Ep 128: Owning Your AI and Capitalizing on Proprietary Data with Andrew Brooks

[00:00:09] Hey everyone, it's Bob. Welcome back to Elevate Your AIQ, your go-to source for insightful conversations on human-centric AI readiness, talent transformation, responsible innovation, and the future of work. Today, I'm excited to share my conversation with Andrew Brooks. He is the founder of Contextual, which is an AI solutions platform company built around the idea that businesses should build and control their own AI capabilities rather than lock themselves into a single LLM provider. Andrew's path here runs through decades of entrepreneurship, including building and building and building.

[00:00:39] Andrew Brooks, building and selling a smart home company called SmartThings to Samsung. And he brings a grounded hard-won perspective on where AI generally changes how businesses operate versus where the hype outpaces reality. We get into how smaller and mid-sized companies can compete using AI, how the skills employers are looking for are shifting, and why staying curious and hands-on with these tools matters more than ever. Stick around for a conversation full of real-world lessons and stories and a healthy dose of optimism about where AI is.

[00:01:09] Where things are headed. Thanks, as always, for listening. Hey, everyone. Welcome back to another episode of Elevate Your AIQ. I am your host, Bob Pulver, and with me today, I'm looking forward to my conversation with Andrew Brooks. How are you doing today, Andrew? I'm doing great. Thrilled to be here. Excellent. Yeah, likewise. Looking forward to this conversation. You have been doing a lot of cool things over the course of your career, and I thought you could just give my listeners a whirlwind tour of, you know, how you got into consulting, technology, and entrepreneurship.

[00:01:39] So I'll let you kind of give some background. Yeah. I jumped right out of university in the late 90s, just kind of as the thing called the internet was heating up pretty broadly and ended up joining what was then Anderson Consulting, which is now Accenture, largely because I thought it was going to be really interesting to just get a kind of a broad-based business process, technology, change management view of technical innovation within companies.

[00:02:06] And so that was great. It fit me well. I got to work on small projects, big projects, but everybody also was leaving and starting companies, enjoying startups and getting hilariously rich on the stock market. And, of course, you get a little bit of FOMO when you're in that situation.

[00:02:22] And so one of my partners at the time had left to start what would be considered a managed service provider, basically providing operations management, monitoring management for cloud or data center-based infrastructure. And he and I built a good relationship, and it seemed like an opportunity to leave the traditional consulting world and dip my finger into the startup world. And that was actually when I really kind of went down a more deep technical path.

[00:02:52] I helped instrument systems for how we monitored them, how we managed them. I had to learn all different layers of the technology stack. The company's name was 7Space, which was the seven layers of the OSI stack. And then I, in that time, I started selling. I was a sales engineer. I ran a sales organization and so got a pretty broad exposure to what it meant to be in a startup environment.

[00:03:17] We sold that company to Sun Microsystems, which is now part of Oracle, back in 2005. And that was my, at that point, I was kind of hooked being an entrepreneur. So I had met a long-now friend of mine through a networking experience, and I was leaving Sun. And I said, hey, are you doing anything? Are you thinking about finding some companies? And, you know, it's kind of literally I hand-wrote him a letter saying, if you're doing stuff, I want to do stuff.

[00:03:45] And that kicked it off. And so can obviously do the 10,000-foot quick example, but have since then built a small business marketing company. We sold that to Reach Local, which at the time was Google's largest ad reseller. We actually took that public as a company, so got to experience that. Left Reach Local, founded a company called SmartThings, which is a consumer Internet of Things company, which we sold to Samsung.

[00:04:12] If you've got Samsung TV, you've got SmartThings or fridge or washer. And then after Samsung took a little time off, did some ultra-endurance races, kind of recharged the batteries, doing that stuff. Started to just do some technology consulting to see what was going on in the world. And then AI became a thing. And so I leaned in on that. And that's the company I'm working on right now, which is a company called Contextual, which helps deliver AI solutions for businesses.

[00:04:42] So it's been a journey across some different industries, consumer, small business, technology. But, you know, have gotten some luck, have had some good outcomes, and it's been a real pleasure to be part of it. Nice, nice. Yeah, that's quite a whirlwind story there. I think we'll come back to a couple of those prior experiences just as we dig into this, because I think there's some threads that I want to pull on or thoughts I want to connect, especially on the smart things.

[00:05:11] And just how we're trying to orchestrate a lot of different platforms and tools and things like that in the agentic world. But tell me a little bit about Contextual and the sort of framework that you've built to allow. It seems like it's a pretty robust toolkit that lets people build a lot of agentic capabilities, you know, agent-specific things.

[00:05:36] And then how you, I know you guys also, you know, think a lot about, you know, being sort of responsible by design and really thinking about, you know, the role, or I should say the sort of partnership and synergy between, you know, human workers and operators and the technology itself. Which is a delicate balance that people are trying to play right now. It can be a heated balance. It can be an emotional balance, for sure.

[00:06:03] So the underlying belief that we had when we launched Contextual a number of years ago was that AI was going to be transformative in two different ways. Number one, of course, AI-assisted development was going to make the speed with which software could be created, you know, that much faster. And obviously that's been borne out time and time and again with, you know, some of the vibe coding trends, but just the fact that LLMs are quite good at reproducing code.

[00:06:32] So we knew solutions were going to be, therefore, less expensive to build. But also AI itself in an agentic capacity or even just in a direct, you know, a task capacity through a more deterministic system was a new form of electricity that you were able to build systems and capabilities that you were just not going to be able to do with, you know, a bunch of if-then statements in traditional software.

[00:06:59] And the combination from our perspective was the breakout from our perspective was going to be markets that historically would have never been able to afford or invest in purpose-built systems, purpose-built solutions, either largely because they were going to be too expensive or they would be too complex to build.

[00:07:21] And probably both were suddenly going to be able to make that investment, that especially lower mid-market and mid-market businesses would have a new right to leverage AI as part of their own moat deepening. That before, you know, they had their ERP and their CRM and, you know, maybe tech was a, you know, a tooling, but it wasn't a differentiator. All of a sudden, these businesses were going to have the right and the opportunity to create these purpose-built solutions.

[00:07:48] The reason we built our own platform, our own orchestration layer is that we also believed it was very important for those businesses to, the phrase we use is, own your AI. And what we mean by that is do not become too embedded in a single provider or a single model because you need to be able to be reactive to what's happening in the space. You know, Gemini 2.5 Flash is going away and replaced with 3.5, which is three times as expensive.

[00:08:18] Okay, is that something you're willing to tolerate or do you need a layer that provides, you know, a framework above which AI becomes a pluggable tool and your system itself remains yours? And so, as a result, we're a company that does basically design, build, operate. We work with our clients to understand what is AI good at and how can it be applied to your business?

[00:08:42] We then, you know, build those systems on the platform and then host, operate, and manage those inclusive of ongoing maintenance and tuning and updates for our customers. Because these are living systems. You can react real time as capabilities evolve in the market.

[00:08:58] So, you can basically, you know, I'm a small business, say, you know, I don't know if you measure that by employees or revenue, but let's just say I have a thousand person, you know, company doing, I don't know, say 15 million a year or something like that. That's probably a sweet spot there. So, they're doing well.

[00:09:20] They've got a pretty good trajectory, but they want to future-proof, you know, what things look like, you know, three, five plus years down the road. They could basically come to you, to your point, not say, oh, we're going all in on Anthropic or we're going all in on, you know, Google or whatever.

[00:09:39] And so, they come and they have a platform where you basically can help them select the right model for the right task, workflow, et cetera, based on cost and other factors, I'm guessing. Yeah, and ultimately, model selection is just one of the pieces. And certainly, we're opinionated on models based on the task that we're seeking to deliver.

[00:10:05] However, they're looking for us also to help design what that system looks like. What is a new human interface that their employees might interact with this system through? You know, when a lot of people think of AI, they think of the chatbots. Well, that's just, that's one form of interacting with an AI system. Very often, the solutions we're creating are not, you know, don't have a chat style component because that's not, you know, how the agentic or goal-seeking behavior needs to manifest.

[00:10:33] It actually has to manifest in a headless way and in another system. And so, they're seeking for us to do a few things. Number one is to have reflections and hopefully experience the type of problems that AI is good at solving or opportunities that AI is good at adding value to. Within that, we really seek to work with our clients around what's your right to win? What's your moat-deepening opportunity?

[00:10:59] Do you have unique data that we can do something special with, with AI? Do you have a process or an intelligence? For instance, hey, one of our clients is a commercial refrigeration company. And they are really good at estimating what it's going to take to do a large commercial refrigeration installation. And a lot of that information is up in the brains of these people who have been doing it for 30, you know, 40 years because they know all the gotchas.

[00:11:26] And, you know, one of the stories I like sharing is this guy is like, hey, we never hang half-inch pipe on the ceiling because people hold onto pipes. You got to put three-quarter-inch pipe on the ceiling. Otherwise, a human is going to dangle from it and hurt it. And that's that, like, that's institutional knowledge that we can build moats around with AI. We actually call that project Digital Greg because the individual with the most knowledge is Greg. And, you know, we're getting him into an agentic system.

[00:11:54] And then lastly, it would be customer relationships. How do you interact with those customers? So we look for data or process or intelligence or relationship moats that we can deepen with AI. But then we're also trying to go on a journey with those customers. Our solutions might start with one use case. But generally, we're having conversations from day one around what's the series of transformations, both efficiency, operations, finance, customer-facing, new products.

[00:12:23] What might that journey look like from an AI standpoint? And again, our belief is by owning your data layer and your business logic layer and how you interact with the AIs and, of course, your prompts and your evals and all the things that make a system work, customers can deepen that mode. And suddenly it is a defensible IP asset.

[00:12:45] Do your customers and prospects, do they know that going in or is this an eye-opening exercise when you get there? Because I feel like a lot of the conversations I'm in, I would say 80% of them, they really do go in with this sort of automation, you know, cost savings, cost avoidance kind of mindset as opposed to, I think you would categorize it as, well, like you said, deepening their moat.

[00:13:14] But what, you know, the capacity building, the effectiveness, the innovation that they can foster amongst that, you know, combining their own know-how, that, you know, tacit knowledge, and where do they really have quite an advantage that you can elaborate on? I think it's consistent with what you just described. Oftentimes we're being engaged in an automation or an efficiency project.

[00:13:38] And that's because it's relatively easy for somebody to look at an organization and say, hey, we have 13 people doing this task. That seems like an opportunity that's ripe for AI-based innovation, especially if it's a, you know, human intelligence task. There is some intelligence being applied to it. And to be clear, we love grabbing onto one of those. It's a great mechanism for building belief.

[00:14:02] It's a great mechanism for conditioning an organizational muscle around AI change management. It's a great opportunity to create advocates within the organization that, yes, this works and it works well.

[00:14:13] And it's also a great opportunity to educate the organization that when we think about the spectrum of AI tooling, certainly you have on one side, you've got your traditional, what we would consider kind of AI assistants, your, you know, your cloud co-work, your chat GPT work, things that are enabling to the individual versus a full AI business system, which might be multiple agents working in concert and doing different things and multiple tools.

[00:14:40] And therefore something that looks a lot more enterprise grade as a capability. Helping people have expand their understanding of what's possible is often a trigger to more creative brainstorming. And then you get a little bit into a situation where once people get a little addicted, they're like, hey, I'd spend two hours a week doing this and four hours a week doing this. Can AI do this all? And sometimes you have to say not everything's an AI problem.

[00:15:09] Some things are processed and some things are just workflow and AI can and cannot play a role depending on what that is. Yeah, I do see people falling into that trap. Like once they, once their eyes have been opened, it's just like AI is the, is the shiny hammer. Exactly. You know, what do you got? What problems you got? And can I throw AI at it?

[00:15:31] And sometimes that means you're leaning a little bit too far over the edge and, you know, maybe over engineering a solution that, you know, more traditional solutions could solve. Or maybe just straight rules-based, you know. Just, yeah, a traditional deterministic process automation. Yeah, that too. Yeah, exactly. Yeah, no, and I think at some level you have to, there's a really key ROI analysis that has to go into that. You have to have a sense of how you're going to value this. What's the value that you're going to derive?

[00:16:00] It doesn't always mean cost takeout. It might be unencumbered growth, you know, taking off the handbrake of growth. One of our clients processes tens of thousands of work orders and invoices via an email inbox each month. And the goal there was to not remove all of the people who are doing that task. It was to make them much more efficient so that they're doing, instead of 50 a day, they're doing 150 a day.

[00:16:24] Because then they can grow as an organization without actually adding to a team that historically would have been a labor-intense team. So from a platform perspective, I guess this is one of those threads I wanted to pull on looking back at your sort of IoT experience and some of the infrastructure kind of components that you worked on earlier in your career. Like, is this sort of an all-in-one solution?

[00:16:53] I'm not saying you do every single thing, just, you know, here you go, ready-made, everything in a box. Right. But just because I think one of the things that people don't necessarily think about, like you can, you might be learning how to build an agent and maybe you're building it, learning how to build one agent to talk to another agent.

[00:17:12] But this is a system that you're building and you need to think in terms of systems in every sense, meaning, you know, security, privacy, all those sort of traditional infrastructure layers. But now some of this agentic AI is almost another infrastructure layer. So are you encompassing a lot of those facets as well? Yeah, absolutely. And I think the corollary to smart things is a good one.

[00:17:42] With smart things, we believe there was going to be a proliferation of connected devices. It was, we were seeing it happen, you know, back in the day, you had some lights and you had some locks and you had some thermostats. But we said, you know, battery power going up, processing efficiency going up. The standards of connectivity were being kind of normalized. And we said, if there's going to be this explosion of smart or connected or connectable devices in the home, an organizing layer.

[00:18:10] And in that case, it would be organizing around how do you connect to them across these different standards? How do you interact with them across different mechanisms? What's the what's the what's the not everybody just wants to turn a light on and off with a with a button on their phone. And you might want your home to react to you. And, you know, one of my favorite examples back in the smart things days was a gentleman who sent us a video of his his whiskey collection in a cabinet. And as he approached it, the lights came up and music turned on.

[00:18:40] And it was this, you know, angelic reveal of his whiskey collection. And it's a it's just a fun example. But the point is that it was a system. That's that's how he wanted to interact with his with his whiskey experience. And the same is true for these agentic or A.I. centered solutions. And of course, the reason I'm saying agentic or A.I. is some things, you know, people would say, hey, an agent really has to be goal oriented. It has to have some autonomy in order to determine how to do its work.

[00:19:09] Some things are A.I. that is not agentic. It's just, hey, please extract the information from this document I'm about to send you. It's much more task based. But no matter what, the systems themselves are often their data hungry. And so you have to have a structured data layer. You're going to be integrating with a number of third party tools, be they A.I. tools or traditional systems. I got to pull this from Salesforce. I got to pull this from NetSuite.

[00:19:36] You're going to be thinking about how A.I. can interact with those third party systems in a safe way. For instance, we might say we're working with a client right now who's on some scheduling work for consulting projects that they deliver. And so they have their their resource availability in in open air, which is a NetSuite environment for tracking these sorts of things.

[00:20:01] Well, we want to use A.I. in a read only context to pull availability out and make a recommendation as to what the staffing strategy should be. But we actually might want a more human led booking activity. We don't want the A.I. to have kind of free form ability to book. And so when we think about systems, you're thinking about all of those things. Where's the data? What data is the client? What data might we enrich from third party sources? What systems do we have to integrate with? What's going to happen?

[00:20:31] Is this a does this have to be a real time synchronous system or can it operate in an asynchronous manner? What sort of checks do we want for the humans to have? How can we take feedback into the system in order to make the solution smarter over time? That's ultimately what a full solution needs to do. And that's why we built the platform the way we did to to encompass all of that. So I talk a lot about like being not just responsible use, but responsible design.

[00:20:59] And so I was curious from the standpoint where every everyone is a builder in theory, right? Like how you work in collaboration with the clients to make sure that they actually understand what you guys have been brought into to design, build and operate. Like how does the is it is that a collaborative effort or do they just like, you know, don't have a technical staff?

[00:21:28] I mean, hopefully a mid-market company has some type of their own IT staff. Unlike, you know, they're not mom and pops. Yeah, they're going to have a CIO for sure. But they likely don't. Our customers likely are not going to have a team of developers, right? They might have their Salesforce admin who knows how to write, you know, scripts and capabilities in Salesforce, but is not a classic independent or full stack developer.

[00:21:52] I think to answer the question, the alignment is very important, but it's less about necessarily architecture where our clients are expecting us to be subject matter experts and bring some the intelligence around how to how to build these systems. It's much more on what is this thing going to do? How are you going to interact with it? And what's the measure of success here? And how are we going to escalate when things go wrong?

[00:22:19] Because it's you're not going to get, you know, 100 percent accuracy in all in all systems. And so I kind of liken this to. AI assisted development, to your point, is so fast, but it's like firing a rifle. It's the bullets going to be down, you know, field so fast. If you've missed your direction a little bit, if your aim was over here versus over here, you're already down the field and you might have missed out on an opportunity to do something right. So we do a lot of work with clients.

[00:22:48] It's kind of a measure twice, cut once strategy of what is the requirements here? What does that PRD look like? It's easier than ever with AI to do clickable wireframes. It can jam out HTML, you know, pretty fast. Arguably, AI is a little bit color and font crazy when it does its own design. But you can, you know, you can you can wrestle it a little bit.

[00:23:10] We want clients to be interacting with those systems and tell us like, yes, I can see how I would be using this before we go too far down the path of building. Now, what what that what I'm going to now give a kind of counter to my argument, which is sometimes you don't know what the second third order benefits of the system are that you're going to build until you've built it, until you see the data flowing through it.

[00:23:34] One of our the client that we manage invoices and work orders, we were processing that to extract data around basically repair and maintenance work for forklifts in in warehouses. But it also became clear that that as a as a service offering, they wanted to look for cost savings. They manage these on behalf of their cost, their customers. They wanted to look for cost savings in labor rates or parts or contract compliance or maintenance window compliance.

[00:24:01] And now that the data is coming through an AI system, layering that sort of rule and intelligence on top becomes easy. And so that's another reason we believe pretty strongly in in an orchestration platform is you get these second and third order benefits from having the data flow through that you might not have even expected or certainly may not have been your the first reason that you were going down this path. Yeah, no, that makes total sense.

[00:24:27] And so one of the things I know you've talked about, I think we talked about last time was around customers perception of their own like maturity and readiness for AI, like where their data, like you said before, like, you know, it's was a Greg. Greg's the guy who has all the expertise and it's all in his head. So if Greg's sick, let alone leaves the company, you've got big problems. Exactly.

[00:24:54] But not necessarily where the data isn't documented, but it could just be messy. You just haven't focused on it because you didn't know, partly perhaps to your point, like you didn't know that certain data might be useful. Correct. Down the road, right? And so that's not necessarily a reason to not start. I think that's right. You know, certainly the reality of garbage in, garbage out is unchanged, right?

[00:25:21] If you give garbage to a system without the boundaries around maybe this is garbage, evaluate this, double check this. You know, you're not operating with kind of good diligence. But I think the if you accept, hey, we AI might not be able to match data point one in system A to data point two in system B directly. And if you could, you wouldn't really even need AI because you would just do some regex matching or something like that. But it can interpret.

[00:25:49] It can add a confidence to its conclusion. You know, one of our customers is a import, a trade services importer. And so they bring a lot of products in on behalf of their clients from China. And the number of manufacturers in China is an enormous number of individual entities, you know, ranging from small mom and pop shops on the corner to, you know, large conglomerates.

[00:26:14] And you might get data on an importer record that is difficult to match to what's in the data set of their core system. So you got to pull a bunch of them and give it into AI and say, help me figure out which of these it is. And AI can be good at that. And if you're designing the system right, then you use a step at that point to clean, to create, you know, you know, cached information about that particular organization in this case.

[00:26:42] And we see that as part of our job is to help customers get comfortable with, well, let's figure out where we're going to catch the bad data. Why is it inconsistent between systems potentially? And what can the system actually do to be a, you know, a cleaning vehicle, you know, through that? I would say one of the areas that I would agree with clients if they push back and they say, hey, we don't want to automate a bad process. And that makes sense. Don't automate a bad process.

[00:27:09] I'm not sure that you're going to make doing something faster and less expensively that's bad is not necessarily going to achieve the outcome you want. And so I see this as an opportunity for companies also to look at those processes and say, is there a different, we've always done it this way for whatever reason. You know, these are how the organization has evolved. You know, it's we merged two companies and this is the complexity that emerged. This is an opportunity to kind of re-reflect against that and think, is there a new way to do this? Could we flatten functions?

[00:27:40] Could we streamline capabilities? Do we need to do this step at all? And I think that's an important part of knowing you're ready. All that to say, the failure to act now is starting to become a little bit of a, you know, a failure to be a good steward of your organization because the learnings are so important. And, you know, you can't wait for the perfect model. The models are revving every 10.

[00:28:06] I read there's a new frontier model every, you know, 10 days or something like that, 15 days. So, you know, the pace of change is just too fast. You're not going to, it's not going to stabilize. So you need to get into the river. Yeah. I see a lot of people getting tripped up by this and I think they just, yeah, they just need a good sort of navigator through some of this because, you know, the train's not slowing down.

[00:28:31] So people are just like, like, it's like, which door do I jump on as this train is moving? And I know it's, it's challenging. You know, I don't mean to make light of it. This is a big decision, but it ultimately, I just don't know if people really have a choice. I think the winners will be AI at their core and at the right levels. They'll be human where that matters.

[00:28:55] If your brand is based on quality of human interaction with your customers, I wouldn't, you know, jam an AI down their throat. But you can certainly equip that person with an AI that remembers everything about that customer from all the different sources of information and puts it right at their fingertips so that they can be the, you know, the person that's like, hey, you know, how was Susie's birthday? Is she still playing soccer? And that's a silly example.

[00:29:22] But you can equip your team members with that kind of differentiated tooling. You know, some of the areas that I, we push back on a little bit is if customers come to us and say, hey, we want to do AI, you know, content generation, which is easy to do. AI is happy to spit out words. That's what they love to do. You know, we're very, you know, kind of opinionated around, is that a good thing? Is that helpful to you?

[00:29:47] Is that, you know, is putting a bunch of slop out on LinkedIn going to actually achieve your objective? Or is there another way to think about using AI? No, I think people do need to ask themselves those fundamental questions. And some of it goes back to should human, is this a task that should be retained by a human being who knows how to tell stories and who can write and communicate well? And is this something that requires empathy?

[00:30:17] Right. And so even I'm sure you have a lot of clients and sort of, you know, like you said before, like in trade, different trades, you know, vocational trades and things like that. It's like, look, we're not talking about actual, you know, robotics, you know, humanoid robotics showing up and fixing your HVAC.

[00:30:38] But you could be doing more stuff with AI on the front end with all that data on the back end for customer service and support, upselling on contract renewals, you know, accounts receivable. I mean, there's all kinds of domains, you know, within almost any business.

[00:30:55] And so certainly I don't want to say that AI is the solution for everything and that every business has to be, you know, go full bore into AI and start, you know, displacing human workers. Not at all. I mean, I think the future of work is humans plus AI and you just have to figure out where your balance point is across different domains and divisions of your company.

[00:31:21] It's one of the reasons that, you know, kind of plastered on our website in our brand language is giving your team superpowers, right? That's how we see this. We see this as a team enabling, a human enabling, a human, you know, activating toolkit and tool set. So that's the lens that we bring to it, which doesn't mean that, you know, it's always there are certainly some areas where full automation makes sense.

[00:31:47] And you would like to do it fully automated because maybe that human's time is better spent on other activities. You know, as if we can automate those cost savings analyses across all these different vectors and then just give a nice report, then the human can go into the conversation with their client saying, look at all the money we saved you. By the way, here's our recommendation for equipment choices that you're going to make with your new warehouse that happens to be in the desert southwest with, you know, high heat load. And therefore, you should have this battery base.

[00:32:16] Like it changes the role that you can play versus, oh, we, you know, we were chasing invoices and work orders and getting vendors paid. That can go away and the thoughtful collaboration can be the bigger piece. Yeah. I mean, I think this is another area where people are so heads down in their day to day and what they're doing that they don't even necessarily realize. First of all, that just because they did it this way before that they have to continue doing it that way forever.

[00:32:46] But also, they may not even realize what their people are capable of and that there's other things that they could be doing. There's always there's always more work to do. There's always projects that don't get funded. There's always things you'd love to support if only you had more bandwidth or more money or whatever. And so I think that that factors into this.

[00:33:05] And then I think people, even in your own job, people where you're fearful of AI, you don't want to train it to basically be your backfill and take over your job for you. But the more you know about it, the more I think you can work with it and figure out what else it could do. So to your point, Andrew, around giving humans superpowers, how do I augment what I'm already capable of?

[00:33:34] And what else could I be doing to prove that I can provide more value to the organization, to the team? If only I didn't have, you know, 40% of my day doing, you know, administrative things or redundant work or things consuming my time. I think that's where the cost savings, cost savings shouldn't be an end unto itself. It's like, what do I do with that savings?

[00:33:58] And how do I reinvest in the people and the company to get into that capacity building? Yeah, we, I'll give you an interesting example that I think kind of reinforces that one of our clients manages vacation rental properties on the Gulf Coast of Alabama. They've got about 3,500 units. And in any month, they're doing 12,000, 13,000 work orders out in those units, fixing things, repairing stuff.

[00:34:26] And this ranges from you need to change the batteries in the remote to you need to fix, replace the light bulb in the, you know, in the foyer to the fridges making a noise, the ice makers clanking, the toilets running, you name it. So there's a lot of work to keep those units up and happy. And they're rental units, so they get a little, you know, they get roughed up. That's just kind of the nature of rental units. Well, because these are owned rental units, meaning individuals own them and this company is providing a service to manage that process.

[00:34:56] When a work order happens, it's going to go back to the owner. They're going to see the information. And so historically, they had managers having to just look at all of these work orders to make sure that the information in the work order and in the ticket once completed was appropriate to get back to the owner. Do the pictures actually show that we fixed the toilet or is it pictures of people's feet? Do the, you know, is there or somebody's thumb?

[00:35:21] Does it actually, is the language we use to describe the work we did appropriate for delivering that to the human owner? And so the AI system is designed to review the tickets and check the photos and confirm whether or not this work appears to be consistent with expectations, consistent with quality standards that they've set to streamline that process so that humans don't have to be in each area. It flags when it's got a concern.

[00:35:50] So it brings attention to the issues that really must be dealt with. And a couple of things happened with that. Number one was, unexpectedly, the AI system started spitting out audits of the line items on the ticket itself where they would say, hey, you said you replaced the light bulb while you were there, even though you were there to do a toilet repair. And the light bulb isn't showing up on the list of supplies that you use because somebody just grabbed it from the supply closet, screwed it in and shipped it off.

[00:36:20] Well, if that's happening a couple thousand times a month and it's five bucks here and seven bucks there, these are suddenly important pieces of revenue that were lost. And so, OK, great, we were doing an efficiency thing. We actually got a revenue recovery benefit out of it secondarily. But now the thought goes to, well, if we're in there and we're taking pictures anyway, should we consider taking pictures of of areas where some renovation might help increase rental rates?

[00:36:49] Because we happen to know all these different what each unit rents for and whether or not it's renoed. And so you can easily see an AI solution that says take a picture of the living room. Here's your current living room. If we refresh the paint and the carpet, you know, and replace that those light fixtures, it's going to look like this. And we think you'll be able to get an additional seventy dollars a night, Mr. Owner or Mrs. Owner from making that investment. By the way, we are a service provider who will do that renovation work for you.

[00:37:17] And so, again, this this journey people can go on of we had an automation, but now we have data flowing through. What more can we do with that? What new business opportunities emerge? I think is really important to to not just get fixated on the cost takeout. It's the it's the long term value creation. I think that's such a great example of connecting, you know, additional value based on the connections that you've already made,

[00:37:45] because you could also say, well, this we know these units have these fixtures and these fixtures go through bulbs twice as fast as these other units or these dishwashers. This brand of dishwasher is likely to break. So anyone if you hear of anyone doing a reno, not to get. Don't do that one. Dishwashers, whatever. That's right. You end up with data that makes you an information partner that you might not have expected.

[00:38:15] And yeah, to your point, all the way to like, hey, these types of batteries last longer in these remotes. You know, who knows what it is? And obviously you can you can screw things down to potentially too nuanced in that same solution. When we first launched it, the rules said if you if you do replace something, you got to take a picture of the serial number of the asset that you replace it with. Well, there's not serial numbers on batteries that you're sticking into a remote.

[00:38:38] So, you know, there's you work within the boundaries and kind of loosen the system until it's doing exactly what you want. And that's that's part of getting started. Right. Is to learn, you know, where the levers are that you tune on these systems. Yes. I think it's another value point for for your customer themselves. Right.

[00:38:59] I mean, I just remember my brother bought a place out west so you could go skiing and he's got three kids who are now basically adults and they want to go skiing too or whatever. So he was looking at a service. He can't be out there all the time. He was looking at one of these services and like he was just kind of looking at a cost and getting a getting a feel for their trustworthiness and things like that.

[00:39:21] But if he knew that he could pay a little bit more and these people would be that much more active or looking out for all these little things that would cause him one less headache a month or whatever it is, those things matter. Absolutely. Yeah. Hey, we managed 50 other units that look like yours. We know this is going to be an issue for you. We're going to be out repairing three of them. Do you want us to do it proactively? You know, there's there's all sorts of again, because I can be so data voracious.

[00:39:49] You can put so much into it and look for patterns so much faster. I think it's a it's a it's it's transformative. Yeah. No, I think that's a great, great example. One thing I want to ask you about was about six months ago, you guys got acquired, right? By a private equity company. We announced it in January, but the actual transaction was completed last September. So we're almost to a full year since that transaction will be a full year and here in September.

[00:40:15] And so, yeah, that's an interesting shift for me professionally. Having done VC before and having sold to strategic acquirers before selling to a PE firm was a was a slight shift. In this case, our sponsor is Southfield Capital. They're a Greenwich, Connecticut based lower mid market PE firm.

[00:40:37] And they they specialize in what they would consider essential business services for, you know, you got to buy them no matter what physical security, managing your forklifts, you know, things like that.

[00:40:47] And we had an aligned belief with their leadership team that AI based transformation for businesses was an essential business service that ultimately all businesses would would need to embrace AI to be to be, you know, to defend their their modes and their differentiation in the end of the future. And so it was just a very well aligned and consistent vision of where the industry was heading.

[00:41:15] And I also believed it was going to take time, especially with our target audience, which is lower mid market. And so, you know, this wasn't a VC up into the right SaaS experience. We're a services organization. We're doing the hard work on the ground, you know, to solve real problems. We believe that's where the value is in this ecosystem. But it's you know, it takes time to build that. Yeah, I mean, one of the things that was intriguing to me is, you know, I have I have a friend who works for a PE firm.

[00:41:44] I have another who's a CFO of a portfolio company of another PE firm. And we've been talking about this for a couple of years now, how this could be. Well, first, it was when I was really focused on responsible AI. I was thinking, well, whether you're a VC or PE firm, if you're making these kinds of investments, it should be part of your due diligence. Are these people being responsible by design?

[00:42:12] Do I know that I'm mitigating some of the risk as things evolve from whether that's a legislative standpoint, regulatory standpoint, from a financial risk or even reputational risk standpoint? You know, how do I have to look at this? But PE particularly, obviously looking more concerned about, you know, long term, you know, investments. And it just seems like one of these like sort of shared, like what an incredibly valuable sort of shared service if you could provide this.

[00:42:42] Essentially, to your portfolio and how they operate. Now, hopefully, if you'd taken those prior steps, those folks have already, you know, adopted AI and they're on that journey. But that doesn't mean that they've figured it all out because I can tell you from firsthand experience, when you're trying to build systems like this, if you don't have a good partner and you don't have anybody showing you how to put all the pieces together, it's a mess. Yeah.

[00:43:11] I mean, there are so many different tools. I just installed, I'm in this sort of boot camp, this summer camp to build some agentic systems for myself and for small businesses. And I thought I had a whole bunch of AI tools already. I think I installed like another six. Yeah. Just for every little thing. I got to store my credentials here. I've got to push this. I need this to push to production. I need to have this GitHub repos.

[00:43:38] I have to have like all these things that I thought originally I was like, oh, I'll just I want to really embed myself in Claude. And I have Google Workspace. So I'll use Gemini. And then maybe I need one or two other things. No, apparently I have like a dozen other things that I have to install. So it's crazy. Real systems tend to get complex.

[00:43:58] We would see some of that as agent sprawl potentially within an organization, which we think is a justifiable argument for why you invest in a partner who has a platform like we do. I think a couple of reflections, just the PE and AI adoption reality today. Certainly a couple of years ago, the motion was your companies better have an AI roadmap. And we're seeking to understand what that is.

[00:44:28] I would say that's certainly now progressed to, well, your companies better be implementing. They better be able to show that they know how to do this. Because we've seen a shift of PE firms talking about EBITDA only, which is the cost savings angle to, you know, alpha, which is, you know, differentiated value levers and creation. And that can include new businesses, new capabilities. So that's a shift.

[00:44:53] And I think they're experiencing some of the same things you experienced in trying to do some of the agentic stuff for small businesses. I'll get a list of tools that a company has, and they have 50 tools that have an AI capability of some form or shape. And they're asking, well, which of these are good? And of course, we might have opinions, but we don't implement, you know, we don't implement third party tools. We build solutions. And so, you know, we also have an opinion on that.

[00:45:22] But I think if you end up implementing 50 different tools on 50 different select data sets, you miss the opportunity. And to your point, I think you increase your risk from a governance standpoint that something is going to go wrong. Something's going to change data somewhere that trickles down and now you don't have traceability. How did it, you know, how did it happen? When did it happen? Those are important bits to the story. Yeah, absolutely.

[00:45:48] I think it also ties to some of the things I know you've spoken about, just, you know, the human involvement. I've been super critical of late when people talk about human in the loop because I don't think it's nearly specific enough. What is this loop that we're talking about? But just, you know, being in the driver's seat, right? Being the captain of this ship and making sure that whatever you pull together, you do have that oversight.

[00:46:13] And yeah, I mean, if you're a CIO, CTO, I'm sure you want fewer, you know, throats to choke and fewer failure points. But you've got to think about how you're architecting that whole solution. And then, of course, you know, as people get more involved, then you need to know that people are, you know, educated and discerning and are using their judgment and critical thinking to make sure that this is all flowing as everyone expects.

[00:46:44] Yeah. I mean, everybody knows, you know, it's, it's, I was reading an article today that was like, it's cute that we call AI mistakes, hallucinations, because it makes it softer, but they're errors. They're not, you know, they're, they're, they're problems, especially in system design. And so for us, it's really around what are we trying, what is the human's job in, in, in this role?

[00:47:08] Is it that we're trying to take away 80% of their job and give them the final product that they can then scan and say, this is good. And I, and I accept this and I moved it forward. Is, is it that through interaction with the AI's results, we can create new rules or new golden examples or new, you know, intelligence that can go in in context learning to make a system better and getting smarter over time?

[00:47:33] Is it that we, you know, we don't expect, we expect to actually just have an asset for the human that makes them better on a call and makes them smarter? You know, there's lots of different ways to think about what the human's role is in, in interacting with the AI output. And I think that comes down to objective setting, right?

[00:47:52] If, if the, if the goal is automation, then initially what we're going to be doing is using humans to validate work, to, to, to make changes that we can trace back and improve the system itself so that we can, you know, get to, to full automation.

[00:48:10] If the, if the goal is facilitation, then we might be producing information in an entirely different way that is, you know, at the right time at their fingertips to, to make, to make decisions, including with digital Greg, right? You know, the goal is make sure we don't miss the obvious or the things that make estimating commercial refrigeration nuanced and special and unique. Because if we mess up a little bit, it could be hundreds of thousands of dollars of margin.

[00:48:38] And so it's, it's to expand the, the corpus of knowledge that gets supplied and flag things that, that might be really economically damaging. And so I think what, what, what's the AI's job? What's the human's job is an important part. And to your, to your point, it's not just, oh, I'm in the loop. I'm not, we don't want to train a bunch of people to be checkers of AI work. We want to say, what is the facilitation that the AI as, as your teammate is going to be doing? And how does that manifest? I guess a little bit on the technical side.

[00:49:07] How do you see jobs changing? Like more, I think every, every job is changing to some degree, but like, I guess when we think about engineering, I feel like everyone's using engineering as a suffix for like every job these days for, for, even for non-technical people. So, so what do you, how do you see this evolving in terms of the technical talent that you may have?

[00:49:36] I mean, maybe just think about the talent that you, if you were going to hire someone to be an engineer today, like what, how would you describe their, their day-to-day work? Like how much of it is really actually doing some of the things that they were doing, you know, a year or two ago versus how you think about it now? Yeah. No, it's, it's, it's, it's something that we, we think about and plan around quite a bit within our organization.

[00:50:02] We are of course very AI forward in our create, in our system creation. We, you know, we're, we're using codex or cloud code or whatever against our MCP server that is trained up on how to build things on our platform, ranging from primitives, how do you manage connections and AI routes all the way up through, you know, full business solutions. What is complex invoice processing out of an inbox look like? And so within that, we seek now people who are strong in system design.

[00:50:32] They can think about the entire system, right? They're not just pulling a Jira ticket and saying, oh, my job was to add a new, you know, X to this one chunk of code. We're seeking system designers because you're fanning out agents who are building the system and you're observing that. And so that's a, that's a, I think a, a, a, a, a, a, a, a more of an architecture, but how the business system works.

[00:50:58] We're seeking people who can predict upfront the unexpected. Oh, you've told me that you handle single page invoices. Well, you don't, there's, you're going to get a hundred page invoice. You're going to get a, you know, it's, it's going to be a mess of different things and we need to architect around that. And that's okay. So I think those are two skills. We, we seek individuals who are not, not just yes clickers on cloud code saying, yeah, go ahead and do that.

[00:51:26] Because as everybody knows that AI can be confidently wrong and very confidently wrong. And, and so you've got to be able to see that and flag that and say, why are you doing it that way? That doesn't make sense to me. Let's reconsider that. And so, you know, we, we look at stats like how many interrupts does a good engineer have in their sessions? How many messages per five hour period are they communicating? Is this just a set it and let it cook?

[00:51:54] Probably not for the type of things that we, we do. And, and I think that's, and it's important to know when you've got to get your hands dirty. To your point that the concept of what's, you know, what's the pull request and who's reviewing it? It's shifting a little bit than a year and a half ago or two years ago, right? Where I can have a, a pretty comprehensive business system up and running in, in 48 hours because we have templates and we know how to do these things.

[00:52:21] And a full platform that comes out of the box with the capabilities you need to do it securely and well at, and at scale. And so, you know, the speed of what you have to digest is, is shifting as well. And so we certainly seek on what we would consider our solution engineers, our engineering talent. They are going to be very business forward individuals. They're going to understand the business problem space. They are going to understand how AI is playing a role in that.

[00:52:48] But they're also going to understand the, the, you know, pro code, code when you need it, elements that aren't AI to stitch the thing together and, and stitch together with, with, with expected scale from day one. Nice. Yeah. There's a lot, there's a lot to think about. And I think a lot for people to reflect on in terms of their own, you know, sort of upskilling.

[00:53:12] And I think it all ties to, you know, a component of, of AI readiness that I don't know that people necessarily focus on, which is their own adaptability. Exactly. I mean, I, I certainly believe anybody who is AI curious in a white collar job in an office, they, you should be using these tools beyond just chat, right? You can go and use Claude Design and, you know, have a, create your own artifact and, and, you know, think about how a system would work.

[00:53:41] Just don't assume that that thing that you've been built over the weekend is, it's, is a production ready system. And be careful, right? Don't be slinging your credentials around into, into AI. AI is happy to put credentials in, in plain text code. And you just, you gotta be, you have to be thoughtful and careful, but the learning is, is super valuable, right? You know, you can, everybody's got a hobby that they could build some custom software to support their hobby. And it's fun to see it. It's fun to be a creator.

[00:54:09] And I do think that, that's one of the things that AI enables is, is that you can be more, more quickly creative than, than you ever were before. And by touching those tools, you get a sense of what's possible and how fast it is improving. Yeah. I mean, you know, I haven't committed, I don't have a wait list of, of customers waiting for me to finish this summer camp. This is just something that I thought was, was important for, for me to know. Obviously I want to, I don't want to be hypocritical.

[00:54:38] I want to elevate my own AIQ. And I, you know, even if it's just for myself, even if it's just for me to help out my entrepreneurial friends who don't have the time or the wherewithal to, to dig into this as deeply as, as I have. It's important for me to understand these concepts and also to, you know, keep up and have intelligent conversations with folks like yourself, Andrew. Yeah. And, and, and what to trust? Just one more anecdote as we wrap up here.

[00:55:04] You know, we, we hit, we hit the Stripe MCP and MCP is just kind of an interface that AI in theory can interact with and figure out how to, how to, how to pull data back or write data. We said, Hey, how many customers does this client have in Stripe? And the AI comp, the MCP company said they have a hundred customers. One zero zero. Exactly. Precisely. And we said, Hmm, that smells like a pagination question more so than anything else. And sure enough, right. With a little bit of digging, you're like, Oh, that's not right.

[00:55:34] Of course it's not right. So you have to, by using it, you also can start to get a sense of where it's confident errors might be. And, and you can then be defensive is the wrong word, but proactively intelligent around, and around how you test and reinforce and, and check the limits of these systems. And, and by getting hands on you, you, that's one of the best ways to do it. Absolutely. Andrew, this has been a fascinating discussion. Thank you so much for spending so much time with me.

[00:56:02] I just want to give you an opportunity to just close this out with any final thoughts, words of wisdom for, for those, you know, businesses or individuals just trying to figure this all out. Yeah. I mean, I, I'd put the, I'm in the rational optimist camp here to your point. AI might change jobs, but we've been changing jobs for many, many years. And we find new uses as information and tooling and systems become more available. The economy grows. And I think that's still the case here.

[00:56:30] I wouldn't, I wouldn't be too fearful of the, the doomsayers. So no reason to not be excited about such a powerful capability that we all now have, you know, in our, in our, in our pockets. It's amazing. All right, Andrew, thank you so much again for spending so much time with me. I think this is really insightful for my audience. So really, really appreciate it. I will include your LinkedIn profile, if that's okay, in the show notes and contextual.io is the site, correct? You got it. Contextual.io. Excellent. Thanks, Bob. All right. Thanks. Thanks, Andrew.

[00:57:00] Thanks, everyone, for listening. We will see you next time.