’s so interesting. Just days ago, right, it’s been, the market’s changed because of, speculation on-
[00:21:36] Kevin Lee: … the AI market. You know, we’ve seen it go down, come back up.
[00:21:42] Greg Kihlström: Right.
[00:21:42] Kevin Lee: You know, new models are getting released. I think one of the things that we will see become front and center are, the model, models out there, the availability, and the cost to run them. Token costs are going to change. I think that’s gonna be-
[00:22:01] Kevin Lee: … drastically different in what we see in ’27, and, when organizations partner with the right organizations, they’ll be able to help them harness those cost savings very quickly and operationalize them. So that’s what we’re interested and focused on, so I think that’s one of the things that we’re gonna see.
[00:22:21] Greg Kihlström: Love it. And last question for you. what do you do to stay agile in your role, and how do you find a way to do it consistently?
[00:22:27] Kevin Lee: I listen to your podcast.
[00:22:28] Greg Kihlström: Awesome. Love it.
[00:22:29] Kevin Lee: I, you know, I think podcasts and, bite-size content are some of the things that really help keep me apprised and, and up to date on what’s going on. As you already know, it changes every hour.
[00:22:45] Greg Kihlström: Yep. [laughs]
[00:22:46] Kevin Lee: Right?
[00:22:46] Greg Kihlström: Every day, every single day.
[00:22:48] Kevin Lee: Every single day there’s a new model. The GPQA score for this model versus that model. There’s a new GPQA, rubric. You know, it, it’s just-
[00:22:57] Greg Kihlström: Love it
[00:22:57] Kevin Lee: … always changing, so, instead of trying to consume it in large chunks, this bite-size, format is something that’s really helping me get, stay up to date. And, also being okay with knowing that it’s going to change, and not having strongly held opinions, right, and being able to, adjust with the landscape is what’s helping me, you know, adapt quickly.
[00:23:24] Greg Kihlström: Yeah. Love it. Well, again, I’d like to thank Kevin Lee, CTO at Nice, for joining the show. You can learn more about Kevin and Nice by following the links in the show notes.
