Insight ON 90% Adoption, 8,500 Agents, and What We'd Change If We Started Over

Audio transcript:

90% Adoption, 8,500 Agents, and What We'd Change If We Started Over

Mark Hughes (00:02):

There were hiccups along the way, no doubt. There were cases where we thought we could use agents and maybe made some changes in the business on the assumption that we could use agents and turned out not to be able to get as much value of them as we though we could, places where we thought we had the data available for things that we didn't and had to backtrack a little bit. I think that there are probably a fair number of tools where the value was overstated. None of those are failures really. They're all just learning experiences.

Jillian Viner (00:30):

How's your AI adoption going? Two years ago when we talked to people, they were stunned at how boldly Insight approached AI. But fast-forward to today, we have access to AI tools for over 14,000 teammates, a 90% adoption rate. Those teammates have created over 1,600 agents, saved an estimated of 60,000 hours and shared over 45,000 AI celebrations. And the wins, they just keep adding up. So today, the number one question that we get asked is how'd you do it? We're going to take you behind the scenes to how Insight became AI Client Zero. I'm Jillian Viner and this is Insight on AI Transformation. Mark, it's really nice to finally have you in our studio.

Mark (01:14):

Thank you very much. It's nice to be here.

Jillian (01:16):

Insight's been through a very long AI transformation journey going back, what, two years now?

Mark (01:22):

Yeah, a bit more than that. Yeah.

Jillian (01:23):

Yeah. What's been your seat at the table? Where were you along for that ride?

Mark (01:28):

Mostly hanging on for dear life. Fair. It transforms within itself over and over and over and over again. And so you're constantly feeling like you're playing catch up. So I was involved fairly early, more as a consumer of it with some of my teams than as a driver out of it. But as I moved into the CTO role, I took more of a leadership role. And then last year, all of the GBS resources who do AI work, we moved them into a central team. And so I've been providing strategic guidance since then, shall we say.

Jillian (02:01):

GBS, the GlobalBusiness

Mark (02:02):

Solutions. Global Business Solutions, that's

Jillian (02:03):

Right. As a consumer of it, you mean like testing, testing the new models, the

Mark (02:09):

Platforms? Well, actually some of the earliest adopters of AI here at Insight. So the Horizon platform there when it was still InsightGPT, just the chat interface, one of the earliest teams to start consuming that was our product data team. So for our digital product data that we use on e-comm and some other channels, we do a lot of enrichment of that. And that team was able to dramatically increase their productivity just by using InsightGPT to help generate some of that content. So they became fairly aggressive early adopters of it.

Jillian (02:41):

I want to hit you with a very unfair question because at this point in the game, a lot of organizations have maybe successfully deployed AI, they've got agents running, and yet when asked about the ROI of it, it's still like, oh, I don't know, vibes. So how do you enter conversations with your leadership team about the value of ROI, particularly now that consumption based usage means that pricing is changing and how we're paying for this is becoming a little bit more scrutinized?

Mark (03:09):

Yeah. The math on value out of AI is changing dramatically. And it wasn't a surprise. Everybody knew it was coming, but it seemed to come very, very quickly. And so it hasn't changed the basic ROI of doing agentic applications, helping with business processes, using AI to help people with reporting and understanding and querying with their data. We can manage token usage quite effectively in those kind of things. Where it's really been a big hit has been things like development tools, applications that are vibe coded that aren't necessarily designed around efficient token usage. And we've had to really think about the costs of those and how the costs of those are changing. We saw a 10X increase in cost for some of our developers who were using AI literally overnight.

Jillian (04:06):

Oops.

Mark (04:06):

Yes. And so essentially, we almost had to retrain everybody because we'd spent two years quite literally saying use it everywhere as much as you possibly can in every possible way. And now we've had to turn around and say, use it, but be thoughtful about how you're using it. Don't necessarily have to use the latest and greatest frontier models. The older models work just fine for things like documentation or generating emails. And so there's definitely, we've had to have people rethink. A lot of teams have been very, very successful with that. We've seen a lot of our development teams in particular changing the way they work with the tools and maintaining the productivity increases that we saw while reducing token usage. So it seems to be working out very well.

Jillian (04:54):

It's not the story that we've heard in the news of employees being very angry because for months and months they were told use it, use it, use it, and then it got taken away because of the cost.

Mark (05:03):

There's certainly an element of frustration around that. The idea that we just completely reversed course on them and the way we were thinking about AI, but we haven't really. All we're saying is it's still really important to use these tools. You just have to be a little bit more thoughtful about which tools you use for which things. And as long as we keep that in mind, we can keep good control of our token spend.

Jillian (05:30):

But what's the value story at the end of the day? You can't say that this particular workflow or agent saved you 10 hours a week and then that translates somewhere to the bottom line, right?

Mark (05:42):

Well, people do say that a lot. It's very, very difficult to find a nice, easy number to talk about that defines the value of the AI tools that we're deploying. And I think the reason is because every workflow that we're using them in is different. The advantages that they bring to one person or one team are very different from the advantages they bring to the other. And so while we could talk in terms of hours saved, what does that really mean? You save a few hours for a developer or you save a few hours for somebody in finance or you save a few hours for somebody on our client operations teams. All of those things have different value at the end of the day. And there isn't a nice single round number that we can point to that we can elevate up to our executives and say, look, this is how much we're saving with AI because it's just different for everybody and every team.

Jillian (06:34):

Do you find that you're thinking more like a CFO these days?

Mark (06:36):

I think FinOps is becoming a very big part of every technologist's repertoire. And to a degree, IT has always had to think that way. It's just much more public now. We went through this with cloud, right? The idea of consumption spending subscription-based licensing has been around for a while, but it's typically been something that we've managed centrally and controlled centrally and haven't really had as much impact on the users of the tools rather than the administrators of the tools. So we've always though this way. We're just being a bit more public about it now.

Jillian (07:14):

Fair enough. So the bottom line though is that you see the value. There is 100% value in having it

Mark (07:20):

In people's

Jillian (07:21):

Hands.

Mark (07:21):

There is absolutely value in having it in people's hands. The numbers are genuinely staggering when you look at them, just Copilot. When we run reports on the usage of Copilot at Insight, on a monthly basis, it's summarizing tens of thousands of hours of meetings. It's generating thousands and thousands of documents. It's summarizing email threads. It's catching people up on what they've missed with teams. And every time I go and I look at those reports, I think most recently there was a number of something like in May, it summarized 49,000 hours of meetings across Insight. Now of course that doesn't mean we had 49,000 hours of meetings, although I imagine that - Oh, I imagine we did. But five people were in that meeting and it summarized it for each of them. And how much time did that save them? We can't go back. The tools have made people's day-to-day, the boring bits or the time-consuming bits just so much easier and so much less painful that trying to take those tools away would be a huge mistake.

And while it's very difficult for us to quantify what tens of thousands of hours of meeting summarized actually means in terms of value for insight, it must mean something, right? People are doing something with that time.

And so yeah, I'm a huge believer.

Jillian (08:52):

Beyond the meeting summaries, transcripts, which are definitely valuable, people all across the org in different business units, different roles are coming up with use cases to help them in their day-to-day job. Some of the times they can build it by themselves. They can run a prompt, they can build a custom agent, whatever, but sometimes it does get to a point where they recognize they need some support from IT.

Mark (09:13):

Yes.

Jillian (09:14):

How do you control that volume of requests and make sure that whatever's being built is worth your team's time or that teammate's time?

Mark (09:25):

That's a really good question. It's one we're struggling. Well, perhaps not struggling, but it's one we're considering at the moment. So we are about to make some changes to our AI operating model and changes to the way that we govern and work with AI. So we're hoping to do some consolidation and bring a lot of the teams that have been very splintered together. Huge reason for doing that is because we are seeing a lot of duplication. And so I think as of today, there's 8,500 agents deployed at Insight. A lot of those agents are doing the same thing. That's okay. It's not actually a problem to have multiple agents doing the same things, particularly if it's an agent that you might build to clean out your inbox or I might build to go and see the most important emails I need to address first. It's

Jillian (10:16):

Tailored to me.

Mark (10:18):

And we're totally okay with having thousands of those. But when you get into the larger efforts where you might have an organization or a team that's putting something a little more complex together, something that's expected a lot of different teammates to use, we do have a duplication problem. And so we'll see sales teams in APAC and AMEA and North America and different go-to-markets all going out there and building the same thing without really being aware that other people are doing it. And so we have to apply some control to that. It's not a good use of our efforts. And it's also a problem for us because while vibe coding is an incredibly democratizing thing, it is important that it's done using supported tools. And in particular, it's important that it's done against the right data and the right data sets. And so we've had an ongoing effort for the same amount of time as we've been working with AI to get our core data systems ready for AI and set up in a way that minimizes hallucination, minimizes the problems that AI will have.

A lot of people either don't know or don't have access to those, don't know they exist. So we have to centralize, we have to apply some governance so that when people come up with a great idea, we can say to them, "Absolutely let us help you. There are three other teams who want to do the same thing. Let's pool our resources. Let's make sure that we get more out of it for that value rather than doing the same things three or four times over and let's make sure we're using the right data to support it." I'm

Jillian (11:47):

Going to come back to the data question, but I do want to understand a little bit more deeply. How are you addressing the duplication problem?

Mark (11:54):

Well, today it's not being addressed particularly well. You're

Jillian (11:59):

Admitting it,

Mark (12:00):

Which is

Jillian (12:00):

Great.

Mark (12:01):

It's refreshing.

And part of that is because we haven't had a centralized AI organization. GBS has provided some central stuff, our horizon platform in particular and some other tools, but all of the different regions and all the different groups at Insight have been free to do largely what they're on. And that was intentional. We've been through this incredibly exciting period where AI has suddenly solved that problem of having to wait for IT to be ready to build the thing for you. And getting people to enthusiastically embrace that instead of seeing AI as a threat was an incredible accomplishment. And all of the AI teams at Insight owned that and did it and did a really good job. But we're now at the point where because of the costs going up, because of the way that we have to start thinking about tokens, because of the risks to our data, we now have to apply a little more control.

And we don't want to stop people doing the things they want to do. We just want to make sure they're using the right tools and doing it with a bit of guidance.

Jillian (13:06):

I'm seeing a pattern here where it's a free for all, everybody go explore, figure it out and then okay, the pricing's changing, so now we got to rethink how we're using models. And also you can go use all the different tools, find the tool you like, go experiment, build things. Okay, now we need to come back and make sure we're not duplicating. So it does seem to be sort of a standard of almost proactively change management setting expectation.

Mark (13:31):

That's exactly right. And so we're not looking to stop people from doing things. The innovation has been amazing. We just want to provide a little more guidance as to how they innovate.

Jillian (13:41):

Yeah. Yeah. All right. Let's talk about the data piece because this is, I want to say, one of the most underrated elements of proper AI deployment. I think everybody recognizes that it's important, but maybe underestimates how big of a lift it's going to be. I don't know. Do I have that wrong?

Mark (13:57):

No, you have that right. And the good news for us is we're a good ways down that path. Solving the AI problem has always really been about solving the data problem. If you give AI good data with good context, it does good things. We started our efforts around that a couple of years ago. We have a central data team and data governance team, and they're working through what we call a medallion model, which is a fairly common approach to this where you have data at silver, gold and platinum levels. And the mass of data that we have at Insight has incredible value, but isn't particularly usable. We bring it into our central systems. Databricks is the one that we use. And incorporating all of those different tables and sources of data together gives us a silver model. It's now in one place. We have some understanding of it.

We then go through a cleanup process that brings it to a gold level. And at gold, it starts to become useful for AI. And so we're at a pretty good point. We reckon that we are almost 100% at Silver that you always find new data sources popping up here and there. So you're never done. And I think that's one of the misconceptions around data is people think of it as a project. Oh, clean up our data and then we're done. It is never finished.

Jillian (15:24):

This isn't like cleaning out Grandma's attic.

Mark (15:25):

Yeah. The cleaning process will continue forever for us, but we'll have tools and processes in place to make it much easier. So once it's a gold, it's useful for AI. When it's at Platinum, that's where we want to get. That then we can say with a Platinum data source, you can pretty much do anything you want with AI and you'll get good results. That's our end goal. We'll probably be mostly at Gold by the end of this year. It'll probably take a little bit longer to get to Platinum though.

Jillian (15:56):

How long has it been?

Mark (15:57):

Two years.

Jillian (15:58):

That's a long journey.

Mark (15:59):

Yeah. And it's actually been a lot faster than a lot of organizations do it.

Jillian (16:03):

Why? How?

Mark (16:04):

I think because there was a recognition at leadership level that data was important as well as AI. I think the weird thing about AI is you always feel like you're playing catch up. We mentioned that earlier. And because it changes so much, because it changes so fast, because there's always something new, you look out there and you look at the hype and you think, oh my God, we're so far behind. We've got to keep rushing. But Insight has actually been cutting edge on this stuff all along and we continue to be so. And that's our solutions teams, that's the teams that work with our clients, but it's also our internal teams. And the client zero story, the idea that we know and understand this stuff well enough that we can go and tell that story to the world effectively is really important to us internally.

And so leadership has recognized that from the beginning and there's been an understanding that if you want to do AI right, you have to do data right. And so there's been a willingness to invest in it. Yeah.

Jillian (17:00):

The speed and the feeling of being left behind is very real. And when you have people like yourself who are in the trenches in this thing every day, you can understand why or how you'd be able to keep up with it. But how? And that's even hard for you, right? So I think the other question that a lot of leaders are grappling with is how to help their team stay on the forefront of this. So how has insight helped people at least keep up with the current pace or a step behind or a step ahead? There's different stages for sure across our teams. Yeah. And

Mark (17:32):

For a long time we were a step ahead. Flight Academy in particular, I think was an amazing effort. Whose idea was that?

So that was the COE, that was Stanley Quinn and those teams that were involved in that. The idea of taking 14,000 people who should be terrified of this technology and making them wildly enthusiastic about it had lots of different parts. We have the AI champions, we have Flight Academy, and Flight Academy as a training mechanism was very, very well executed and very effective. We are a very highly AI enabled organization. And when we go out there and we talk to our clients and our vendors, it's genuinely surprising sometimes how far ahead we are. Flight Academy has, I don't want to say fallen behind, but we need to add a bit more to it now. And so I'm hoping we'll get to reinvigorate that a little bit. It needs to be, as we bring more central governance, we want to make sure people are trained and enabled on that.

We want to make sure people understand token usage and how to think about it. And so hopefully you'll see new life in Flight Academy coming.

Jillian (18:44):

Give me the quick elevator pitch of what Flight Academy is. How is it a successful learning tool?

Mark (18:50):

I think if you gamify that stuff, it always makes people more enthusiastic about it. An element of competition helps. A

Jillian (18:55):

Little bit competitive.

Mark (18:56):

Yeah.

Jillian (18:58):

I'm a little annoyed. I'm not further along than I am. It's just

Mark (19:00):

Taking the time for it. What level have you made it to?

Jillian (19:02):

You know what? Realistically, I'm probably at level four, but I just haven't gone in to actually update it.

Mark (19:08):

Well, I will reluctantly admit that I have only completed level one, so you're way ahead of me.

Jillian (19:12):

And

Mark (19:12):

Clearly

Jillian (19:13):

You're beyond level one, so this is a part of it. But it is a good way to keep people motivated, put some internal competition, and it's really nice guided structured learning.

Mark (19:23):

Yeah, it is. And it's great content. Part of the problem with AI is that you've got to constantly be updating the content that goes with it. And so there's work that's required, but yeah, Flight Academy did a great job for us.

Jillian (19:39):

I was literally on a call yesterday with two people at Insight who are on the cutting edge of this AI stuff. And one of them was showing us a new capability and the other person was like, "Oh, I didn't even..." And I was like, "I feel so honored that I get to be on these calls when people who I know are at the cutting edge of this are discovering it right along with me. It's a privilege."

Mark (19:56):

It's been really surprising to me how much is brought to us by people around Insight. Yeah, a lot of

Jillian (20:03):

Curiosity.

Mark (20:04):

Yeah, because yeah, people are excited and curious about this and they'll go out and they'll find a new tool and we'll start to get questions about, "Oh, when are we rolling out this new Microsoft capability?" It's like, "Well, it's only in beta and we haven't really put it through InfoSec yet." But yeah, it's definitely become a world where we are being pulled along as much as we are pushing from a technology perspective.

Jillian (20:27):

That's kind of remarkable actually.

Mark (20:30):

Yeah, it is.

Jillian (20:33):

I mean, I'm sure Flight Academy maybe was part of it. And I love what you said earlier, the journey or the ambition to try to get 14,000 people who should be afraid of this thing to not only accept it, but be excited about it. That's a miraculous accomplishment.

Mark (20:47):

It's a miraculous accomplishment and it also makes it into something they don't have to be afraid of because by embracing it in everything we do, we remove that. I mean, fear of it is a technology, fear of it is affecting my job, those sort of things. The people who embrace it are not going to be affected.

Jillian (21:08):

Yeah. And there's always going to be a human element to using it.

Mark (21:13):

There is. In

Jillian (21:13):

Fact, you have a good perspective of this concept we hear a lot about, which is human in the loop.

Mark (21:19):

Yes.

Jillian (21:19):

Elaborate on that.

Mark (21:20):

So I think having a human in the loop is really important. I do think we think about it in slightly the wrong way because one of the things that AI does, and we have this very clean term for it called hallucination, but it makes stuff up.

Jillian (21:40):

Confidently.

Mark (21:41):

Yeah, confidently. And it will tell you that the sky is purple. And when you say, "Well, actually the sky is blue," it'll say, "Oh yes, you're absolutely right. The sky is green." And I don't think people really look for that kind of mistake when they are reviewing things, verifying things, checking things. And so I think we've got to retrain the way we think about looking for problems because we're just not used to people making stuff up completely out of nothing.

Jillian (22:10):

That's

Mark (22:11):

True. We might be looking for stupid mistakes, simple mistakes, maybe somebody intentionally trying to mislead, but we don't look for complete fantasy, and that is what AI does. And so yeah, it's a new way of thinking about that.

Jillian (22:27):

I call it the razzle-dazzle effect

Mark (22:28):

Because

Jillian (22:29):

Especially if you're kind of new to AI and it produces something that looks so high quality, like it would've taken a person hours to do it, on the surface you're like, "Oh my God, this is so impressive. This is amazing." And if you have to go a level deeper, then you see all the flaws and the clearly AI written text and everything. But initially, it's truly yours razzle dazzle by

Mark (22:52):

It. It is. There's a rather good AI joke going around right now where the overworked person writes five bullet points, asks Claude to generate a 12-page document for it, gives the document to their boss who then asks Claude to summarize it to five bullet points because they don't have time to read the document.

Jillian (23:12):

That's actually quite perfect. Yes.

Mark (23:14):

Yeah. There's a lot of slop and the responsibility for working through the slop and finding it, it's almost like it's shifting upwards. And so yeah, there's a lot of new ways of thinking about work that we're going to have to deal with. Over generation of AI, over creation of slop, generating these incredible presentations and documents and everything else. Yeah, it can be a little addicting, but sometimes we really just need those five bullet points.

Jillian (23:45):

That's true. That's true. In addition to having the gold standard or maybe the platinum, is there diamond level

Mark (23:52):

Data? There are levels. There's semantic levels above it, but platinum is where we. Yeah.

Jillian (23:57):

Outside of that, it also, I think, comes down to applying AI to the right workflows and understanding what's a good AI use case versus not an AI use case. And I've heard that from people across Insight that they get these use cases and they look at it and they're like, "This actually isn't something that we should apply AI to." What's your model for that approach?

Mark (24:17):

I think the first thing is you should almost never look at an entire workflow, but you're much better off looking at individual tasks within a workflow. So

Jillian (24:26):

Break it down.

Mark (24:27):

Yeah, break it down. I believe that you should find the best process first, then decide where you want to apply AI to that process. There is a tendency for groups to try to use it as a bandaid for a bad process, and now you've just got AI doing a bad process. So in an ideal world, I would say you simplify, take the complexity, unnecessary complexity out of your process, pair it down to the most efficient way to do what you need to do, then look at it say, "Right, what can AI help me with in these steps?" And that way then you get the best of both worlds. Hopefully you're finding more efficiency in your business and then you're using automation to help solve for that.

Jillian (25:12):

All in all, the success at Insight as Client Zero has been a really exciting, interesting story. I'm sure it has not gone off without a hitch though. How would you describe Insight's AI adoption?

Mark (25:30):

There are times when I'm very cynical about it, and at the moment I'm actually very optimistic about it, so you'll get the optimistic view. I kind of want the cynical view too

Jillian (25:40):

Though.

Mark (25:42):

No, I think we've done very, very well. I think that if I were to point to one thing that I would change going forwards, and that's not the same thing as saying I would change it in the past because I think it served a purpose, but I think that we are past the point of AI for AI's sake. Now we need to be looking at it from the perspective of it's a tool in a toolbox, where can we apply this tool to get the most value? Whereas in the early days of our AI adoption, it was very much just you must use AI for everything because it's AI. And that was great for getting people enabled, that was great for getting people enthusiastic, that was great for learning, but that's something we need to think a little bit differently about. I think our journey was wild.

Jillian (26:33):

That's

Mark (26:33):

A

Jillian (26:33):

Great

Mark (26:34):

Singular

Jillian (26:34):

Word for

Mark (26:35):

It. Yeah. There were hiccups along the way, no doubt. There were cases where we thought we could use agents and maybe made some changes in the business on the assumption that we could use agents and turned out not to be able to get as much value of them as we thought we could, places where we thought we had the data available for things that we didn't and had to backtrack a little bit. I think that there are probably a fair number of tools where the value was overstated and we may have to go back and reexamine those and say, "Do we really want to keep these around? Are they really worth supporting?" But none of those are failures really. They're all just learning experiences. And so yeah, as you can tell, I'm quite optimistic about it at the moment.

Jillian (27:23):

What would you tell another organization or another leader that's been maybe having a hard time getting started or not really seeing a great value or output from it?

Mark (27:36):

Well, if it was a leader of an organization outside of Insight, I'd say we can help you. I think the first thing is just go for it. It's moving and changing so much that you can't wait for it to reach a point of stability or you can't wait for it to reach a point where it's all well understood and not constantly in flux because that's going to be a long time yet. And if you do, I don't think people will necessarily get left behind entirely, but there's definitely a catch up that would have to be played. Dive in, start to learn, and you'll get much more advantage out of the change as it's coming. And then just if you don't like change, it's not a game to be in. That's the

Jillian (28:24):

Truth.

Mark (28:25):

Yeah. And I mean, that's the statement we're saying to everybody at Insight all the time right now. We have a lot of change going on. We've got the changes to our core platforms with the Workday migration that's ongoing. We've got all of these AI tools being introduced. We've got the changes that those bring to our day to day, the changes they're going to bring to our processes. Yeah, people who are uncomfortable with change aren't going to have a particularly comfortable few years, I

Jillian (28:50):

Think. Yeah. I want to say it was maybe 2020 where we talked a lot about change fatigue

Mark (28:56):

And

Jillian (28:56):

It feels like we might be coming up on that again.

Mark (28:58):

I think so. I think there's nothing we can do about it because for the next few years at least it's going to accelerate and not get slower.

Jillian (29:04):

Yeah. Yeah. I didn't get a chance to talk to you very much about the Horizon platform. You mentioned it at the very beginning.

Mark (29:11):

Yes.

Jillian (29:13):

Tell me a little bit more about the Horizon platform. Why was that something that Insight felt we needed to build? Was it something that we intentionally set out to build?

Mark (29:21):

It was. And this is something that Shannon Meyers and team put together early on. She did an amazing job. She deserves a great deal of credit for that. I call her the

Jillian (29:32):

Godmother of Horizon.

Mark (29:33):

Yeah. And she continues to do an amazing job with AI for us. But we originally built it out because it wasn't something we could bring in. There were models - Like a chat platform? Yeah. And so ChatGPT was the big transformative announcement and then suddenly everybody wanted AI. We weren't able to use ChatGPT because OpenAI did not mean. To our enterprise security requirements. So we couldn't bring that platform in. We had no choice but to build one for ourselves because data security in particular, but also some other aspects had to be the most important thing for us. And there's a lot of little problems that you deal with when you're starting to put these platforms together. The example I always think about is something like an Ask HR agent.

There's data in there that you can see that I shouldn't be able to see and that I can see and that you shouldn't be able to see. And that's also true of our finance agents where somebody in executive leadership might be able to see forward-looking data, whereas somebody like me wouldn't be able to. And the platforms that were publicly available back then just didn't make that kind of discernment. And so we needed a platform that allowed us to put those barriers around different organizations' data, different people's data, different levels of security. And so we built it. And we continued to develop it, adding new features. For a while, it almost felt though that we were sort of competing with the Claudes and the copilots of the world. And so we did actually take a moment and say, "Will this platform continue to have value for us?" But I think now as we're moving into the age of token pricing, token-based pricing, it's taken on a whole new life because now we're looking at using it as an agentic orchestration platform, which helps to choose which model gets used for different things.

We're evaluating whether we can have models running locally in our data centers rather than going out to the cloud for the models we use today. And so it's going to become a very effective way for us to control our AI costs. Interesting. Yeah, Horizon keeps gaining new and interesting stuff. As the technology changes, Insight as an. And again, I don't know why this surprises me, but it always surprised me. We're right on the cutting edge. And with a platform like Horizon, as these new things come up, we've got a tool that we can learn with and use with and make all of those things available to all of our employees. So it's a really cool and interesting and valuable platform.

Jillian (32:13):

Yeah. So initially started off as just like we can't use what's available because of security reasons, but we know people are going to use it anyway,

Mark (32:21):

So we need to

Jillian (32:21):

Come up with a

Mark (32:21):

Different solution. We need to come up with something to solve. Yeah.

Jillian (32:23):

Then it was the segmentation of information for right access, right person. And now it's more of a economics being more prudent about the budget for token usage.

Mark (32:37):

That's exactly right.

Jillian (32:38):

And I know that new features are rolling out all the time. What next problem do you think it's going to solve?

Mark (32:43):

I think there will always be, we have an active product owner team and I think there's some. I'm not going to give it away too much, but there's some very interesting features coming up with things like user created skills. So for example, you could define a skill and deploy it in Horizon, which would then be available for everybody else to use as part of their prompts rather than necessarily as being an agent that they wanted. We like the idea of, and I say we, the product owners and the team that runs that, they're really good and I'm not really a part of it. The idea of user generated stuff, the idea of being able to share what you do and find valuable in the AI platform so that other people don't have to search for it. And Amelia, the agent that does the AI celebrations was sort of a first step towards that and you'll see more stuff than that.

But I think the real core changes to where Horizon will bring value are actually going to be under the covers. The idea of, for a long time we let anybody use any model they wanted.

Unfortunately, that's too expensive. So some controls around that, being smart about which model to pick to solve which problem. The stuff we're doing with InsightIQ where it automatically routes to the right agent. So you don't have to know which agent to ask your question of. You just ask it to InsightIQ. Insight IQ routes it to the right agent, finds your answer, brings it back potentially multiple agents. And then finally the idea of being able to use locally deployed models to save on cost and have the system route to them in a way that's completely invisible to the user so that you and I don't have to think about it. Horizon will think about it for us.

Jillian (34:39):

So it really is like the front door to every AI lab. It's the front door to our data and it doesn't matter what's changing in terms of what new models get released, which labs fall ahead or behind. That's

Mark (34:57):

Right.

Jillian (34:57):

It's all kind of like we have control over that destiny.

Mark (35:00):

That's right.

Jillian (35:01):

That feels like a big win.

Mark (35:02):

Yeah, it is. Horizon is something a lot of organizations don't have and it gives us an element of control that we otherwise wouldn't. You're absolutely right.

Jillian (35:15):

When you're talking to clients, partner, vendors about the journey that we've been through and the capabilities that we have today, maybe even a teaser of what's next, what do you find most often catches their attention? What's unique about the way that we've done this compared to other organizations? I mean, you even called out that even if we feel behind, Insight is actually ahead in a lot of ways. How is that possible?

Mark (35:43):

I think when I have that conversation with clients in particular, they seem surprised that we jumped into it as wildly and enthusiastically as we did. Fair. We really embraced it and embraced it from top to bottom across every part of the organization. It was a challenge to keep up with that on the IT side and the GBS side. And not a lot of companies did that. Of course some did, but there's a lot of risk that comes along with that and we had to embrace that. I think it was very challenging for the InfoSec team in particular. But yeah, I think that's probably the thing that stands out the most. But then also just how far we've come. We're a very sophisticated AI organization at this point. We use it more broadly than almost any other company that I'm aware of. We have it rolled out to more of our employees.

I mean, every single one of our employees has access to Copilot. We let people choose if they want to use Claude or the other tools that are available. We've embraced it incredibly broadly and that is also surprising.

Jillian (37:05):

Very good. Well, Mark, thank you so much for your time today. It's been an absolute pleasure.

Mark (37:09):

You're welcome. Thank you very much.

Speaker 3 (37:11):

Thanks for listening to this episode of Insight On. If today's conversation sparked an idea or raised a challenge you're facing, head to insight.com. You'll find the resources, case studies, and real world solutions to help you lead with clarity. If you found this episode to be helpful, be sure to follow Insight On, leave a review, and share it with a colleague. It's how we grow the conversation and help more leaders make better tech decisions. Discover more at insight.com. The views and opinions expressed in this podcast are of those of the host and the guests and do not necessarily reflect on the official policy or position of Insight or its affiliates. This content is for informational purposes only, should not be considered as professional or legal advice.