Insight ON The Tokenomics Behind Company-Wide AI That Actually Scales

Audio transcript:

The Tokenomics Behind Company-Wide AI That Actually Scales

Shana Meyers (00:02):

If we gave everybody a Gemini and a Copilot license, that'd be very costly. We can do it for a 10th of the cost. That's much more palatable and it gives access to everybody. It's not just these specific teams that are heavy technical teams have access to those tools. Everybody can have access to those things regardless of their role.

Jillian Viner (00:27):

Ever since the AI Labs changed their pricing structure to be usage-based consumption, it's really forced organizations to think hard about their AI budgets. So the question that we hear a lot these days is how do we enable our organization both securely without overspending on licensing? And the answer might surprise you. I'm Jillian Viner and this is Insight on AI Deployment. I was talking to some folks a lot during the Insight AI client story interviews and everything, and your name comes up a lot.

Shana (00:59):

Okay.

Jillian (01:00):

I like to think of you as the godmother behind Insight Horizon, which is our enterprise AI hub.

Shana (01:09):

That's amazing. I love that title. I love it. I think

Jillian (01:12):

We should make it official. Give you some swag there. But what's so impressive about this platform is that number one, just the adoption of it. I mean, this platform serves 14,000 teammates. The adoption's been pretty high almost since day one. And people can go in there, select the model that they want to use, create custom insights. But this platform almost happened by accident.

Shana (01:39):

Kind

Jillian (01:39):

Of. Kind of. So

Shana (01:40):

Can

Jillian (01:41):

You tell me the story? Take me back to the beginning. What

Shana (01:44):

Was

Jillian (01:44):

The original mission that you were working on that ultimately became the Insight AI platform, the AI Hub?

Shana (01:51):

In the early days, we just knew we wanted to prove the use cases. And so we started and we had a very prescribed use case with legal and we were competing against a couple of SaaS giants out there and we had this very prescribed use case, people testing, legal, external counsel, everybody testing. And hands down, we were equal to or better in every category compared to these two SaaS giants. And we said, "This is amazing. I was so excited." And then we got together and they decided to choose one of those others out there. And I was crushed. I was like, "This is terrible. We've put all our heart and soul into this and we really thought we did something great." And they just kind of went the other way. And one of the SVPs making that decision came to me and he said, "You guys have been amazing, and if you tell me that this is what Insight is going to do and this is our future, I'll support you 100%." He said, "However, if you think your resources would be better spent on different use cases, I'll also support you there." So that was a tough one and I had to internalize that a little bit, but he was right.

(03:15):

We weren't going to compete against them trying to do the same things. They had hundreds of people focused on those platforms. We weren't going to do that. So we took a step back and that's when we said, okay, what is going to give us the most benefit? That's our data. Our data, that's our moat, our soul, our context. Nobody else has that. Nobody else can compete with us when they talk about that data. So then that's when we kind of focused on bringing our data to the platform. So originally within a couple weeks we rolled out InsightGPT is what we called it, and that was really just our internal use of just one model at the time. And that grew into the platform that we built today, which we internally call Horizon AI that serves up multiple models and all of the features and functionality that we have.

(04:17):

It is available externally called the Enterprise AI Hub.

Jillian (04:23):

What is it exactly? Because it's not a harness.

Shana (04:27):

No, it's a way for us to serve up the models, have agents pointed to a curated data set that we use internally, also connected to technology behind it. We can connect it to other applications and we continually add features and functionality, but it's a secure way for us to give access to every teammate without a specific license for Gemini or Copilot, those types of things. SharePoint's great, but if you think of Insight Insight and some of the SharePoint sites, they have 80,000 documents out there. How many of those are current, do you suppose? Let's just say it's 20,000. If that's the case, those 60,000 that are not current, are they going to give answers counter to that? So how do we fix that and focus on our data to get real value? Because sometimes finding that Insight's been around for a long time. We've got a lot of data.

(05:33):

Let's put it to a good use.

Jillian (05:34):

Yeah. Not an issue that is strictly an Insight

Shana (05:39):

Issue. No, everybody has that.

Jillian (05:42):

Why would you say that this platform is more valuable than just using Claude or just having an enterprise ChatGPT license or Copilot? Why this platform?

Shana (05:55):

It has a lot of flexibility and we can tie it to things that those other things can't tie to easily. Those other platforms, we would have to go to them and work with them. This way we're in charge of our own destiny. We can choose our own adventure and pick our own path and we can do multiples. Why be stuck with one when we can do all of them? You

Jillian (06:21):

Could switch out, we're going to do Anthropic and

Shana (06:23):

We're not. And connect to all of our different data sets in the back. That's not as easy with some of those other ones and sometimes more costly.

Jillian (06:34):

Interesting. You're talking about APIs?

Shana (06:36):

Yes. So if we gave everybody a Gemini and a Copilot license, that'd be very costly. We can do it for a 10th of the cost. That's much more palatable. And it gives access to everybody. It's not just these specific teams that are heavy technical teams have access to those tools. Everybody can have access to those things regardless of their role.

Jillian (07:07):

I think about too, how often teammates may get access to a software or tool or something. They use it for a little bit and then they forget about it, but you're paying for it for a whole year. So if you were to give everybody access to multiple platforms and then they find that they're only using one, it takes someone being really proactive to be

Shana (07:25):

Like - Yes, it takes us a lot of time to curate back those licenses and then what's the right timing to do it? And these are all monthly costs. So Horizon, we're just paying for the tokens that we're using with whatever the model is.

Jillian (07:42):

Sounds brilliant.

Shana (07:43):

Yeah, we like to think so. Put a little dust up your shoulder. Yeah.

Jillian (07:48):

How long did it take you to get the first iteration of Horizon, which is what we internally call it, but it's like the AI hub that we build. How long did it take for that to go live?

Shana (07:59):

Our first version went within six weeks of Azure OpenAI launching. Fast. Yeah, so super fast.

Jillian (08:07):

How long does it take today?

Shana (08:08):

Because

Jillian (08:09):

That was two years ago.

Shana (08:10):

Right. I mean, we are continually adding and new features and models and things like that, so pretty much on a weekly basis.

Jillian (08:19):

The custom insights that you mentioned, which we're taught, those are really personalized agents or long form prompts that people can save and use over and over again. Tell me a little bit more about those. What's been the greatest benefit of having that feature in this platform?

Shana (08:34):

I think when teammates, especially on teams where you have a few people on a team and they're doing the same kind of work, you have a prompt versus they have a prompt, which one works better? This way you can solidify that, take the best of both, and both of you can use the same ones. You're getting the same results when you use it. So just being able to share that knowledge, what you've built with others, that's an easy way to do it.

Jillian (09:01):

That makes sense. Especially we hear a lot sharing use cases among peer groups is important because what use case you use may not be helpful to me in my role,

Shana (09:11):

But

Jillian (09:11):

People in my role can share things.

Shana (09:13):

Absolutely.

Jillian (09:14):

What has that meant for adoption? Do you see that it's increased outcomes?

Shana (09:20):

Yeah, I think so. And the ability to get documents out, that's always been a big one. But I think Flight Academy really helped and I don't think either platform would've been successful by itself. And I think that those two being tied together, we went on training people, getting that education, driving adoption, that kind of helped hand in hand and really listening to teammates, what are that features and functionality that you need or want that will make you want to use it? We don't want it to be something that's forced, but if you think about that as a platform, there is no other application at Insight more used than your standard Microsoft products on your desktop. So Horizon is used by more teammates than any other application that we have, which is amazing to me. But that tells me our teammates get it. They want it, they want to do better and get rid of some of that work that just drains your brain.

(10:30):

I don't want to do that work. I want something or someone else to do it for me.

Jillian (10:36):

Yeah. Flight Academy really told them how, but the Horizon platform gave them the capability to actually go do it, the hands-on learning. And the outcomes of that together, I think the stat is from a couple months ago even, but we do log people's AI celebrations, sharing those wins, over 45,000 celebrations across the org. And it's not every use case, it's just whatever someone takes the time to go document.

Shana (11:07):

Absolutely. Right.

Jillian (11:09):

What's been the most interesting or wild thing that you've seen come out of those AI celebrations?

Shana (11:17):

So it was interesting in the early days too, there was a lot of contests. What's the best use case? Send in the best use case. We'll have a panel of judges. I judged a lot of those contests. And one of the very first ones I did was for finance team. And it's a lot to wrap your head around technically just for even people that are technical, but general population, I was like, "I'm not sure how much we're going to get or if this is going to be valid." Early on we had that contest, there was probably 235 submissions. And when I read through them, all of them, I would say probably 95% of them were viable. And a lot of them were very similar across the teams. And I was amazed by that because I though, if this is your finance team who's not necessarily technical, they get it.

(12:20):

It was amazing to me and they were very similar. How do I find that information? How do I get rid of that soul sucking work? That's what everybody wanted because of those manual efforts I talked about that we've added over time because we didn't have the technology capabilities.

Jillian (12:39):

You could just waste hours searching for one document or one person who had the answer to something specific. Yeah. That's a really big signal too, that if you've got multiple people even across multiple departments coming up with solutions for the same problem, that's a pretty big problem.

Shana (12:57):

Yes, yes. And it's probably, a lot of it's never been on anybody's radar because that was just the way we did business. So I love to talk to people about what's the art of the possible? What if we could take that 37-step manual process and make it a two-step automated process with someone, a human in the loop to validate the results? But that's the exciting part. That's why I said I think there's a lot of opportunities for those things.

Jillian (13:26):

Have you seen that come to fruition?

Shana (13:28):

Absolutely. Absolutely. I've seen where we've had cost avoidance where we were so far behind on doing certain tasks because it was very manual and we had 10 people working on that. And to keep even current, we would have to hire 20 more people. Well, we automated that and we put AI in there and now we have down to five people that are validating it and they've gone on to do other things. That's amazing. That's very small when you talk about 14,000, but if you can take that in every department, every area across insights or any company, just imagine those results. That's huge.

Jillian (14:11):

There's a temptation to continue to measure the ROI in terms of time saved. I love your perspective on this. So tell me why that is a not great model for measuring ROI.

Shana (14:23):

It's really hard, right? And any executive's going to say, "Okay, we calculate that time and the average cost of the teammates that would be doing that work, that's let's say a million dollars a year." Okay, well, the first thing they're going to ask, "Can I get 20 heads? Do I get a check? How do I get that back?" I can't give you that. But I think we need to change that, flip that a little bit and say the ROI is what does that enable us to do net new? What are the other things we can go? Is it new clients? Maybe it's focusing on those clients to expand that book of business or whatever it is. How do we share that story? So we think about it as an investment and not just I get a check back in the mail because that's really hard to do and nobody wants to give that up.

(15:21):

And frankly, that's what scares people if you go into it saying, "Hey, we're going to cut a hundred heads in order to do this." You're not going to get the best of people to help you bring that use case to life.

Jillian (15:35):

Yeah. How have you encountered people who have approached AI with resistance or fear?

Shana (15:42):

Honestly, we haven't had a lot of that, but I think proving to them what it can do and how that can make their life, their role better, that's really what we've done. And a lot of it, sometimes it gets uncomfortable because we're questioning this is what they do every day. They've done that 27-step manual process for the last five years. I couldn't do it, but that change is hard sometimes. But what are the benefits that they get and what is next for them and how that makes their life easier? We haven't had a lot of that, frankly. And I think it's because we don't say, "Hey, you're going to have to give people to get this."

Jillian (16:29):

And you've actually sort of walked the front lines to see what some

Shana (16:32):

Of the processes

Jillian (16:33):

Are. Tell me one.

Shana (16:35):

We've been working in the warehouse recently in sustainability, and I am so convinced now more than ever that the way we do agile development, vibe coding development, we have embedded ourselves in the warehouse. If that team would've called us and said, "Hey, we want an application and here's our list of requirements," we could have built them something. I don't know if it would've worked or would've been very successful. So we got in there and we did the job with them and kind of captured all of that as we were developing it. So we've had a developer right there, has a seat right on the lines, receiving the boxes off the trucks, going through the whole process. I mean, it's been amazing, and I can't say enough about how that. We are so much further along than we would ever be if it would've been a traditional project.

(17:31):

I think AI changes project management, how we think about those things because there's not a. A lot of traditional project management and projects, you have an end goal. This is what I get at the end. We don't know what the end is yet. We know what the next step is and what we want to do. We have a vision. Technology and things are changing so fast. Six months from now, that might look completely different. And that's the way we've treated Horizon, what's next so that it doesn't get stale and old and people don't want to use it. What's the next thing that's driving that roadmap?

Jillian (18:13):

I want to go back to something that you said earlier about the model selection. People were really excited when that came to the platform, and especially now with the consumption-based billing, there's a lot of questions about what is the right model? There's model wars, and it seems like every week one of the labs is dropping the latest model. And insight, our teammates are particularly spoiled because we have access to pretty much all of them. Yeah, right. Yeah. Sounds expensive. It sounds almost impossible for other orgs, and yet we have that capability. How is that possible?

Shana (18:53):

So I think with Horizon, we really found a way to do that technologically where not everybody had to have a license for everything, so we could serve them up these amazing capabilities without having to have a Gemini license or micro copilot license or any of those things. So it does cost money. Everything costs money. But in a lot of those cases, we were able to do that for a fraction of the cost, building out that infrastructure to serve everybody and using those models just like any other company would use the consumption, but at a much lower cost by doing it in-house on that platform. So when we give people the freedom of choice and to try it, try each model for the same prompt or your insight that you're creating, that gives them a lot of freedom and a lot of power to do it their way.

Jillian (19:50):

And

Shana (19:50):

What's

Jillian (19:50):

Been the benefit to the business doing

Shana (19:52):

That? I think people are out there experimenting and they've come up with these. I'm amazed all the time. People ping me and say, "Hey, I'd love to show you what I've done and see if you have any feedback." And they've done amazing things and automated things that they didn't even. They're not developers, they're not technical folks, but they've been able to do that on their own. That's amazing and that's really powerful. And the one thing we wanted to make sure enable people to do it securely. We also knew in the beginning if you block those things or stop people, they'll always find a way around it. That puts us more at risk and makes my job harder. So let's enable them to do it securely and give them access to all the tools and then they can forge their own path. And when it gets bigger than just their team, their department, whatever, then come to us for help and we've been able to drive some of those other higher level use cases.

Jillian (20:55):

You mentioned that Horizon is connected to our SharePoint, which has thousands and thousands of documents, maybe 20,000 of them are current. How did you address that?

Shana (21:09):

We also learned early on that when you take a use case, seems pretty simple, should take a week or so, was not so simple because there was so much data. And so we found that the more finely curated data sets we could have, the better the results would be. So in those cases where we. The AskHR, we've very finely tuned, they've done an amazing job at curating that data set, and we only pull from that set. So we're very particular in those use cases to understand what they're trying to get out of it, how that data is working. And we do a lot of testing with them, whoever our stakeholder is and ourselves against the different models, what gives us the best results, because those are set to a specific model when you have your data set. So it's been a challenge. We have a lot of data,

Jillian (22:12):

As does every org. You mentioned AskHR, that's one of the agents that we have. It's one of many. Explain how that works.

Shana (22:21):

It's great, and that was one of our first persona based, meaning if I go to it and I say, "What's the next holiday?" It knows that I'm in the United States and that I'm going to pull the US holidays. And if a teammate in Manila goes and asks the same question, it'll bring back the Manila holidays. But it'll also, if I want to ask what the Manila holidays, it'll tell me that as well. But again, they curated that data, they manage that data, HR does, and so they are constantly testing and adding new data. And that's the other beauty of Horizon. They manage their data. They don't have to come to us. So if they own that SharePoint site or wherever the data lives, they continue to add, update, edit, whatever data there, and it automatically captures that next time a question is asked.

Jillian (23:16):

The business functions are responsible. They're

Shana (23:19):

Stewards of

Jillian (23:19):

Their own data.

Shana (23:20):

Correct. And I think that's part of the project management and things that's changing. It's a partnership, it's collaborative. You maybe previously would put in a ticket and say, "I want this functionality," and somebody would just do it in the back. You wouldn't know who that was. Now we have to work together and you have to tell me what you want it to look like and that outcome, and we work together to get it there.

Jillian (23:46):

Everything has a clear purpose and a clear user experience. Yes. Yeah. And the agents that have developed have really come from, again, what you talked about earlier with groups coming to the table with the same solution. It's a really clear signal that there's a broader use case for this.

Shana (24:03):

Absolutely.

Jillian (24:04):

Interesting. I'm going to go back to the warehouse for a minute.

Shana (24:08):

What

Jillian (24:08):

Are you building? What are you making?

Shana (24:09):

So we've making an app, and today's sustainability, they were actually receiving. These are returned pieces of equipment for different customers, and Insight being one of those customers, they received that equipment. They had no way to track. You don't know what comes in a box, could be a laptop, two monitors, whatever it is, but they have one tracking number. They were cutting the tracking label off the box and then making photocopies and attaching it to each piece of equipment and sending it through the process. Well, sometimes that would get lost, it would get blown off or whatever. So now we're tracking that, each device to a tracking number. They scan it and they take a picture. So customers, there's different grading that happens. So if you have a gash off the top and it's more than three inches long, maybe this is how you're going to repair it or maybe you're going to dispose of it.

(25:13):

And so a lot of that equipment, if it comes, it's going to be auto disposed of. Let's get that out of the path first because otherwise it was going all the way down the line before. So now we can. Okay, this is going to go to disposal right away. So a lot of those manual efforts, so they were receiving a thousand packages a day between the two sites, a thousand. Wow.

(25:37):

So if you think about teammates cutting off labels off a thousand packages. Oh, forget it. Yes.

Jillian (25:42):

Forget

Shana (25:43):

It. I

Jillian (25:43):

Don't even like opening more than two Amazon

Shana (25:45):

Boxes. Right. Yeah. So I mean, just that stuff, let's make that easier. Again, for scale, we might not be moving those teammates to doing other things because we're freeing up time. We may be increasing their ability to take 2,000 packages a day.

Jillian (26:06):

It sounds like almost every use case that comes to you, there's a lot of work that has to go into evaluating what is the process, what happens and why does it happen this

Shana (26:14):

Way? Right. Yes. And a lot of times you have to ask, "Why do we do this?" And sometimes we don't know. It's just, "I've been doing it that way forever. The person before me did it that way." So then we have to ask questions like, "What happens if we don't do that piece anymore? Let's see what happens." So there has to be some risk in there because we're kind of redesigning, re-imagining what that could look like. And what if you connected it to this? What if you had this? It is kind of an interview, so to speak, for those people to get them to reimagine what that could look like and how that could make their lives better.

Jillian (27:04):

Yeah. And iterative because the other thing you've talked about is you may start a project for one specific use case they've identified. Cool, that's done. But now what if?

Shana (27:17):

What if, yes, yes. And that's the great part, and we have the flexibility to do that. Now we're going to add that in there. I think even with the warehouse project, we've changed that roadmap multiple times because we got further in the process and I said, "Oh my gosh, you know what we really need is X. Okay, let's focus on that, then we'll get to the other pieces." So it is very iterative. We just don't know what the end looks like. It's not a traditional project in that sense.

Jillian (27:50):

Yeah. What advice would you give to another organization or other leaders that are really contemplating, they know that they need to bring AI into their business, or maybe they've already chosen a platform. Why might they reconsider something like the AI enterprise hub?

Shana (28:07):

I think just enablement, right? Enablement, and that drives adoption, and that drives thinking through and making great decisions. I tell people all the time, our job is to figure out the technology, how to use it, those things, but the power of those use cases come from each of the individual teammates. Empower your people so that they can help you. This is not an IT does this for an organization. This is we enable people so that everybody can start seeing that ROI.

Jillian (28:45):

One question I love to ask, I think I asked you this in our first interview. Thinking back on the journey, what was the thing that you would point to now as like, "Oh, we failed, but it was a helpful failure?"

Shana (29:01):

I think that very first use case that we went and tried to do, and we were successful, but ultimately we learned a valuable lesson that we were not going to compete with others in the marketplace where our benefit and our value is, our data, our moat, that's our context driving that, that's where our value is. So I think that failure was the precipitous to everything that came after us.

Jillian (29:35):

I heard at the time, but

Shana (29:36):

Look

Jillian (29:37):

Where it left things.

Shana (29:38):

Yeah. I think I hung up the phone pretty hard there virtually on that call, but yeah, I was like, "Yeah, dang it."

Jillian (29:46):

It's not as satisfying when you can't.

Shana (29:48):

I know, right? It's not a physical slam down the phone, but yeah, that one hurt. That's been

Jillian (29:54):

A really hard conversation.

Shana (29:55):

Yeah, it was. But I give a lot of credit to that SVP who He didn't tell me why. He kind of led me down that path so that I could realize it on my own. Took me a minute, but I got there.

Jillian (30:11):

Have you had to have that conversation with other teammates?

Shana (30:14):

Yeah. That's not going to be our focus. That comes natively AI in different tools. That's not where we're going to spend our time. We want to make sure people are doing securely. We'll provide the governance and security, but we're not going to put efforts into that specifically. So yeah, for sure.

Jillian (30:38):

It sure is. It sure is. Yeah. Well, awesome. Thank you so much for your insights. I appreciate it.

Speaker 3 (30:43):

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.