ExpertCast By ActualTech Media Unlocking Business Value: AI in Enterprise IT

  • Automox
Unlocking Business Value: AI in Enterprise IT

ExpertCast

Integrating artificial intelligence into core operations has shifted from an experimental initiative to an urgent strategic imperative for modern organizations.

Led by a veteran industry authority, this ExpertCast offers a practical roadmap for converting machine learning capabilities into measurable corporate return on investment. Participants will explore real-world implementation strategies that automate routine service management tasks, enhance predictive infrastructure monitoring, and optimize resource allocation across complex architectures.

The session highlights emerging software platforms, actionable integration frameworks, and proven methodologies for overcoming common deployment hurdles. By focusing on high-impact use cases rather than theoretical concepts, this presentation provides technology leaders with the technical insights required to evaluate cutting-edge tools, refine existing workflows, and scale automation effectively.

Equip your team with the knowledge needed to harness intelligent systems and drive sustainable digital transformation across your enterprise.

Why You Should Join:

  1. Discover cutting-edge platforms to accelerate automated enterprise IT workflows
  2. Learn practical deployment frameworks directly from a recognized industry authority
  3. Identify high-impact operational tools that maximize overall digital investment return

Agenda & Speakers

  1. Howard M. Cohen
    Howard M. Cohen Compelling Content Creator, Technologist, and Writer ActualTech Media
Full Transcript Keith Ward Happy Friday, everyone, and welcome to the actual tech media expert cast. My name is Keith. For those of you who don't know, I am with actual tech media, and I am super excited to be your moderator for this event today, which we are calling unlocking business value AI in enterprise IT. It is so good to see everyone today. Thank you for those thumbs up emojis and all the others. We love having you engage with us, and that is what I'm gonna be talking about right now as I go through some housekeeping items. Alright. First off, as always, we're gonna begin in the chat tab on the far right hand side of your console. This is where you can say hello. This, I like to call our virtual Starbucks or Pete's Coffee or wherever you get your coffee from. We just hang out here. We get to know each other. We talk amongst ourselves. We share stories. We send emojis. We do all of that fun stuff. So let's get into that right now. Let's say hi to a bunch of you that are out there. Howard, who we're gonna be hearing from very soon, I have a feeling, says greetings from Maricopa, Arizona, not Maricopa County. I have no idea what that means. Maybe Howard will tell us when he comes on. Alright. Cheryl is in Nashville. Great to see you, Cheryl. Greg is in Northwest Washington State, and Matt is in Ohio. Jim is in Northern Virginia outside of Washington, DC. Not that far from me. Robert tells us, bonjour from Winnipeg. Bonjour, monsieur. Greg is in Texas and Robert in New Jersey. Rob is in San Diego, and Judeo 4 is in Winnipeg. Oh, way up there north in Canada. Alright. Great to see everyone. Nick is in Seattle, and Jordan is in Colorado. I think that is how we say it correctly. So that is exactly what we do there. Let's keep that going. So, the other thing you need to know in that little window to your right is there are two other tabs. If you go to over from chat, you will see that it says q and a. And amazingly, q and a means questions and answers. So if you have questions for the brilliant presentation we're gonna be hearing momentarily, that is where you put them. So try not to put them in the chat, try to put them in the q and a. It just helps us keep them a little bit more ordered and easy to filter. So in the middle, in between those two is the docs section. Now there we have got all kinds of great content that goes along with the webinar today. Check it out and see what we have now. If something goes wrong with your picture, with your video, with your audio today, what you're gonna wanna do is let the amazing Deanna know also in the chat tab. She will hop in to help you out, and she is always right on top of it. Every time I say her name, she is ready with her bears football, emojis there. I love that. Even though I'm not a fan of the Bears, I am a fan of emojis. Alright. Next up here. Lower right hand corner, you see the word Best Buy and a $250. That is because I am giving that very thing away. I'm actually not giving that away because that doesn't really exist. But I'm going to be giving… Handing out a Best Buy card or actually you're gonna be winning one. You're gonna get an email saying, hey. You've got this money. I think that's how it works. I really don't know. But one of you is gonna get a $250 Best Buy gift card, I promise, at the end. Now you do have to be in attendance during the live event to qualify for the prize. So with that now, let us be done with all this intro. Oh, sorry. One more thing I forgot. I should be paying closer attention to what I write here. Upper right corner, you see closed captions are available. Now to get that, you will go to the lower right hand part of your screen in the corner, and you will find the drop down. There are lots of languages available. So if English is not your native one, pick one that is or at least is a lot closer to it. Okay. Now expert casts are special product neutral looks at technology problems presented by an expert in the field, but the support of our sponsors is what makes it possible for us to bring you this content. So I want to give them a special shout out today. Now there are resources available from our sponsor. You can see right there, which is Automox over in the doc section. So I encourage you to check that out as well. And now a few words from our sponsor. I will put on my announcer voice. Here we go. All right. We are supported today by Automox. Cyber threats move at AI speed. Your response should too. Automox closes the distance between vulnerability, discovery, and action so IT and security teams can patch what's patchable and mitigate what isn't. With AI powered automated vulnerability mitigation worklets, Automox turns emerging threats into ready to use engineer reviewed automations even when no patch exists yet. You know those as zero day exploits. That means faster mitigation, less manual work, and every Windows, Mac OS, and Linux endpoint hardened by policy every day. Patch, mitigate, remediate at the speed of AI. To learn more or get a demo, visit automox.com. I feel like one of those YouTubers doing the, the the sponsor promotion with that. How did I do everyone? I don't know. Probably not so great. But with that, we are now ready to get to the main event. It is my pleasure to introduce you to our presenter today, the great Howard m Cohen. Now as you can see here, Howard is a compelling content creator, technologist, and writer. We were actually gonna go with a non compelling content creator at first, but wondered if that would provide the most bang for your buck. So instead, we hired the very compelling Howard who is ready to go right now. And there is that compelling content creator. How are you today, Howard? Howard M. Cohen I am okay, Keith. Can you hear me? I just got a flash on the screen about muting. Keith Ward Can hear you just fine. Just. fine. So. we are… Howard M. Cohen gotta tell you. As… You did great. Your your announcer voice, you know John Stewart, you know Scott Galloway, but you're getting close. Keith Ward I've I've got a a little more work and then I can be raking in the big bucks, I think, Howard M. Cohen But Keith Ward on national or something. I don't know how that works. Howard M. Cohen you, you you started off making me feel like I'm on with Heather Cox Richardson, who calls out all the cities people are coming in from. That was really great. Maricopa, the city of Maricopa where I live, is in Pinal County, Arizona. We are 20 miles south of Maricopa County. Now most people know Maricopa County as a place where they had the frauds, and we had sheriff Joe Arpaio, who was out of his mind. We don't have anything like that here. We're very calm. We're a little city 20 miles away from anything. We've got beautiful horses running around in the desert around us. It's a nice place to be. Keith Ward It's. it's also by… It's also by the way, Howard, for for those of you who are cinephiles out there, the home of h I mcdonough and the raising Arizona set. So that's why I'm a big fan of Maricopa, so I'm old enough to remember that well. Howard M. Cohen Okay. Well, Keith Ward There you go. Howard M. Cohen time you're out here, we have to get together and hoist a few. Keith Ward Absolutely. Absolutely. I I cannot wait to get out to Arizona. I love it out there. Howard, I am gonna be done talking here. It's time to stop listening to my voice and start listening to your mellifluous tones. So I am gonna queue up your first slide here and hand it over to you. Remember, folks, keep asking your questions for Howard. Put them in the q and a tab on your right when he is done this amazing presentation. He's gonna be doing q and a. So, Howard, I'll see you for that on the other end. Take the floor, sir. Howard M. Cohen Sounds good, Keith. Thank you. Okay. Welcome, everybody. Great to have you here with us. As Keith mentioned, I'm Howard m Cohen. My official title is senior resultant. Resultant because since I started in this business forty some odd years ago, I'm all about results. You know? You you can tell me all the great things you did, but it doesn't matter if you didn't get the results you were going for. And that's why we're here talking about this today. The senior part, because as you can see, I'm getting on in years, and, that's kinda why you're not seeing me on video. I'm kidding. I just think, I want you to focus more on my content than on me. Okay? So that's, that's the story there. And we're gonna talk about integrating AI into core enterprise operations. And I want you to kinda think about that phrase for a while because it really differentiates what Automox helps you do, and I was very impressed with what I saw. But it also suggests where you can get the most bang for your AI buck. So let's let's get on with it. Any of you who have ever been in any of my presentations before know that I like to start every presentation with a word of the day. Today's word is a lot of fun. Today's word is outcomes. Here are a whole bunch of people celebrating a wonderful outcome after years of study. They've gotten their diplomas. They've graduated. Success is theirs. Now the fun begins. And outcomes are what everybody's talking about these days. That, you know, are you getting the outcomes you need from the things you're investing in? And that's really the goal. I mean, that is that is the underlying goal for everything. How does that relate to AI? Very good question. What we see today are employees who come running into their boss and say, hey, boss. Look. We got AI. And it doesn't matter which one you got. Now the boss, of course, thinking about the bottom line, thinking about, you know, business improvement, asked the question, what are we gonna do with it? Now if you have a great answer for that, I suggest go see the boss right now. However, if you're like this guy and you really don't know what you're gonna do with it, then hold off. Because what your boss wants to hear about is not necessarily, is not necessarily that you have it. They wanna know what it can do for them. The phrase that I like to use, and I've been saying this for a long time, you know, I break the secret that nobody in… Nobody invests in technology for the sake of having technology. They don't buy technology for technology's sake. For most people, it's really not about what it is. What is the technology? Is it chat GPT? Is it Claude? Is it Gemini? That's not what it's about. It's solely about what it does. And even more important, it's what it does for you. And let's go a step further. It's what it does for you that brings value to you. So let's stay focused on that. So right now, you have a whole bunch of people who are thrilled with their AI tools, and you're spending oodles of money on some of those AI tools. You're letting everybody have a shot at checking it out, and they're all happy, but they're not exactly sure what to do with it next. Now we'll be talking about use cases, as we go through this presentation, But let me tell you that a lot of people find finding use cases to be their biggest challenge. The right way to approach it is to find the problem. And if you identify a problem and you can identify that, hey. AI can help us with this, then, yes, then it makes all the sense in the world to start investing in AI solutions. And there are many good people out there who can help you do that. There are many good resources for you that can help you learn more about that. But I strongly suggest to all of you, don't worry about, should we get this one or that one? Worry about more about what do we need to fix that this could fix? What do we need to improve and make better that this can improve and make better? That's really the key to going in the right direction with AI. The gap isn't the access to the AI itself. It's integration in how your business goes, how your business works. And, you know, I make the point that everybody everybody today is using AI. We hear from everybody. They're all checking it out. They're all playing with it. Some are playing games. Some are writing emails. Some are making prettier pictures of themselves. You know, that's great. That's fun, but that's a toy. It's not really a business application. It's not what we're talking about. On the large enterprise side, when there was a recent survey, almost three quarters of the of the enterprises surveyed said, yeah. We use AI regularly. Then when we dug down deeper, we found out that only 10% say it's core to what they do. They haven't really integrated it deeply into their business even though in many cases, it would be the difference between where they are and where they could be. On the small business side, 86% of small… Of US small businesses have adopted AI. Now I want you to realize that includes restaurants that are printing new menus with it. It includes any kind of company that's printing flyers with it, writing better emails and all… You know, the the rudimentary things. And there is nothing wrong with that. When you think about paying $20 a month to be able to do those things or even being able to use… Do them for free in most cases, why not? And so, yeah, it's it's very, very likely that 86% or more of small businesses are… Have adopted AI, but only 15% have actually started to run their business on it. And once again, I'm gonna suggest very strongly that when they realize that they can, a whole new world's gonna open up to them. Now the name of this, session was from pilot to production, and there is a stat out there. I believe it is 75. Oh, we'll we'll hit it in a moment. But there's a large percentage of pilots that fail. So the theme of this particular slide is why most a pile… AI pilots stall. It was an MIT study, and it's almost a… It's over a year ago now, but they said that 95% of generative AI pilots never scale into measurable business impact. They never do anything for you. MIT published that, a very, very, very, credible source, and then everybody started to disagree. Everybody started coming out and saying, yeah. But you didn't let the AI use its tools. It can't do anything without its tools. You didn't let it have this. You didn't have… Give it access to that. You know, the fact of the matter is it's true. MIT just made challenges to a basically naked AI large language model, and they derive their answers from that. There have been many studies since that actually gives us much better statistics. So the the learning gap was that the tools that they could have been using, if they don't adapt to the workflow, they're gonna stall. And the model quality wasn't really the issue. You know, they… They've tested all the different models. Yeah. That wasn't the issue. It was the connection to tools and data, and we're gonna talk about the importance of data too. This pilot misplaced his money. No. It's been studied and found that over half of AI budgets today are going to sales and marketing uses, producing marketing emails, producing sales letters, producing brochures, laying out huge campaigns. They're capable of wonderful graphic treatment, and so that's wonderful. But what you're gonna find after today's conversation is that the strongest returns on investment are gonna show up in the back office where the core operations are, and that's really the theme of this one. Okay. One more pilot. We then return to what I call the build versus buy versus broker conundrum. I left out broker. Back when we were talking about cloud and even before that when we were talking about virtualization, we basically said that you can build your own. I mean, you can build your own cloud center in your building. You can buy the services from somebody else, or you can go to a broker, and they can be doing a lot of other things for you. You may be using them already, and they can also hook you up with the necessary services. So there is lots of different choices, and your decision is gonna basically be, made based on how much you're willing to invest in personnel, in training, and and other resources like that. Okay. So now that we understand a little bit more about the the the… Not the pitfalls, but the easy places to slip up, let's talk about where real improvements lie. Where can we find real improvements? You know, there are winners out there that have used AI, and what they're doing is really paying off. They're getting significant return on investment. Let's take a look at where that's happening. I'm gonna suggest it's probably not happening at the pari mutuel window. I have yet to see an AI that can call a horse race, but that's neither here nor there. Customer service. We see a lot of use cases in customer service. There are systems out there that can be your customer service. They can be answering the phone. When they do, they identify from the incoming number who the customer is and they call them by name. You have agents doing that with screen pops. Well, the AI can do it automatically. And it's been found that, not for nothing, but most of the problems people call in, they're pretty much the same. They're consistent amongst different customers. So 85% of queries from customers can be solved by an AI based customer service agent. That means that the customer service agents you have today can be doing proactive calling. They can be reaching out and driving more business for you. HR and employee service. There are many cases in which almost all, like, 94% is the number they came up with, of common HR questions are handled by an AI agent. The employee never needs to wait to hear from a person. They just get the answers they need, and you know that an informed employee is a happy employee. The IT service desk, you know, you have a large compute, estate running in your company, and you need to have people who can help people when they run into a problem. You need fewer people. You can have people building the new infrastructure rather than fixing the old when 90% of employee requests are handled by an AI agent. The finance back office. I'm gonna show you an example where a process that took four days was reduced down to eight seconds. And in terms of those of you who are using IT to run your operations and cloud to run your operations, you can see from 35 to 80% cost reduction in things like Kubernetes and other platforms that you use. Notice this about all of them. Each one of these is a high volume activity. But as I pointed out, most of them are pretty predictable. You usually get the same inquiries from most calls. I like to call them frequent flyers. And you… If you know what those are, you can answer them pretty easily. AI can certainly do that. AI can take that a lot further. And good news, you have plenty of years of data that you can feed into your AI to give it a good sense of the context that it's operating in. And with that, it can provide the kind of responsiveness that you are looking for for your user community. Here's a good example, and this is an example that I use to talk about how you need to be patient. You know, just because it's computer doesn't mean it's instantaneous. And and IBM took six years to create an overnight success. It's a platform they offer called Ask HR. And, of course, they sell it to companies that can run their HR business on them, their HR comp… Department on them. Six years. They now enjoy 94% containment. In other words, the the questions don't have to go further than the first contact. They get an answer right away. They're supporting 2,100,000 employee conversations every year. 2,110,000 calls be answered by a system. Think about how many people hours that might take. They've enjoyed a 40… Their customers have enjoyed a 40% lower HR operating cost over four years. So as they use it more and more, as it learns more and more about the company, and as it provides better and better service, they're seeing more and more of a savings, and that will continue to grow. That's got a long tail. What is most impressive to me is that amongst their customers, they've enjoyed 99% adoption amongst those customers' management. Adoption is the biggest problem we face. It is it is the element that defeats almost 75%, three quarters of projects of any kind that are launched in IT, they fail because nobody uses them. The user just refused to switch to the new platform. So that 99% number is just remarkable. And what made it work, first and foremost, they they they identified two tiers. They identified what the AI could definitely handle and got a clear definition of what had to be forwarded off to human operators so that the customer didn't have to end up frustrated and saying, can you transfer me to an agent, please? The AI kinda knew that before they asked, and that is very helpful in improving the customer experience. They wired it. They in… They integrated it with their PSA Workday, with their ERP, SAP SuccessFactors, Concur, and the agent completes about 80% of the tasks instead of just answering questions. It's amazing. They refined this over six years. I I have to reemphasize that. It took them six years to get where they wanted it to be, But most customers have turned those savings into hiring more engineers, more salespeople, driving more and better business for themselves. Of course, when you think about it, who else but IBM could make this claim? Most of the large language model model makers have only been in it since, like, '19… Since 2022. I know that IBM has been in it since at least 2015, 2016. They've got a much longer history. Watson was introduced back then, and it became Watson x. It's got over a decade of development and solution and learning underneath it, and they've learned that being patient is extremely helpful. They've also learned that in a world where people are talking about managed intelligence, integration is just as important as intelligence. It is all about the data. I will never deny that. I will say that every time. But the data is completely useless if it is not in a form that can be integrated with all the systems that are gonna be working with it. So integration is every bit as important as intelligence. I talked about customer service. In customer service, the big goal for us is to resolve the problem. And the more we deflect the problem, the more we forward it to some other department, or the more we tell the customer, oh, look it up on the website, god forbid, the more they're gonna get like this guy. We know of a software giant who we cannot identify, where now 85% of their customer queries to customer service are resolved without the involvement of a human. No forwarding it to a human. No handing it off. No referring it. No delaying it. No deflecting it. It's handled. 30,000 plus conversations a week, and they're enjoying an 85% agent accomplishment. That is fantastic. And the way they got to it was they started off narrow. They started off with a small set of things that the AI could do, and then they learned from what they didn't get right and learned and learned and improved it moving out from their core further and further out to the periphery of what they needed to know. And today, they have literally trained that system over time to give them 85% resolution over 30,000 plus calls a week. There is a city in Texas. Not a happy place to be for some people today, but, their 311 information line now enjoys, and this is amazing too, ninety percent first call resolution. It makes all the sense in the world. If somebody's calling in because they have a broken something, a broken photocopy or a broken computer, you know, a broken phone, it may be that that broken device needs a part. And so the customer service person's gonna have to say, okay. We're going to obtain the part and we'll call you to make an appointment to come back and fix it. You instantly have a less than satisfied customer. Their customer experience just went down the hill, a bit, maybe a little bit, maybe big. If you had the parts all in stock at one time, yeah, you could, you could answer all of them on the first call. That would be great. It's unlikely, and it's certainly impractical. But the fact is, in many cases, it's not something that's broken. It's just that something is not working properly in software or the user is making a mistake, and that can be resolved by an AI agent. And thanks to that, they are enjoying about a ninety percent first call resolution ratio. That's unheard of. And they're handling 12,000 plus resident requests, and I believe that's a month. I apologize for leaving that out. They have a very small team, but the scope of what they're supporting isn't huge. It's not like, you know, supporting major manufacturers. It is supporting a city, and there are very specific set of city requirements that they can support. And they make a clear service promise. The customer or the customer, the resident always knows what to expect. And, again, we see it. Launch narrow. Be able to satisfy a narrower set of problems with AI and then slowly but surely integrate more problems into the AI, but having people handle them in the meantime. It's gotta be a smooth, slow transition. If you do it that way, you will fail less often, you'll disappoint customers less often, and you will enjoy a much better customer satisfaction rating. The one thing, and I I… I'll say this personally, and I think you'll all agree, you need to keep a clear path to a person. It has to be easy to get to a person. I've encountered some automated phone answering systems that ask me a thousand questions. I've I've I've recently encountered online chats where when I ask a question, the first thing it does is ask me six questions, and I have to hit six buttons. And one of the buttons… None of the buttons may pertain to what I wanna do. I don't know what to do next. You don't wanna lose customers like that. You want customers to know that they can talk to a person if they ever feel like they're not getting what they need. We talked about routing and deflecting versus resolving. K? And the goal is to resolve. You wanna have a service desk that resolves as many problems as possible without routing it to anybody else. Yes. Your level one people are going to encounter problems that are too complex, and they're gonna have to move it upstairs to a tier two or tier three, support person. That's going to happen. But we have seen large companies accomplish seventy three percent first contact resolution, not the 90% we saw a moment ago, but on average, 73. We've seen a 63% faster time to resolution the more AI is handling more calls, more more requests. We have measured in some in some companies an average of about 21,000 manual hours saved every six months. So if I guess 42,000 a year, I don't know why we didn't say that. Imagine 42,000 times your average hourly salary to an employee, whatever it is. If it's $10, you've just saved $210,000, and I'm sure it's, hopefully, more than $10. And a lot of these companies are now achieving that 99.2% customer satisfaction. Remember, none of us create customer satisfaction. We can't. The customer does. And so what we need to do is obtain customer satisfaction. We have to earn customer satisfaction. And one of the best ways to do that is to resolve instead of routing or deflecting. What got automated first? Not the hard problems. Once again, they started with the easy stuff, triage, routing, and password resets, simple resolutions, the mundane middle. I… That somebody used that phrase, and I just logged on to it. The mundane middle of every ticket. If the agent can handle the mundane middle, fewer and fewer calls will get to your human resources and take up your human resources. So the more an agent can take on, the less a person has to take on, the more that person could do much more productive, much more proactive things. And, again, we're seeing an average now of three… 74% adoption. That's a very good number. It's a very good number, and it is the number to watch. The more your people adopt it, the more you know it's getting used. If it's getting used and you're not hearing complaints, you can start to expect to see better customer satisfaction scores as you move forward. So the advice here, automate that mundane middle first. Yeah. The hard problems will keep. People can keep on handling them as the AI gets taught how to handle them. And once you're confident that the AI can take on a set of problems, then and only then do you transition it. It's kinda like taking a tier one and making them into a tier two and then taking that tier two and making them into a tier three. We all wanna do that because it it it conserves the investments we've already made in personnel. And so this is a wonderful way to do that, not with a person, but with a system that we've invested a lot in. Okay. Here's that financial, example, and it's an employee, a saleswoman. This saleswoman does not look happy, but it isn't because a customer she's been after has just said no. That's not the case. She would be in their face overcoming objections and beating them to the ground to close that deal. But the finance department told her that there's some questions with her commissions. Now if you wanna know the easiest way to upset any salesperson, mess with their commissions, play with their commissions. They live and die by those, and they are very hawkish about tracking what their commissions are. When your calculations disagree with their calculations, they are unhappy people. It used to take about four days for most finance departments to come up with an answer when there was a problem with a sales commission, And the HR department had its hands full trying to deal with those disgruntled salespeople. Let me tell you something. I've managed hundreds of salespeople in my career. The one thing you do not want is… There's two things you do not want. You do not want a disgruntled salesperson, and you do not want a disgruntled spouse. Easy to remember. Now in the study that we've quoted here, it takes eight seconds. Solving the same problem takes eight seconds. There are security guardrails in place to make sure that the AI doesn't either cheat, get cheated, or make mistakes because you don't wanna make mistakes with money. But we have successfully been able to use AI agents to assess a salesperson's problem and solve it within eight seconds. You think that's gonna be a happier salesperson? And it wasn't that the process itself was sped up. This was a complete business process reengineering. They literally replaced the entire process. And the research, which came from MIT, by the way, found that the back office is the place that delivers the strongest returns to an IT department. There's less outsourcing. There are fewer handoffs. It's just faster to close out each service call. There's a lot of opportunity there. And if you're investing in resources for your support department, talk to your back office first so you can gauge and you can scope and you can size the force that you're bringing on. On the agent side, everybody remembers when cloud came first aboard. And with cloud, you can see the clouds on the screen there. You were able to say, oh, I need more storage. It gave you more storage from the shared resource pool. Or I need more memory to run this process. It gave you more memory. This is going too slow. I need more processing power. It gave you more processing power. You could throttle it up. You could throttle it down. Unfortunately, all too many people throttled it up, and then they finished what they were doing and walked away. So you had all these resources that were being paid for because they would… Had now been provisioned, but they weren't being used. You know, a lot of people call them zombie virtual machines, zombie resources, and they cost corporations insane amounts of money until providers realized that they needed to take on those tasks for them. And a lot of providers made themselves heroes to companies like yours when all they did was keep an eye on what was throttled up, what was throttled down, and what utilization looked like. Comparing the two, they moved those throttles to where they should be, so you're using what you pay for, and you're paying for only what you use. That's the problem I'm talking about here. We found environments in which, they were using Kubernetes for, for for for a lot of their orchestration. The average CPU utilization assigned to Kubernetes was only 8% utilized. 92% of it went unutilized, but it was provisioned and being paid for. They were using about 20% of the memory they requested. 80% was provisioned, but it was not being used, but they were paying for it. And ops was paying for GPU utilization, and with AI, that's huge. They were paying for 95 more than they needed at the time. These things have to be monitored. They have to be managed. You have to constantly be rightsizing them. A lot of companies say, well, I've got somebody who goes in and checks to see what the utilization is, and they do it once a quarter. You can be paying, through the nose, tons and tons of money from one quarter to the next. Maybe you do it once a month. It's still too long. AI can do that for you automatically without involving a person, and it can be doing it for you constantly. Imagine somebody sitting there constantly moving the throttles so you're only using what you pay for and only paying for what you use. Unbelievable. And we found that doing that reduces the number of crashes they run into. So you need monitoring, but you also need monitoring that will predict. Now that it's got all that data, it's seeing how much is being utilized of everything all the time. It can calculate and estimate what your future requirements are gonna be. So if the shared resource pool is insufficient, you can increase that. You can adjust that to fit what your people are going to be needed going forward. That's insanely powerful. So, you know, the the catchphrase to remember here is continuous beats periodic every time. Anything you're doing periodically, look at doing it with AI and letting the AI do it continuously. And, you know, as I said before, if you're rightsizing quarterly, you can't keep pace with with the changes that are going on. You know, I said every hour. I should have said every minute. Okay. So we've looked at some ways in which people are finding the problem that they need to solve, and by solving them, they're really winning. Here are first a few things that are not working that you should probably try to avoid. First and foremost, there's a a a, credit company. You can buy home appliances through Klarna. You've probably seen it in some of the stores you shop at as a consumer. Klarna decided that a lot of what it does doesn't need people to do it, and they could do it with AI. And so the CEO required that they fire about 700 customer service agents and about 1,200 other people, and they brought their headcount down from 5,000 to an an impressively lower 3,800. The very next thing that happened was the phone started ringing off the hook. Customers were complaining bitterly. Nothing nothing was going right, and it was because people had a better touch and feel for the customers. They had a better personal relationship with the customers. As much as they could train the AI, they couldn't train it fast enough to understand the customers. And so what did they do? They bit the bullet, and they went back and they hired back all those people and put those people back into place. And all of a sudden, their customer satisfaction rates are going back up to where they ought to be. The, Commonwealth Bank of Australia, cut out 45 call center agents. They use a a voice bot instead. And the number of calls being answered promptly fell by 2,000 a week. The union for those agents showed the the bank, look what you did. You know, you have fewer and fewer customers who are getting answered, much less satisfied. And you're using your team leaders, your managers, and you're paying a lot of time and a half overtime, sometimes double time overtime, to fill the gap that you've created. Don't do that. And the bank, to its credit, apologized, hired those people back, and the numbers turned around immediately. These people measured the outcomes. They measured the outcomes that they were getting, and they realized that cutting headcount reduced the quality of their outcomes. So I recommend don't look at headcount first. Look at the outcomes first. Work on the idea of how much are we going to hurt? How much are we going to lose? How much are we going to damage if we decide to reduce the number of people and increase the number of AI agents we're using? I had my own example of this. Let me really quickly because I see we're we're we're we're getting along now. I use Claude to do a lot of the work that I do. And And it happens that one of the processes I developed in Claude code, I traditionally use Sonnet 5.54. Back… I started with 4.6. Love Sonnet. Did the job great. And then somebody mentioned to me that, you know, you're paying… You're burning more tokens with Sonnet than you would, and they just introduced Haiku 5.5. Why don't your, why don't you have your people… Why why don't you go ahead and move from Sonnet down to Haiku and see if you're burning fewer fewer tokens? Sounded good to me. It's easy to switch. So the next time I ran the process, it was switched to Haiku. It could not get through the process. First, it crashed, and I told you what to do to resolve that crash. Then it did several things just wrong, and the results were completely garbled. And I explained what good production looks like. We went through several iterations. I eventually put out an inferior product. It's the one thing I don't ever wanna do. And so you can imagine I stopped using Haiku 5.5 immediately. I don't worry about the tokens as much as I worry about will my end readers appreciate the product that I'm delivering to them. And believe me, they will notice the deterioration in quality long before you do, long before a dashboard says, uh-oh. Danger. Danger, Will Robinson. Believe me. Your customers will get there first. There are hidden taxes when you try to do that. Now you notice those hidden taxes are peeking out from the… Every corner here. One of them is work slop. 40% of desk workers are getting low quality AI output, and that's costing them. Every time they go to do something, they're losing about two hours of time and about a $186 per worker per month. How many workers do you have? Multiply by a 186. That's the beginning of your loss. So just improving your systems to the point where your people can produce quality product, it may be as simple as switching from Haiku 5.5 up to Sonnet 5.5 or Opus 5.5, moving up to better models of Claude. It may be moving off Claude to Chat GPT or to Gemini or to Grok. There's plenty of choices out there, and there's no reason you can't experiment. There's no reason you can't pilot. You always have choices, and you should use that voraciously. I found this statistic, and I found it to be very, very conservative. They're saying that 40 of AgenTic AI projects will be canceled before the end of next year. I think it was Gartner that originally said that that was more like 95% or 97%. I think that's way too high. I think this is way too low, but I do think it's north of 50%. More than half of AI agent product… Projects will just be abandoned, because people rushed. They went in for the wrong reasons. They didn't go through the right sequence of events to build an AI solution, and there are great methodologies available to you. So you need to be prepared to do it do it right. There was a story in the news, I don't know if you caught it, back in April. Uber, one of the most AI companies we can think of, uses AI to to coordinate millions of ride shares all the time. They decided to license an AI LLM, and they figured on their budget for a year, the year being this year, 2026. Unfortunately, their employees burned through that by April, by the fourth month of the year. And now what are they gonna do? I'm not gonna get into how that rolled out. The story was in the news. It was all over the news, and you can certainly Google it and find it. But it highlights a real challenge. You've gotta be realistic in your budgeting. You've gotta start looking for where the hidden costs are, and you have to put guardrails in place so people know what they can and cannot do. That's called governance. And if you don't start every AI project by developing the policies that are gonna run it, the governance that's gonna control it, you're making a mistake. You're only hurting yourself. Agent washing. We work with many, many vendors. I've named a bunch of them. We… I didn't mention, like, Microsoft as an example. There's, Lovable, Cursor, Replit. That's a whole bunch of them, and, they have to unmention. But a lot of them will refer to something as an AI agent solution when it really isn't. It's a glorified chatbot solution, or it's a wrapper around a chatbot solution. It's not gonna give you what you think an agent can give you. And that's approximately a 130 out of every thousand that are being agent washed. That's not bad, but it needs to be better. There are some platforms that are delivering success, and these are the platforms you really need to start to look to. You know, you need to start to recognize hype from real claims. You need to start to learn who you can trust. So here are four players that are really delivering results. Workflow platforms. Now these examples here are ServiceNow, Salesforce, AgentForce. You wouldn't have thought of them when you thought of AI because they were there before AI. Well, now they've integrated AI. They've used AI to enhance what they can do, and they're sitting there on the system of record that you you use. And so they're acting on the one reliable pure source of truth. So there… That's one great example. Domain orchestration. And I guess IBM is probably… Could be considered the noted master of orchestration, again, because they've been doing it longer than anybody else. But they recognized the importance of integration a long, long time ago. So you won't find IBM selling systems that are just AI, that are just a chatbot, for example, or just a Copilot, for example. They integrate the HR department, finance, production, manufacturing, shipping and receiving. They try to bring together as many systems as possible to share that one common source of truth and to be able to interact with each other very, very efficiently. And so those are delivering tremendous results. Again, it's as much the integration as it is the intelligence. Agent builders. The best agent you're gonna use is gonna be an agent you built yourself. And I tell people all the time, do not think you cannot build an agent. You can learn to build an agent in fairly short order because it is natural language programming. That's not just a catchphrase. It is natural language. You're sitting there and you're telling it, okay. I want the thing you're making for me to be able to do this, this, and this. Now when it does this, I don't want it to do that. I want it to do this instead. I don't want you to do that. You can just go on and go on about what you want it to do, what plot code or codex or anti gravity or cursor or replit or what have you, what they will do is translate that into the code the computer needs to do what you told it you needed to do. You've done no programming. You've done no coding per se. I think Claude code is hardly named because too many people think it's code. It generates code. But even the guy who wrote it, Pars Churney at Anthropic, he no longer codes. That was his life for a long, long time. Today, he doesn't code. He uses his invention, ClubCode, to do all of his coding for him. He just tells it what he wants it to do. Powerful. You can get fantastic results building directly onto the frontier models. This is where I think there's more human learning needed to understand. For example, everybody thinks that it's all about anthropic. It's all… I'm sorry. It's all about Plaud or ChatGPT or Gemini or Grok or any other ones I've mentioned. It is equally about their harness. Many of you probably don't even know what a harness is. You interface with the harness. When you're sitting there in front of the chat, you're connected to the harness, and the harness is connected to the to the large language model. Claude has a harness. ChatGPT has a harness. So do all the others. Copilot… Microsoft Copilot is a harness. It connects to many different large language models, including seven of its own now. But in introduction, they needed to use Chegg GPT as their back end. Cloud Code is a harness. Codex is a harness. Start to learn more about what harnesses are available to you and what they can do for you and how they can help you, and decide if you wanna learn that. And if you do wanna learn that, you should absolutely pick the one that works best with you, the one that you're most comfortable with. Take the time to really research it, and build some things yourselves. Again, as I recommended before, start small. Go to the mundane middle, the things that are pretty easy to take care of, and you will find that over time, you'll become better and better at figuring out. You know who can help you? The harnesses. You ask them, how do I do this? And they'll go into the large language model, get the information they need, and deliver it back to you. In many cases, they'll do some of the stuff for you. And so that's… It… It's all about telling what you want, not how to do it. It's not about what it is. It's about what it does. Now, of course, this all means that you need to look at the market for platforms, and there's tons of logos here. And you're getting information coming at you at tremendous speed. And I strongly recommend to everybody that the first thing you do is identify the people you know you can trust. Not that easy to do, but gauge them also by not who they are, but what they do. You know? I mean, I just used an example. There are a lot of people out there who are commenting about AI. There are thousands of publications, mine included, and there's a lot of us writing about AI. I honestly only read a handful of them. One is Boris Cherny, who I just mentioned. He wrote Claude Code. That is a… That's a reliable source. Andrej Karpathy is probably one of the most popular sources. He was one of the original founders of OpenAI. He's worked at OpenAI as a director. He's now working for Anthropic. Most of the experts I know consider him to be their expert, the expert's expert. You will find lots of those, and I will make no bones about it. I I write an… I write a substack that is really focused on managed service providers called the agentic MSP. You can look at it, agenticmsp.news. Nothing stops you, and there's lots of information there about resources. So if that helps you in that way, that would be great. Remember, as you're looking at these things, that adoption isn't outcome. I said that adoption is very important. Outcome is more important. One does not equal the other. When you go looking at all these platforms, the first thing they'll be bragging about is how many people are on their platform, how many people are using their platform. And, you know, it sounds great. Ask them to prove it. Ask them to show you the case studies of the use cases these customers have created. I recently went on a hunt for that… Those use cases, and I was able to find very few. It may be that nobody's talking about it yet, but I'm sorry. I don't think that all those people who signed up are actively using. So you're getting an artificial number. Look for what's been accomplished with the platform rather than how many people have bought it or even what it can do. It's what it has done that's that's most important. And, again, data is the foundation of everything. If you don't have your data, you don't have it clean, you don't have it organized and structured properly, you're not gonna be able to build anything on it because it is the foundation. However, you're gonna get more and more bang for your buck when you integrate it properly. Now one definition of that is who's gonna actually do it for you? And you can certainly hire or train up people of your own, or you can certainly, find companies that are in that business. They're in the AI deployment business. Amazon Web Services just put themselves in the AI deployment business. They're spending a billion dollars, on putting their employees on your sites. I will tell you that Anthropic did the same thing. OpenAI has done the same thing. Microsoft invested $2,500,000,000 and 6,000 engineers into it. Even though they claim that their great strength is their partner community. They're putting their own people out there. So they're voting that you need our people to do it for you. You can decide that for yourself. Clearly, they realize that you're buying a license means nothing. If you renew that license, that's when it begins to mean something. It's because you got something from it. So ask every vendor when you interview them, who runs it? What are they running it for? Who's actually using it in production? How did they measure? What were the metrics they used to determine whether or not they were getting a return on their investment? Be very particular. Be very specific. Be very demanding, and don't just buy the claims. Finally, and I'm getting real close, Keith, what the 10 Keith Ward Yeah. Yeah. Yep. Howard M. Cohen Start with the process. Pick the process first, then find the products for it. Redesign the entire thing. Redesign, reengineer the entire process. Govern before you scale. Make sure that you've got it working right before you make it big because then you're just repeating the same mistake over and over again. Turn to your line managers who… To determine whether or not it's performing because they're the only ones who are gonna know if you're getting the right results, And bring your people along. They have great insight into what you're trying to build. So I'm gonna skip the ninety day plan. Basically, you kinda have it already, I think, but it will be in the slides, which I believe will be made available to you. But here's the question I'd like you all to answer as you leave. If you're using AI today, which processes are you actually depending on? Which ones, if they broke, you wouldn't be able to do anything else? You may wanna redesign that process. You may wanna make it more bulletproof. And, of course, then you'll learn a lot more about what other processes will go the same way. And with that… Hey, Keith. Why don't you come back on, and let's do questions and answers. Keith Ward Hey, Howard. Unfortunately, we have no time for questions and answers today. Howard M. Cohen I'm sorry. Keith Ward That's not a problem ever when you're presenting, Howard. You have so much great information. Sometimes what you've gotta do is get the information out there. That is the priority as it certainly was today. Just a fabulous presentation. I wanna remind everyone before we let you go, Howard. I did put your… The, the website up there, agenticmsp.news, and I imagine people can contact you via that if they want as well. Folks, I have looked at that site. I've seen it. It is it is amazing and chock full of great AI information of the type Howard was talking about today and a lot more. So as always, Howard, it is such a pleasure to have you on. I learn a ton, and I'm sure the audience does as well. I wanna thank you for being with us on the expert cast today. Howard M. Cohen Thank you, Keith. It was a pleasure. Keith Ward My pleasure. Alright. Thanks. So there we go, folks, and we're gonna say goodbye to Howard. I told you he was absolutely amazing, and there was so much good stuff in there. I was writing lots of notes. I hope all of you were too. Now, we do not give out the slides, but you will be receiving a follow-up link to this presentation. So if you wanna hear it again and you wanna hear any specific parts, you will have that opportunity. So no fears there. Now before we wrap today, I still need to give you… Give away the $250 Best Buy gift card, but I'm not gonna do that until you all help me along by doing… You know what to do if you've been on events, with me before. There you go. Let's keep seeing those emojis coming up. Alright. Nice to see. Keep them coming. There we go. There we go. I think we've reached critical mass to allow me to give away the card now. And the winner of the $250 Best Buy gift card is Bethany Garretson from Michigan. Congratulations to you, Bethany, and sit tight because we are gonna be in touch ASAP to get you your card. And with that, folks, on behalf of the entire actual media team, including Deanna, who has been incredible on today. I wanna thank Automox for sponsoring this event. Remember, without them, there is no event today, and that would have been a great disappointment to me. And I imagine to you with the, just phenomenal information that Howard, was able to drop on us today. Thanks as always for attending it for your great questions. And remember, get in contact with Howard if you have more questions. He is great at this stuff. He lives and breathes it, and, we were just lucky to have him on. Folks, that's all for me now. Have a great weekend. I will see you in the next one.

Please provide citations & attributions for any information you generate from this content, thank you.