Name: Why Most AI Rollouts Fail (and how to be the exception)
Uploaded: 2025-08-13T18:17:08.062Z
Duration: 58 min
Description: Why Most AI Rollouts Fail (and how to be the exception)

Transcript for "Why Most AI Rollouts Fail (and how to be the exception)": Hello, everyone, and welcome to today's webinar, why most AI rollouts fail and how to be the exception. This session is presented as a partner webinar with Assembled and Bluewave Technology Group and excited to get into the topic today. A quick agenda as we get started. First, we'll talk about, a little bit of an introduction, set the stage. We'll talk about the urgency of AI, why we're talking about it at all, and then we'll get to the crux of the conversation, why AI projects failed, what everyone came to learn here today. And then we'll cover off with some of the lessons from the field, some real life scenarios, and what we've seen work and not work in real life. And we will definitely be leaving time for questions at the end of the session. So if you have questions, feel free to drop them in, and we'll definitely leave time for those. As people are logging on and joining in today, just a couple of housekeeping items. There is a chat and a q and a box on your screen. You are welcome to make a liberal use of it. I see some people are dropping some stuff in, so go ahead and find those boxes, drop in whatever you need. This is a broadcast only session, so if you want to participate and be heard, that is the way to do it. The session is being recorded. We will share the slides and the recording after the event. So there's a lot of content we'll be covering today. Don't feel like you need to screenshot or anything like that. It'll all be coming your way. So that will come within a day or two. Joined by joined today. My name is Lindsay. I am on marketing team here at Assemble, and I'm joined by Paul and Mae. Paul is coming from Bluewave, one of our partner technology groups. Paul comes with a wealth experience over twenty five years in the industry. Ton of knowledge working with companies of all shapes and sizes and tackling all sorts of problems. So really excited to have Paul join us today and share some of his experience. Also joined by my colleague, Mae, an AI deployment strategist here at Assemble. So one of the ones definitely on the front lines, one of the first and foremost to talk with you and go about your AI deployment. So definitely has seen it all, heard it all, and, lived a lot of stories in the last few years that we've gone through our our own AI journeys. And with that, I will hand it over to Paul to talk us through some of our AI change management. Sure. Well, good afternoon or morning, everybody, depending on where you are. Thank you for joining us. Well, the urgency of AI. I think if there are a few 100 of us here giving an hour out of busy days, that tells us a little bit about the urgency of AI and why it's on everybody's mind. But let's talk about a couple high level statistics, get a sense for the marketplace and the movements within it and where AI is headed. These, statistics we're going to cite over the next few slides are sourced from Gartner and Salesforce surveys largely from 2024 data. Speaks for itself, 64%, nearly two out of three CEOs consider AI to be their top investment priority. 700% is the projected market value growth from 227,000,000,000 of spend to 1,600,000,000,000.0 in the next five years. I hope my stock portfolio appreciates by that same amount. Wouldn't that be great? But that two thirds of the CEOs consider that to be their top investment priority. The challenge that puts upon everybody here, CEOs included, is that oftentimes there's a charter, getting your roadmap, figure out where it makes sense, but there's not a specificity. All that hard work is what needs to be figured out in terms of where we are going to invest and why, certain investments should be prioritized over others. Next slide. So if two out of three CEOs are saying that that is their top investment priority, let's talk about expectations versus reality. CIOs who are often tasked with much of the work involved in developing the roadmap, 92% of CEOs believe they will implement AI in 2025. Again, this was last year's survey, but 11% say that they are presently realizing value. That tells that really speaks to the urgency. We all know we want it, we all know we need it, we all know that there's a race to get it and deploy it effectively, but not many of us are reaping any value. Certainly, very few are reaping full value at this point. Next slide. A little bit of helpful guidance is that 90% of the CEOs believe that their focus must be to differentiate based on customer and employee experience. So that at least gives us a focus area, a target for where we should be looking for the utility of AI and where we could benefit from it from it most. Next slide. All of that informs really what we believe is an existential dilemma that companies are facing. We've all lived through, you know, tactical changes, outsourcing, transformational changes, premise to cloud, and the like, and all of those are needle moving, considerations. I think the AI phenomenon is a bit more even important than that, truly existential. If we think about how most organizations drive most of their technology spend towards keeping the lights on, systems of operation, network storage, communications and the like, And only 20% towards innovation and differentiation, those technologies that drive experiences, increase market share, differentiate brand. AI is an opportunity to do both. It's maybe to drive down some of that operational spend, but also drive more towards innovation and differentiation. But organizations and leaders are faced with this challenge. Labor is becoming harder to find and increasing in cost. Consumer expectations are always on the rise. CSAT is on in an all time low and maybe part of that is through ineffective deployment of AI. And so we're grappling with this, but the existential challenge is that if organizations are slow to adapt, especially slower than their competition, they will find themselves in trouble, ultimately spending too much money while losing customers and revenue. And so, again, we view this AI phenomena as an existential, concern for organizations to grapple with. Next slide. So given what we all, I think, stipulate and acknowledge as the criticality of of of AI, let's just talk a little bit about some of the high level trends or or impact areas for AI. Enabling business and technical performance, and so this kinda goes pretty deep and wide, but can I make better business decisions with better insights into cleaner, clearer data? Can I leverage AI instead of BI to be more prescriptive and predictive how I analyze my data? Can I automate and simplify manual and tedious work to deliver better experiences and drive down cost? Can I humanize and personalize experiences for employees and customers, again, to deliver better outcomes? Data is the new currency. It's the new oil. How do we graduate from describing what happened or quantifying how much, how fast, how quick, and learn more insight into why and how well and what's next. And that's a significant value proposition for AI, in the space of data analytics. Ajentic AI agents. Now, there's a lot of conversation here. Everybody's heard that. We're gonna talk a little bit more about machine learning and Ajentic and generative AI. But these are goal driven, human like, autonomous virtual agents to execute goals and take action. If if there's a word I've heard a thousand times that I'd never heard prior to two years ago, it's agentic. I think you all might stipulate the same thing. And the truth is is that when you look vendors in the eye and ask them if they're Agentic, they'll say yes. And if you ask them to explain Agentic, they probably many of them don't. It's something everybody is grappling with. I don't wanna dive deeper into that because we're gonna talk about that a little bit further into the, webinar. The the high value agent, all that's really meant to say is that, you know, work is going to be redistributed. And as AI and and human talent harmonizes, and we're gonna figure out for each individual organization how that happens, The work that people do, the human touch work becomes that much more important and is significant also in brand differentiation. So we are going to find more and more emphasis on the high value ambassadors for your organization, be that service, sales, or otherwise. And this is a big one, real time frameworks. If we think about training employees when they onboarded a company or performance improvement or even a contact center agent, or a salesperson, most of the improvement cycles are delayed feedback cycles, right, quarterly reviews, thirty day after corrective actions, very little is being done to get better in real time, at least up until now. But with AI, you know, in our meetings and in our conversations, analyzing and listening and driving insights in real time, we can be guided. We can be supported. We can be coached either immediately real time or post real immediately post real time rather than waiting, you know, days, weeks, and months for corrective actions. And so assuring positive outcomes while work is being done versus reacting to negative ones after is a significant driver and value proposition for AI. Next slide. So this here is only really meant to say one thing, that the way companies buy technology has to evolve. Technology innovation is moving so fast. The traditional RFI to RFP procurement cycle simply take too long, and the answers you get in your RFI by the time you make your thirteen month later decision probably aren't accurate answers anymore anyway. So how do organizations consider reorienting to fast fail, fast win cycles? We're gonna talk a little bit Mae is gonna talk a little bit more about that also, later on. The but this is a very real concern, and we're gonna talk a little bit more about this, but I've got one terrible analogy because I am a sucker for an analogy. Imagine you are seeking a life partner and you have options to go on dating sites, go to places where you know people of common interest may be, but your approach instead is to send a 38 page PDF to 25 prospective candidates that might wanna date you. And you fill that PDF with a bunch of rules and a 100 questions about them, and you only tell them a couple things about yourself. You're probably not gonna have a successful rate of dating, Yet somehow we think that's a good way to find technology matches and partner matches in the, in the technology space. I would suggest there are better ways. Commercial for Bluewave, leveraging an advisor who is an expert in the marketplace who can facilitate the the the matching of key requirements to a short list of concise well selected vendors, so you can focus your time and energy on the productive parts of the assessment rather than the the the effort to screen of a wide and fast marketplace of which eighty percent of who's being screened never had a chance to serve your needs in the first place. So let's just think about how we can reorient leverage adviser resources in your own internal resources to consider fast win, fast fail cycles rather than traditional long term procurement cycles. Next slide. So why AI rollouts fail and there's a bit of this that is also just why does any potential project within an organization fail. So let's get into that. The the high level statement I'd like to make here is that most organizations, in our experience, if there's one sort of common mistake, it's that too many companies focus on who rather than why, and they end up in market before they're ready without necessarily clear intent. And it's okay to use the marketplace to learn and understand the art of the possible, but oftentimes that just gets stuck into these prolonged, let's say, non business case justified cycles where we're engaging vendors and seeing demos and overloading information and creating confusion for ourselves. And so that the general thought here is focusing on why before who. So if any of you actually do this when you have a chance, Google what what is making good decision? What what are what do good decisions require? You'll get all kinds of access to notes and and and platforms and ideas, but here's what it's gonna tell you, and I think we all know this intuitively. A good decision means recognizing and researching the problem that the decision is intended to solve. Gather relevant data, analyze it accurately. Those things are hard to do in business cycles when you're working with vendors. Identify options and alternatives and establish your priorities, anticipate outcomes and risks, choose an act. Sounds pretty simple from something as simple as ordering dinner to choosing a vendor for an AI rollout. Those are the pretty much the the same steps in each decision making sequence. What are the threats to decisions? Why do we make bad ones? Lack of time, lack of reliable data, uncertainty on the on the completeness of our options. We just didn't know what was available. We didn't know what we didn't know. Lack of resources in the decision making process. Clarity on the scale of the problem. Some companies solve felt pain problems that aren't as impactful to solve as maybe a problem that's baked in that you don't even necessarily feel the pain. So that's where clarity on benefit of outcome becomes even more significant. I know what I feel, but which problem if I solve it drives the most return on investment? Which one in terms of effort versus impact has the highest ratio of impact versus effort? These are all the challenges that make decision making hard for organizations. And so if you skip past clarity on the problems and you haven't been clear eyed about the impact of solving the problem, then you're looking at solutions that maybe aren't gonna solve the full depth of the problems and you might wind up making no decision about 60% of decision making procurement cycles with corporate with companies end in no decision at all, and these are the reasons why. So we've get, among stakeholders in an organization, confusion, resistance, underachievement, and frustration. Click the click the button, please. How do we turn that into confidence, alignment, optimization, and preparedness? It's by focusing on the r, the return, before we focus on the investment. Focusing on the why, we're gonna do something before we spend too many cycles on who we're gonna do it with. Next slide. So this is just a super simple high level slide on some key differences of AI. And by the way, it's not one is different or they're all different. Not one is not better than the other. They align to certain workflows. And in some cases, with some platforms and vendors, you literally might you know, you call up to speak to a company, you authenticate your, your identity by offering a birth date or a four digits and something like that could be a machine learning interaction, and then everything else is an agentic interaction. So it's not always just either or. But machine learning focuses on learning from data. It's generally used for structured data analysis and predictive or deterministic flows. And I we have all interacted with chat bots and voice bots in those machine learning models and to varying degrees of success and frustration. Generative AI, and that's really kind of the term that exploded AI's awareness in the in the broader marketplace, both business and consumer with things like chat GPT and deep seek and the like. But gener generative AI creates new data, creates tasks like art and media creation. So when you're speaking to a more natural sounding voice spot today, that's a media creation from generative AI or content or writing stories or modifying code. Those are generative AI creative tasks. AgenTic, we talked a bit about that, autonomous decision making and action taking. I actually joke that I'm 58 years old and I don't know if I'm AgenTic yet. But anyway, the AgenTic AI oh, yep. Before you go. Agentic AI is goal oriented, and we'll talk a little bit more about it. But the general gist here is that if if any of you have ever seen or built, let's say, a conversational flow in a in a deterministic or machine learning model, it looks like an insane flow diagram or Visio chart, because in fact, it really is. And it's all of these if this then that with very specific words and matches it's looking for. And if everything hits, you contain. That's not how agentic is configured. Agentic is based on goals. It's based on guardrails, sources of data. It's really coached as you might coach a human, and and thus, able to handle much more complex multi intent interactions or workflows or drive a car, for instance. Next slide. So we talk about urgency and why it matters and, you you know, and sort of focusing on the why versus the how or the who, I should say. This is something that I I I we wanna offer you as a perspective, as a framework for how to inspect where in the body of your organization the utility of AI could be maximized. We call it frequency, friction, and value here at Bluewave. It's pretty straightforward. If something happens a lot, it's probably a little more important than something that barely ever happens. If something is a little harder to do, takes too much time, is too manual, creates frustration for employees or customers, well, that's friction. And if something and when I when I say something, I mean a workflow, prior authorization in health care, first notice of loss in insurance, you know, back office, customer interactions, anything, scheduling appointments. If if there's a value to the interaction or the work and and however you assign the value, revenue creation, revenue preservation, time spent, employee experience, CSAT. As you start to inspect the body of work across an organization, you you know what you feel, you know what what generally, what's kind of great and what's not so great. But if you put this into a matrix and you start to break down these workflows and you check the box frequency, friction, and value, if you check all three boxes, now you're starting to build a priority matrix for the kinds of problems you should invest in solving first. You're not gonna start with low value, low friction, low frequency. You're gonna start with high, high, and high. And as you build that matrix, you ask yourself some basic questions. And it's it's a little bit of work, but it's necessary work, and it's the work that saves you from wasting time out in marketplace without intent or purpose. So why is this workflow valuable to my business? How is it done today and by whom and by how many people? What data, systems, and processes are accessed to do the work? How much time is spent on it? How often does it happen? What's the the theory? The I wish I could on if I had AI, how would I use AI to support, augment, or eliminate the workflow? You're always gonna think about depending on, again, data that's accessed, how would I manage and secure this workflow as well, and then where do I start? I think if you and we do this with our clients. If you, again, look across the body of work in an organization and bring this frequency, friction, and value, we can start to demystify a little bit of kinda where to get started. Next slide. So let's talk about some examples of use cases. Here's something like 10 or 11 of them or 12 of them. I I'm not gonna go over all of them just for time, but we'll we'll cover the ones that are highlighted. But you can wash your eyes over the entire slide and get a sense for many of the use cases. And you might say, yeah, we're already doing that. And by the way, different models, different ways that you can procure and access AI are available to you. And some are narrow cast and embedded in solution sets. Some are available in these, you know, the global models like Gemini or ChatGPT. But if we talk about some basic business use cases, we already talked a lot about customer interactions. We'll go there because it's one that I think almost everybody here has experienced. Right? You all have customers, everybody's thinking about extending the hours of service and deflecting and containing and handling the menial and mundane and repeat tasks with automation to deliver better faster resolutions and so my people can do the more important and more critical work. And even if that's how it started, people then are starting to think more about how the agentic AI models can start to handle some of those more complex multi intent interactions. But we've all talked to a bot. We've all chatted a bot. So we start here. But this is probably in our engagements with our clients. Nobody's not asking or thinking or talking about this because everybody that we do business with has customers. Another use case is executive dashboards. And this is what I talked a lot about earlier. Power BI, Tableau, organizations that are doing the best to visualize and understand what happened. You know, that's the that's the descriptive capabilities. The graduation from descriptive to diagnostic, predictive, and prescriptive. How do I graduate from knowing what happened to affecting what will happen next? With generative AI query capabilities, a better ability to, interact with, analyze, and understand unstructured data sources versus purely structured data sources, there's been serious movement. So companies that are asking themselves hard questions that have to reconcile Excel files and CSV files and manual processes and and lots of man hour time to pour over data to make certain assessments and decisions around what's next, those workflows are being supported, accelerated, and done more accurately with the the onset of of AI. Next slide. We're not gonna go over this one in detail. You guys are gonna get the deck. This just gives you a little bit more sort of cross functional use case considerations, IoT, cybersecurity cloud, customer and employee experience. There are products and services just abound that deliver capabilities like this, and I and and it it's no easy task, again, to determine which one matters to me the most and then who do I go get it from. Those are all the ways that we help our clients and, you know, but but for now, we'll just leave this slide, leave it for you guys to evaluate, and, obviously, we're all gonna be accessible for follow-up discussions and conversations. So we can go to the next slide. So let's talk a little bit about the blockers because urgency, the why, ultimately then the who, what type of AI, all of those things never come to to fruition or or never come to light if an organization can't get past the blockers. And I'll stipulate that these are all considerations, but they shouldn't be blockers. So lack of security and observability. Well, everybody's wondering about disclosure and governance and GRC and, you know, and the data sovereignty. Organizations should adapt their vendor onboarding and infosec requirements processes. There are questions that need to be asked now that necessarily weren't asked before. Before you were asking a question, if you're a HIPAA, you're a HITRUST, you're SOC two, you okay. That checks my box. I'm good enough. Not every organization, some are much more stringent than that. So there is a a a consideration for adapting your vendor onboarding requirements and asking specific questions about data sovereignty. Also, when you're evaluating vendors, paying closer attention because there's so much movement in the space, so many players emerging from from the from sort of out of nowhere. You have to pay closer attention perhaps than ever to funding vehicles, to years in business, and and also even outright ask organizations about growth and exit plans, because many of the plays are to try to grow up to a certain point to be acquired, and those things can impact you. So there's certain business risk that just comes with the relative newness of AI. It should impact how you evaluate vendors. It shouldn't block you from adopting, AI. There are also platforms that are out immediately available to you today that most folks don't know about that deliver secure access to all the LMS you read about every day, the Geminis, the DeepSeqs, you know you know, chat t p t's, turbos, all of them that align to your enterprise control framework, single sign on, roles based access, attestation around data sovereignty, that can really facilitate and accelerate kinda use case efficiency by having access to these models and avoiding shadow IT and personal additions and all the like, that we could also talk to you guys about those options that are readily available that most organizations don't know about. Uncertainty on ROI. Well, we talked about that before. Focus on the why, get the r figured out before you get the I figured out, get the return part figured out before you start engaging vendors and getting price models. You get a price model from a vendor before they're really sure or you're really sure what you wanna buy and buy when, it's never the right price. It's never the final price. It's never the best price. And and it creates misperceptions as well. You might actually create a bias in your own evaluation process towards the lower costing vendor who in the end wasn't gonna cost less if if you actually ask for pricing at the right time. So focus on the why, define your use cases, project the value, get the return part figured out, and you can have more clarity and and, and alignment to to help support confidence in the decisions you make. Data complexity, I'll just say this about that. It's it does matter that your data is accurate and true, and many organizations struggle with scattered knowledge, inaccurate data. It is not necessarily the case. In fact, it's not the case that you have to go through significant investment to create, like, lake houses and pipelines if they don't already exist to unlock AI value. There are scenarios where that may be required, but there are many that it is not. And organizations that may think, I gotta go do that first, significant budget, AI readiness, it's a big thing. It's not necessarily the case. And so really evaluating and that's part of what that's why we ask those questions. What data do I access? How does this work get done? Do I really need to do these things in order to to start to achieve some benefit from AI? I'll also say this, that there are some tools available that can go into your data store and classify your data, rot, and risk. About 40 to 50% of most enterprise data is unstructured rot, Redundant, obsolete, stale, trivial data that just bloats your storage, complicates AI readiness, impacts your GRC initiatives, and that that data could be classified, could be deleted, could be stored in in different ways. And a lot of organizations are sitting on on a ton of that without necessarily being aware. And then the risk data, PHI, PII data that is stored in places it doesn't belong, like in the text of an email, for example. And so there are very effective low cost ways to get data classification so you can drive down the rot, mitigate the risk, and have more effective data strategies for AI readiness, something we would love to talk to any and all of you about. Bender Lockett, I hear this a lot. You know, who's gonna win? Stock prices, news releases, anthropic, this, that, the other. And the only thing I'm gonna say about that is that if you wait to see who wins the I AI race, the only certainty is that you will lose because there isn't going to be a winner. And if there is, consuming AI from the number one ranked whatever isn't necessary. If you think about your vendor mix today, where do where do those assessments even get made? So you pick the solutions that matter to you. You go through your vendor onboarding, your infosec, you're clear on on on use case, you're clear on return on investment, You don't develop rapport and and and relationship with the vendors you're evaluating. You work with honest disclosures. You tell them the truth. They they tell you the truth. You make the right decision. You both win. No gaps. So this is real. You know, the work people do is going to change the management, the deployment, the the the training of AI inside of organizations. You just look at, you know, kind of the the youth generation and what people are learning on and the technical skills and the majors that people are bringing into college, it the workforce is significantly changing. But I'm gonna give you a sort of a different perspective on this. There's no doubt workforce will will be reimagined and skills will will need to be realigned. Every organization, all of us here, are going to figure out how the harmonization of AI and human talent works best for us, and that's not the same answer for everybody. But there's going to be a day, and it's kind of already here, that agentic AI employees can be sourced, assessed, and onboarded from the open market and deployed in companies much like a typical person is today with a job description to to do work. Anybody you hire has a job description, a role, an expense to your organization with an expected output. A lot of that work is going to be done by Agentix, and you'll be able to access Agentix marketplaces and just build or procure and commandeer already built agents to do work like writing code, analyzing contracts, booking appointments, training employees, generating leads. It's very much starting to happen. And so the most significant impact on the workforce might be how companies what companies hire between people and agentic agents. Next slide. So now we're gonna hand it off to Mae to deliver the the meat of the presentation, the lessons from the front lines. Thanks, Paul. So, yeah, as Paul's been mentioning, I think as you all know and why we're here today, there's been such a surge of interest in AI agents and customer support, especially over the last few years. And definitely for a good reason, done well, they improve speed, accuracy, customer satisfaction, all those key metrics really well freeing up your human agents for that higher value work. But I think the hard truth is, and a lot of what Paul touched on, is a lot of AI rollouts, actually can fail and not necessarily because the tech isn't good enough, but really because of how it's implemented. So I'm excited to share with you, you know, a few of the top lessons that we've really learned from these real world deployments and, really, more importantly, how to avoid and how you all can avoid those pitfalls. So first lesson here is really starting with business outcomes, not AI for AI's sake. If your directive is purely we need AI, you're probably already on shaky ground. The teams that really win start with a crisp problem statement and a measurable business outcome. Your common goal might be resolution rate or deflection or containment or CSAT, really often tied to a specific channel or a specific use case. So it's really important to know which of those your, true North Star is. For example, if we were to set a goal of AI coverage and higher CSAT, while those aren't necessarily mutually exclusive, if you jump to if you jump straight to covering as many topics as possible without that operational plumbing, like making sure your AI agent has robust data connections action capabilities. In those cases, the AI agent is actually gonna have fewer tools than a human will in those cases. And the result is, ultimately, a worse experience for your end customer and a lower CSAT. So when you reframe that statement to automate high volume intense where an AI agent can provide the same experience as a human agent, that's where your CSAT will stabilize and you can really scale your automation sustainably. So what we recommend here, writing that one line problem statement that really defines what success looks like and get that cross functional sign off on what good looks like before you write a single prompt. Now when evaluating potential tech partners, that's really where you wanna be wary of those long and siloed procurement processes that really lead to outdated decisions. Eight months RFPs RFPs make sense in a world where the software is really static, but AI moves weekly. ChatGPT five came out last week. You know, by the time you finalize your RFP, the features that you're optimizing for might actually be completely different if you're in those really long cycles. And I think the clearest evidence where we really see examples for that is someone doesn't buy from us and then weeks later or months later comes back to reevaluate. Or we see customers who actually just purely reevaluate their vendor every single month. We've seen cases who have just siloed RFP processes, cases where someone has a separate RFP for your chat automation, another for your voice automations, another for your agent Copilot. That's three teams, three timelines, three stacks, and really means, essentially, those teams are doing about triple the work. You're delaying your learnings and really just struggling to share those data and metrics across those teams. So, you know, every feature parity difference can change in a matter of days, but, really, who aligns with your vision? What we recommend doing is running short outcome based bake offs with really production like pilots, and that way you can prioritize vendors who really ship fast, co build, and align with that vision of yours. Now once you do make your purchase decision, that post purchase execution can truly make or break you. So even with the right platform, launches can stall when ownership is fuzzy and inputs are messy. Even when a solution is chosen, that execution might break down if you have scattered knowledge bases or misaligned teams or lack of clear ownership. A lot of those same pitfalls Paul was mentioning as well. That technical content, those operational stakeholders, if they aren't on the same page, things are gonna be delayed. You're gonna get suboptimal outcomes. And so, really, I think the key here is that the AI deployment success depends as much on the cross functional coordination readiness as it does the technology itself. And speaking of readiness here, prepping your inputs for AI is just as important here as well. A scattered image only, stale articles, anything like that can really undermine even really the best model. If your knowledge lives across PDFs or screenshots or wikis that don't think with a live system, the AI really inherits that mess. So, when this is the case, instead of pausing the entire program, we recommend really creating a list of critical use cases. So, basically, what's enough to launch a narrow but excellent b one, earn that trust from your team and your broader colleagues, and then justify that process, for that broader kind of process and knowledge cleanup. Either way, AI or not, you're gonna have to plan for that prep and maintenance, so you might as well get ahead of it. And last one here, I think this is probably my favorite lesson. Don't let perfect kill progress, ship then shape. AI is so iterative. If you wait for perfect coverage, you are never going to launch. The winners start really small, prove value, compound. You know, we've seen companies that have hit their ROI in months, by really just focusing on a single high impact workflow and then expanding based off of those results. So pick that one workflow that's high volume, has clear boundaries, launch to a safe segment, whether it's in your sandbox or certain hours or geographies. And then once you see success there, expand one ring at a time to your next intense or additional channels. And I generally say to mindset wise, expect your v one to be imperfect. Your sandbox, it's gonna teach you a lot, but nothing will teach you as much as when your agents start using a Copilot or when your customers start interacting with those agents as well. So just to sum it up there, I think the five key things are, one, define your business case, the problem statement, the target metrics, the outcomes. Choose your vendor for pace and partnership, pilot in weeks, evaluate on outcomes and velocity, not that static checklist. Get really clear on the cross functional team involvement and resource properly, whether it's content, tooling, operations, each with their clear owners, and then really prepare your data and processes. So harden those inputs, understand what are your critical use cases, and connect to live systems where possible, and launch a narrow b one. One high impact workflow, production like conditions, create that learning loop super early on. AI agents are not magic. They are new operations level. So when you anchor to outcomes, shorten decision cycles, align on those owners, prep your inputs, and iterate quickly, you'll move from pilots that stall out to programs that really scale. So with that, I will pass it over to Lindsay. Yes. Thank you, Mae, and thank you, Paul. That was a lot of information, a lot of really great information. Glad we were able to cover it. I'm sure there's some thoughts swirling around in everyone's head as you're kinda digesting this. You both have a knack for making a lot of complex things sound really, really simple because you've lived and breathed it and gone through it, but, there's definitely a lot of me to unpack here. So, thank you for walking us through that. I'll give a quick overview of who Assembled is if you're not aware. Assembled is AI first customer support operations. We layer not only the AI agents and the AI copilot on top of the WFM. We've been in the mark in the space since 2018. About a 130 employees strong across San Francisco, New York, and London. So, really, what we optimize for and specialize in is meeting you where you are in your automation journey. So a lot of the times the case come as the case complexity goes up, that's where you apply a lot of your human agents. You apply your AI agents and some of the tier ones and some of the, lower complexity tickets. So you're able to kinda balance both of human and your AI agents all in one platform. So kind of that two brain approach. And what the outcome of that is, a lot of happier employees and happier customers. Right? Like, your agents are able to solve the tickets that where humans should be in the loop, and you're able to automate the ones that are just mundane and things that you shouldn't add your humans to. So that results in something like two times faster resolutions for companies like Flex Car, that are automating a lot of their, using Copilot to eliminate their draft replies. Something like Honeylove, support in more than one ways, using, case automation, for us in reducing escalations as they kinda go through their processes. Thrasio, a number of other ones that have layered the AI component into their human element, combined workforce management with AI agents and really seen the benefits and the outcomes of doing so. And a little bit more about Bluewave. Paul, you can speak to this one. Sure. Thank you. So, again, thank you everybody for being here today. Just a quick blurb or two about Bluewave. We are technology advisory. We help our clients make better decisions better, industry best advice around the increasingly complex technology marketplace to de risk your decision process. We do that through rigor and process, insightful discovery, glass box clarity, and concise search and selection to create and capture value for our clients. Next slide. It really is about insight and acceleration. We have a team of practice specialists across all the different towers that you see here. I'm not gonna read them all to you, but basically everywhere your technology can take you, we really our our what drives us is output and benefit to our client. As we work very, very hard, we remove all of the risk because of our engagements models engagement models with our clients is a no fee structure. We're vendor agnostic. We provide excellent clear data to drive great decisions for our client so you can speed up your value, increase your return on investment, and save time and money. We do that through a variety of rapid assessments that we've curated over the years to help organizations unlock the untapped ROI that sits inside of every company, and we see it over and over and over. The projects you need to execute, we helped you do that better and faster. The projects you might not be thinking about where there is untapped ROI, let's shrink the unknown and go get that ROI as a team together. We're a well aligned partner if your strategic objectives include the following, leverage your technology to effectively drive growth and differentiate employee and customer experience, to develop and deploy a multifaceted AI strategy that returns rapid results while caring for your GRC initiatives, and if you're looking to achieve effective cost takeout, simplification, or vendor consolidation across the areas of technology you see here on this slide. We look forward to potential engagement with any and all of you. Excellent. Thank you. And just to kinda bring it on home, the value of the partnership, Bluewave and Assembled have a number of joint customers. They see a ton of value in combining the strategic CX planning and with the software solutions, to align with the business needs and create those innovative differentiated solutions. You get the benefit of deep expertise and intimate knowledge of your existing systems and processes. You fit the right solution with the right size and stage, and layer on those proven methodologies to lay the foundation. It takes you all the way from pre implementation and planning through post implementation rollout and change and your partner along the way, as we've kind of seen in some of the customers that we work with together. So that's a little bit about us. If you're interested to learn more, of course, we'd both love to talk to you. So let us know. And with that, we will, as promised, leave some time for some of the questions from those of you on the line today. First question, I will just kick off because we get asked this all the time. Are and if you weren't here in the beginning, is this recorded? Yes. It is. Will you get the presentation materials that were shared today? Yes. You will. Don't worry. They will show up in your inbox. If you don't see them in the next twenty four hours or so, check your junk, check your spam folders. They're likely in there. If not, reach out to us, and we'll definitely get those over to you. So, yes, you will see all of the information today. Another one that I just like to kick off with because, we did have some colorful conversations in some of our planning calls. Paul, I know Mae touched on this, but what is one of the biggest mistakes that you see companies make in AI implementation rollouts? I'm gonna speak kind of bluntly here. Aimless, vendor interaction, road map for the sake of a road map, Over investing again in the who before they're really solidified on the why, that's really the biggest thing that we see over and over again, with organizations is it creates noise, it creates information overload, it creates confusion, and we've literally seen organizations that have maybe tied up the time of five, six, seven, eight valuable stakeholders, add up the time, add up the money, tens and hundreds of thousand dollars of man hours spent before a c suite even endorsed a particular initiative or project or anybody knew it could actually happen. So alignment, purpose, intent, and focusing on the why. Most organizations simply go to market too fast. And when they get there, because they don't have clarity on the problems they really wanna solve, because they're not necessarily sure on on buying timeline and intent, they often and and because typically the vendor screening process may be haphazard, this one called called me Gartner said that. Take a look at Gartner stock price, and that'll tell you kinda where their value may be. The market is way bigger than the than the seven or eight vendors Gartner tells you about. Companies might not have a super effective way of aligning appropriate candidates and shortlisting them so they can invest the time they need, and you wind up over investing time in vendors that could never have met your needs. Hopefully, you figured it out before you signed a contract with them and or your project never gets off the ground because it wasn't subsidized, it wasn't endorsed, or there wasn't clarity on the benefit of change. I think that's the biggest mistake that organizations make. Super helpful. Love to hear it because, hopefully, it can help some others. Question from the audience. Christian wants to know how to prevent shadow AI and focusing on single model or single model not realistic? Paul, do you wanna start with that one? Yeah. So well, here's the good news about shadow AI. It tells you that your organization is already clamoring for and wanting and needing it, potentially even benefiting from from AI. So in in a way, it's shadow adoption. Obviously, the bad news is there's data exposure risks, and you're not really operationalizing it in in in at scale ways that help the organization. Earlier, you heard me talk about platforms that deliver secure access to all of those LLMs that everybody has personal edition access to as well. That's one really it's not the only strategy, but one very effective way is to and we could talk about the specifics of it. I don't wanna get into vendor names and all that here, but platforms that enable access to all the models that people know about and go and figure out and access on their own, and do so in a way that, again, roles based access, SSO, templatization, prompt engineering support, so that people get to the right models for the work they wanna do and not everybody should be accessing the right models. There's really, really a low cost effective ways to shut down that shadow IT operation by giving employees access to the very same tools, but to do it again within the framework of enterprise controls, and governance and visibility. Awesome. Hopefully, that answered the question for you, Cromwell. Mae I have anything to add to that one? No. I think plus one to Paul said. I think that covers it. Awesome. Another question, that I think is on everybody's mind. Mae, can you tell us a horror story about a knowledge base that you've worked with and how difficult it was, to let maybe some people feel like theirs isn't quite so bad? Yeah. I think, I tried to touch on this one a little bit lightly, but maybe to give everyone the real, lowdown is we run into knowledge bases that I think the ultimate the worst is nonexistent and existent in some places where it's duplicative and also has incorrect and out of date information. I think the hardest case is is when it's truly just tribal knowledge passed down from agent to agent and there's nothing even written down. Thankfully, I have worked through some of those instances with customers where you focus on that one use case. How can we kind of improve and iterate, at least build out the process, or documentation for those first initial use cases and then really make the case for conquering, knowledge bases beyond that. But, yeah, anything that's images only, doesn't sync to live systems, doesn't have the right information or the right images, or just lives in people's brains, always makes it a little bit of a trickier rollout, and may maybe where you wanna scale down and focus on that specific use case first. And by the way, if I can add to that, I I mentioned this a bit earlier. At at Bluewave where we have developed a number of accelerated methods to find and deliver value for our clients. One of those assessment methodologies is a no cost sliver of your environment to do data classification. Again, we talked about rotten risk. Well, that's exactly what Mae just talked about. She talked about redundant, she talked about obsolete and stale data. There are simple, low cost, and effective ways to clean up that mess without having to invest again in large and expensive data lake house, pipelining programs, and projects, classifying the data. If you're going to go do that, it's it's like if you're moving out of your house to another house, don't you usually try to filter down what you really need and what you don't before you move? Same concept. So that's an assessment that we can execute with any of our clients at no cost to show you a glimpse into the environment and kind of a sense for, is it a panic button or do we have a life level concern for for the scenario of of rot files living in your knowledge bases and data stores. Totally. And I might just add to that as well. I think there are there are cases where you're just kind of improving knowledge all the time and also a really good product. If your knowledge is already in a place where you have everything relatively documented, it exists in one place, It's a source of truth. While you can always make things a little bit more legible for AI, really good models are actually starting to get ahead of that and translate things into more AI ready versions of your knowledge base. So once you're kind of in that place of at least everything's written down somewhere, you shouldn't have to do too much work beyond just getting those, key pieces of information down on paper. You're talking about vectorization? Exactly. Yep. So what you're saying is, even if your knowledge base isn't quite up to snuff, you're not you should be scared of, new AI rollout and deployment? Right. No knowledge base is ever gonna be perfect, and I think also the LMs are just getting so much better at reading through, and ingesting a ton of that information. So as long as we have something to work with, we can typically find our way there. Awesome. Awesome. Thank you. Well, it looks like that is it from the audience. I wanna thank you both again so much for everything that you presented today, all the information that you went through. Again, I know we kind of touch on a lot of meaty topics. We probably could've spent a whole day, a whole workshop on some of these, even the use cases themselves. There's a lot in there. So, hopefully, everyone got a taste of what is up, and how you can do more, learn more. So, again, I wanna thank everybody for joining today. Thank you for your time and your attention. If you want to learn more, you can go to Bluewave.net or Assembled.com to learn more, and, of course, the partnership joint solution is available to you as well. We do have a session coming up, on September 12, why you need modern workforce management in the area of AI, what's changed, why it's changing, and what it matters for you. So join us next time. We'll be sending link to join that one as well, and hope everyone has a great day. Thank you. Thanks, everyone.