Get control of AI costs before the bill arrives

AI cost management is the practice of tracking, allocating and optimizing spending associated with AI tools, services, models and cloud resources. It helps organizations understand where AI investments are occurring and whether those investments deliver value.

AI costs often span multiple technologies, including cloud services, SaaS applications and consumption-based AI platforms. These costs are frequently distributed across different teams and budgets, making visibility challenging.

Organizations can improve visibility by centralizing cost data, establishing ownership, implementing governance processes and monitoring AI consumption across cloud and SaaS environments.

This webinar is intended for IT leaders, FinOps teams, cloud practitioners, SaaS managers, finance leaders and anyone responsible for managing AI investments.

You'll learn practical strategies for reducing AI waste, increasing visibility, improving governance and creating accountability around AI spending.

Victoria Anderson (0:07 – 0:13)

Okay, great. Mark and Chris, we are live. You can take it away.

Mark Bradley (0:13 – 0:39)

Awesome. Hello everyone. It's Mark and Chris. Um, we'll go ahead and do an introduction and then uh get started here. Uh, welcome everyone. Thanks for joining. Uh, my name is Mark Bradley.

I'm one of the uh product managers here at Flexera and I'm super excited to uh share what we're doing with uh AI costs before the bill arrives and I'll pass it to Chris to introduce himself.

Chris Hayes (0:40 – 1:15)

Hey, thanks a lot Mark. Um let's say a good morning, good afternoon, good evening depending on the time zone where you are in the world. I apologize in advance. I'm having technology issues. So instead of you know video like Mark you're just going to see that big picture in my head this whole time.

Um so my name is Chris Hayes. I am a principal solution architect at AnglePoint. Um we've you know worked in the industry for several years are familiar with multiple tools and really happy to be working with the technology experts at Flexera like Mark and team to discuss how you manage your AI costs. Um so with that Mark let's uh let's take it away.

Mark Bradley (1:16 – 9:11)

All right. So today, Chris and I are going to cover what we consider is probably the most significant shift uh in enterprise technology history. I think uh all of us are seeing this change. Uh I think we're all here because we want to learn more. We want to hear what other people are saying. Um and we want to validate that Chris isn't an agent on this call, that it's the real Chris. So as we go forward, we'll identify and track uh show how we track AI cost. Uh we'll show how we give visibility uh into AI investments before cost gets out of control. We'll talk about um understanding what's being used and how it's being used. We'll help you uh with some visibility and show you ways that uh we can help reduce waste and then establish governance and then kind of really dig in. And I think to help kick it off, um I think what we're going to do is do a poll question.

All right. Which team is responsible for AI spend at your organization?

And you can have multiple choices. And I think this was a good question.

When we um you know as as you click away on here when we were uh we did some road shows and uh you know Chicago, San Francisco, New York and when we would ask this question you could see that there was kind of a perplexed look because it's not an easy question to answer and I think it's been different in each of the different organizations.

Uh sometimes it comes up as an initiative uh from procurement where they've negotiated a deal. Sometimes it's the FinOps group because uh they want to take advantage of some models uh to help enhance their solutions. Um others have just went out and did the $20 shadow IT uh and did that and so now it's kind of everywhere.

So the sprawl is real and looks like what we have here is the combination of uh FinOps teams and infrastructure teams kind of leading the charge. Not a single uh owns it and then unsure and by all means uh let's use the Q&A section, the messaging section uh share your thoughts, share your ideas.

Um wait, what's it saying? The sound quality is terrible. this end even if it's uh light mode. Hoping for the recording. Oh yeah, let us know if the sound quality is not good. Okay, just checking there. Making sure we had a pulse on the discussion.

Cool. I didn't know if it was between my keyboard and chair, but some sometimes it is.

All right, so it looks like we have a good poll question here. Looks like um one, thanks for voting. two. Looks like uh FinOps teams and IT infrastructure teams are leading the charge.

Um cool, cool, cool, cool. Yeah, share your thoughts in the uh in the messages.

Let's uh let's move on to the next uh let me see where I'm at here. We did that. All right, so technology is the largest area of spend after people and AI vendors want to make sure that they become the first. And the reason we kind of put that in there is because we're seeing that it's evolving. We're seeing a lot of companies that have uh made some changes on their orgs and then allude to the fact that they had AI to do what that individual was doing prior.

Um, but before we get too far deep, um, I've been in the industry for a while and as I look through some of the names, there's a few of us that have been in this industry for a while and we've seen this before.

And I want to take us back just for a moment to 1999. And the reason I want to take you there is because we had a huge change. That was like our last really big industry change. uh Mark Adre, he talked about how software was eating the world. He wrote an article in the Wall Street Journal uh I think it was August 20th of 2011 and his thesis at the time was software was no longer supporting the business. Instead, software companies were becoming the business. So now you're a technology company solving an industry problem. And what was interesting in and about this particular time frame, um, Bezos was on the Tonight Show talking about, well, Jay was giving him a hard time. How is it you're a billion-dollar company, but yet you kept losing money every single quarter? And the prior quarter, uh, the new quarter is more lost than the prior loss. And in 1999, we were still able to go to retail stores. We still did, you know, our normal shopping in a way that we had grown up doing. And what was happening then was that he was investing every one of his uh profits into a bigger picture. And Wall Street understood the strategy of what he was doing and they loved him for it. They took care of his stock. The stock kept going up even though they kept logging these losses.

And he had a couple of famous quotes. We are famously unprofitable company. It's a new math. And he was able to sit in that chair and explain it. And although Jay didn't understand, Wall Street understood.

And the other part I wanted to add, Flexera was there. We had a product called InstallShield helping. So longevity. I think Flexera was 12 years old at this point. But nonetheless, what I wanted to show was that we've seen this before and now Gartner comes out and I'm thinking deja vu. Here we go again. But instead of talking about millions and billions, we're talking about trillions.

Now, I couldn't wrap my public school math around trillion. So, I looked it up and I have a very interesting way to look at this. Humans have a hard time with big numbers. But if you took those numbers and converted it to time, so for instance, a million seconds is 11 days. A billion seconds is 32 years. A trillion seconds is 31,700 years.

Now, the investments that are happening are all happening on the AI side. They're building the data centers. They're building the infrastructure. They're improving the models. We're about to have one of the biggest changes I think worldwide. It's going to explode. And it's that pent-up demand. And it reminds me of that moment before of what happened.

So when we see this and we understand what's about to happen, I think the topic for today becomes even more interesting and it starts to put into the relevance what's happening. So as we look to the next step, there is a difference between the different models.

Um model choices and cost is it's a cost decision. the gap between budget and frontier models can be up to 500x in token consumption. So as you're optimizing AI, it's not just about technology, it's also about people who use it. And so I just kind of wanted to call out where we're seeing these massive changes.

Chris, I'll pass to you for this part.

Chris Hayes (9:12 - 15:09)

Thanks. So, in case any of you are not readily familiar with AnglePoint, we're an IT management consulting firm and we service at least 20 of the top fortune 100. So, a lot of what we see in the market um kind of echoes some of the challenges and topics that that Mark was alluding to. We have a huge technology and spend change. I'll stop short of calling it a revolution, but it's a huge sea change here. And here's what we're seeing as we talk to a lot of our strategic and enterprise clients.

The top point is that this is not a, oh, I'm going to go with OpenAI, I'm going to go talk to Anthropic or I'm going to, you know, I'm all in on copilot this and this. There's a proliferation of AI amongst nonAI providers. Every single publisher, if you're, you know, I see some people saying, "Hey, I manage the SAM team or here's where the function sits, this and this." All of the publishers that you're managing in your portfolio, almost all of them these days have all new fancy AI feature, right?

They are enabling these features and along with them new modes of charging you and consuming that technology and having pricing. So it's not just a great I have my handle on anthropic spend.

It's what about Microsoft? What about Salesforce? What about this random SaaS solution with a new fancy AI feature?

Complexity is coming for the whole portfolio. The next big challenge is that we find across our client base just basics visibility.

Where is this stuff? How are we consuming it? Is it on premise? Is it in cloud? Is it in SaaS? Is it a jimble jambled mix of everything? And especially if you're an organization taking a more democratized hey AI for everyone type approach, you are using multiple modes of technology consumption and model consumption. So how to get visibility on that whole state.

The next one is thinking a little bit more about the next couple actually thinking a little bit more about value. Great. We're spending more and more and more and trillions are going to be invested. Amazing huge change. This increase in spend a lot of times for our major clients are not tied to specific use cases. I can't tell you the number of technology executives we talk to who say, "Oh, well AI is my strategy." No, AI is a technology and a solution. This is not a strategy.

What value are you getting back? And it's darn hard to talk about this and to manage when you're not thinking about things. This is akin to a unit economics type um approach for FinOps for cloud cost management when you don't know what your cost per successful outcome is right let's say you're using an AI agent and you're trying to save time per customer call or ticket closed etc. you need to as a technology leader be able to understand and manage to what that cost is per closed ticket or minute that you've saved etc.

Otherwise, you're throwing money at AI and you turn into that, hey, AI is my strategy. I'm spending a lot of money. You need to understand the value.

Not surprisingly, and us ITAM folks who have been at it for years and years, this is a very familiar story. Unfortunately, this is codifying some of these silos. Finance, oh, I need predictability in this thing. I don't want it to go off the rails. How do we know when this is happening? How can you um either uh cloud cost management team or ITAM team or procurement team you know where it sits in your organization how can you provide me in finance a good budgetary number we can live and die by how does this get better and better and info security not surprisingly hey where is our data going where is it being stored for these democratized AI and shadow IT type implementations do we even have a clue so finance and info security are starting to scream very loudly that they are not integral partners in a lot of the deployments that we see across our client base.

And last but not least, where we are, we collectively kind of as a management firm looking at these technology approaches with our clients, we find that adoption is a lot of times in almost a crawl, you know, think crawl walk almost a proof of concept or crawl type phase. So the motus right now is spend as much money as you can enable the business enable that innovation. Yes, we'll think about governance but right now it's an AI gold rush. And this gets us to the next question.

So if we can flash that up, does your organization have an AI cost reduction goal?

So, yes, absolutely. I helped dependent. Oh, wow. Okay. So, that's right off to the races. No, we don't have this at all. We're still in PC mode. Or, yeah, depends. Not sure. Maybe we kind of do.

I'm very curious. Again, our experience in consulting with our many um clients from several different public and private industries. They seem to be in that crawl phase where it's not really, right? So, this is kind of bearing out what I would expect, but let's see what you guys think about this.

Uh, that's a good comment, Christina. You're saying that it's not necessarily a reduction target, but you're putting in place some guard rings, right? Spending caps, etc. This is where finance comes to that conversation. We see that as well.

Mark Bradley (15:10 – 9:11)

Wonder if we should put max token. Max tokening.

Chris Hayes (15:13 – 15:42)

Yeah, token maxing. That's token. Yeah, that's true. And we laugh, but really that's again a lot of what we're seeing is just spend on AI, right? The board and the C-suite folks are hearing spend as much as you can on AI. This is your gold. That's an expensive goal. All right, maybe we'll give it one more minute and uh and we'll go on.

Rob, you're spot on. Maximize the value.

Mark Bradley (15:43)

Yes.

Chris Hayes (15:44 – 22:01)

Okay. So, not surprisingly, it seems that, you know, some of you are saying I, you know, really have no idea or we're partway there. Um, some of you are saying, nope, not at all. We are in the motus of employ enable a business, enable this technology. That's not a bad thing. That's where a lot of again big industry players are. And then some of you the minority but some of you are saying yes absolutely we are getting more mature and getting our hands around this and let's say planning for the future.

All right let's go to this next one. So when we think about this it's a little bit of a busy picture but really in the middle there of let's say the pyramid what would that be? cement or sand or whatever holds the pyramid together, right? We think about the total cost of AI, not just hey tokens. Tokens actually when you think about the tech stack, they make up probably 20 to 25% of your total AI calls. You need processing, you need inference, you need training, you need infrastructure, you need compute, you need storage, etc., etc., etc.

There's a huge tech stack there, right? So you think about those technology value dimensions at the bottom. We're building the pyramid. Your classic and not surprisingly, we have kind of the triangle here. Your classic iron triangle. Do you want something that's cheap? Do you want something that's fast? Or do you want something that's secure? You look on the right there for the strategic decisions and you say, okay, in the business input lane, you're taking your commercial model for the business. You're taking what strategy you have in your operating model and then the technology decisions that are driving and balancing against do we want this fast, do we want it cheap, do we want it secure, etc. You want to be able to take that as a technology executive, as an organization.

Do we want to optimize in this? Do we want to invest more? Do we want to get rid of this? Do we want to dive in? One other thing I'll mention before we go on to the next slide is a healthy AI management posture and portfolio especially since we're a lot of these you know companies in crawl and PC um phases it has a healthy kill rate if AI and your proofs and concept are not delivering value you need to fail fast you need to say right this use case is not driving value it's gone so what this does as you look at the total cost and balance this against each dimension of spend, performance, and risk, you'll climb up that pyramid and drive that outcome, that value that someone was talking about in the chat. That's exactly where we want to be.

So, thinking about the governance for the approach, right? So, this is a little bit more frameworky, but it is intentionally. So, this is an interoperable framework, and you see there, this was clipped from um FinOps Foundation. This is adapted from their inform optimize operate framework. This is interoperable with IT asset management. It is interoperable with uh the Tokenomics foundation. Right? This inform optimize operate idea and framework. One can take and say I can manage ITAM. I can manage SaaS, I can manage cloud, I can manage AI etc. It's important to have that so we can align all of the capabilities and disciplines.

So just very briefly and this is if you see these kind of lighter arrows here this is a continual cycle. This is not a oh I've done inform I'm done. I've done optimize I'm absolutely done. It goes it continuously improves. So at the inform phase relative to AI what we're talking about is that baseline of visibility.

We want to understand where these costs are going. Ideally put in place mechanisms to allocate them back to the business and start thinking about that cost per successful outcome or unit economics of what value is this driving? How do I measure this from a visibility point of view in the optimized phase this will fall out of inform? Oh, we are missing this this this key data point. Well, we want to think about how we get that additional data or how we fix things that we are missing. How we in other words optimize.

And here after the informed baseline, a good next step would be we want to focus on value and reducing cost. Things that are driving value, can we do this cheaper? Things that are not driving value, like I mentioned, get them out, kill them, get them out of the portfolio.

Here you would look at things like right sizing your models, routing to a cheaper model, thinking about architecture, the other stack behind AI or you could think about a little bit more cloud cost management or FinOps type approach. Again, this is interoperable. Thinking about your rates and reserved instances and commits with your hyperscalers for example.

Then after optimize, we want to go to a state of continual operation where we embed that governance. We want to have accountability for the spend. Hey, this is not driving value. Someone is accountable and needs to address this. We want to have good forecasting with again that accountability to say we have an accurate number and it's going to get better and better. And we want to continue to track that value. And then before I hand it back on to Mark, this loops around. We inform and we get more visibility. Oh, is there another issue? Great. We address it and optimize and we run to operate. So this keeps on going and going and going. And by this this interoperable framework we can manage. So at the very end of the day really what the question is how as a technology led organization and I think you know every single organization today is technology-led. how one manages that visibility across all of the domains across ITAM across on-prem across SaaS across server across data center across cloud across AI how do we start with that visibility and how do we think about managing through continuous improvement so Mark I will hand it off to you.

Mark Bradley (22:01 - 30:22)

wow thanks Chris that's awesome it kind of it kind of brought me back to the uh to the ITIL Deming cycle days of continuous improvement and constantly looking for that next thing uh as we go. And it and it parlays perfectly to uh to the question of are you equipped to manage the economics of all six areas of AI spend? So Flexera was looking at this um over the last three four years as it was unfolding uh we could see AI data uh AI utilization start to filter through our different areas of ITAM SaaS and cloud.

And as it started to grow and grow, we quickly said, "Hey, we need a framework to start with recognizing that this has to evolve to help address uh how quickly AI is growing and changing. And so these were the six categories that we saw. And as we started to dig into them, we saw a technology buyer, which is kind of on the left. Then we saw the technology builders that were on the right. And then as Flexera was embracing AI in the same way that many uh software companies were, we started to see areas of opportunity.

In fact, if I double clicked on the AI data cloud section in particular, if I double clicked on Databricks and Snowflake and and BigQuery, Flexera saw that, hey, wait a second, we have to control the AI within these sections. And you probably saw recently where Flexera acquired a company called Chaos Genius. And the and the group within Chaos Genius allowed us to deep dive into that section of AI and find significant savings so well that we said let's acquire this company and apply it to our platform so that our customers can see those significant savings that we found. And as we continue to dig through, each section in itself looks somewhat simple, but then as you start to really dig in, you find that it's a fullcourt press and that we're being infiltrated from all different areas. And through that, there's tremendous opportunity. And what we're seeing is we're seeing um we're seeing startups coming and taking a section of it and saying, "Oh, we have cloud spend and here it is." And then you and then you start to look across the landscape and like well what about this or what about that and what about this and so we look at this as the landscape itself.

I like to say that cloud stress the operating model AI broke it because AI spend is everywhere. It's in apps, models, agents, platform, infrastructure, cloud services, and I imagine it's going to continue uh to evolve, but I think this is a great spot to kind of start with and then as we go into the portfolio. What I want to do here is I want to show how Flexera is perfectly aligned to now take this next transition as our market is breaking out.

Um and as we were and you know from that prior screen as we were building that out luckily we were struggling do we do we try to put it in asset management do we try to put in SaaS do we is it a you know it kind of goes back to that first poll question which group should really manage it and then the question became has anyone heard of Tokenomics you know if in the chat if you're familiar with obviously you're probably familiar with FinOps but are you familiar with Tokenomics and Tokenomics kind of said the same thing that we were seeing that we need to break this out and make this a brand new category. So as we uh as we go through this one of the things within Flexera they have this concept called see everything control everything and for a long time we've been able to meet the market demands let's take IT asset management in the in the chat who remembers their first audit which company was it that you were tapped on the shoulder to go through and begin the audit process mine was Oracle uh I got to learn about Oracle and then that was my indoctrination into ITAM uh to work through that and then as I thought about it remember when software licensing used to be complicated now AI is like hold my tokens I'm about to change everything and so when we look at this one of the key areas that I want to call out and why from an ITAM perspective it's always come down to our discovery, our ability to go out and find, in this case, the platforms

where AI runs SaaS management. When you go over to that section, who remembers when Adobe told Wall Street that we're not going to sell any more boxes of software? We're no longer going to have people installing on their own machines.

We're going to have them use our platform going forward. And they flipped the model because they could see that they had a captive audience. Their customers absolutely loved their software and the only way they could really grow is if they acquired another company. But after you had most of the market, they had to figure out another way. So SaaS management's born, subscription is born, Flexera is there and helping to address that need. And at the same time in parallel you had AWS,

Azure, GCP all creating all these SKUs and all this capability where I don't have to spin up my own data center. I can use a credit card and spin up my own servers and bring them up, bring them down and experiment and do all kinds of craziness. And AI kind of feels like that moment as well. We're all told to embrace this. And so here we are at the next step which is AI cost management.

It's starting to uh evolve and again the competitive advantage digs right into this discovery layer.

Go back here. The next step is the knowledge area. Now the knowledge area. What I like about this is after we've gone across your ecosystem, we've used all of our different discovery capabilities. Then we get into the knowledge layer. So you probably have multiple scanners from all over the world, data centers, on-prem, cloud, etc. You're pulling in all this data. Sometimes there's multiple scanners finding the same data. And that's where the normalization comes in.

We begin to normalize that data, remove the duplicates. We cleanse it. We go into classification.

What's the difference between a desktop and a workstation? Then we go into category subcategory. So now that we've categorize this data and we can tell where it needs to be plugged in to be managed and then becomes the business context um both in business context and enrichment. I can't help but think about there was a movie where Nike was trying to negotiate with Michael Jordan to be the shoe of choice. And there was a line in the movie that kind of connects here was a shoe is just a shoe until Michael stepped into it. Your data is just data until you can do this

business contact and enrichment. So I wanted to bring those two together because that's part of you know it's a lot of data but then when you can bring it up and contextualize it be able to put it in its business context have it enriched then it sets it perfectly for the next layer which gets into the insights our customers tell us all the time Flexera you already have all of our data why don't you derive the insights why don't you tell me what I need to be aware of u bring that up and set that up so that I can just work through that versus you know driving through all the minutiae of data that we can gather.

So this is where we start to do the insights using natural language. We have uh many of the new uh AI dashboards that we have um where you're going to be able to um through a prompt ask it to interrogate your data to help address it and then say build me a chart, build me a graph. um give me a table with uh the following data sets and then as we continue to evolve that you're like you know what I like that report I want that one to be you know automated so I see it on a regular cadence um we have where we've launched uh within ITV some of this capability and then I think it's a fast follow as we work through FinOps then the next layer starts to get into what do we do with this data And this is the part that I find most interesting. Um when we were in San Diego talking uh we had a session out in San Diego and it was the first time I took the uh I didn't have a driver in the Uber seat.

Um I think it was a Waymo I think it was and it was it kind of blew my mind because the cab came to our hotel. We went in. Uh, it knew where we were going. Let me pick my music. Let me uh see what was going on and it drove me to the office and it was autonomous. It was pretty cool. Um, and I thought, well, how does that relate to what I'm doing here? This is the layer and level whereby now that we've gathered all this information, now that we understand what's happening throughout the industry, our next level is to start to build this. Now, this isn't something that uh you can just turn on and hand it over to customers. This is something you need to partner with them and work with them to help build it out, help validate it, help see that is the information that we have, is it answering our question, where are we going with this? So, this autonomous layer, I put it here so that we could at

least talk about it and learn about what would we do uh to help address that.

Now, we've had APIs and MCPs uh internally. Our internal ITAM team has already built MCPs to help address what they're doing. And having what that did was it validated to us that the data we were surfacing gave them the confidence to start to look into that. So, more to come on that. It's a very interesting layer. I'd be curious to see where customers are in terms of that topic and also um if this something that they they'd like to explore.

So today um what I want to set up for and I'm going to turn on a demo here in a little bit but uh the key part is is this introduction to AI cost management. This is the area that we believe is the next phase of our industry. uh by seeing Tokenomics spun up uh we see that there's an industry best practice that's being approached.

Many Flexarans are involved to help with the different practice groups uh to help evolve this and it really starts with our ability to master our discovery layer. Again, we saw a lot of this earlier in our data feeds. Now, we've gone out and really dug into uh some of these key uh discovery points so that we can get the data where it's at. Um, let's go to the next one here.

The other part I wanted to talk just a tiny bit about was AI energy and sustainability. We also recognize that in our communities, we're seeing uh a challenge to the data centers. We're seeing our

water bills and our electric bills going up and we're questioning, you know, why in my neighborhood? And so being able to help track and monitor that. This is something we have today within Flexera. So we have a CCO which is our cloud sustainability uh module. It's an add-on and it gets it grabs all the AI components and then with our partnership with Greenpixie we have the ability to uh to calculate this. And the idea is it to make you aware of it and then one as you're going through uh

regulation two trying to make sure that where you're placing your workloads uh it's in a good spot in terms of uh water and electricity each data center has its grades in terms of uh you know where the grid's located and water availability but the main point is business balancing the business value with cost carbon water electricity So I just want to put a little commercial there for uh for sustainability and then the key part

Chris Hayes (35:23 - 36:16)

maybe really quick mark to sorry to build on that we're seeing that as well in our European facing customers and our globalized customer base increasingly sustainability is another um almost discipline that's going to fold in. So if we have all these data points then we're able to look at um water conservation or um tons of carbon emission etc. and hit those goals as taking the visibility and you know ostensively we're taking visibility and we're creating actionable intelligence to drive valuable outcomes. So if we have this data across the sustainability stack, if you're, you know, a European uh organization thinking about CSRD or other kind of disclosure and regulatory requirements, it just makes sense, right, to use this visibility to drive that. So that's a topic that we're seeing increasingly um in our client base. Just wanted to call that out.

Mark Bradley (36:17 - 39:17)

Uh thank you, Chris. It's um it's a topic I've been passionate about and you know I it's one of those you can meet resistance but you can also see a lot of insights um in that and I um as we've been

digging into it working with many of the European uh companies as well as uh down in Australia we have a few companies we're working with there that see this as a significant driver um one to set themselves up for success down the road.

And then two, trying to do the right thing by uh where they place these loads and the awareness and the impact.

Cool. Great points. And uh we're getting close to the demo. We'll have a live demo here that uh that I'll play. But um the key part here is um very good question. How do you reduce the cost of tokens with very complex agentic AI problems across the corporate data?

You're going to love the demo. We cover that right to the token. Um and one of the key areas we do have um like for instance in the in the AI platforms within AWS, Azure and GCP, we can see at a transaction level where which skew is a AI skew. we can see based on that skew what the GPU is and then within the GPU there's some specs that help us calculate uh the water electricity um that's being used and then based on utilization and then that kind of ties back towards the grid the grid location so there's a whole um what they call greenhouse gas calculation and then there's some ISO certification uh process that that we follow. And that's part of the reason that we partnered with Greenpixie was to dig into the math behind it so that we can present it as it's happening based on your transactions and whatnot.

But the demo uh covers that. And what I wanted to do on this slide was right now based on this framework um we feel that we've got a pretty good model in terms of coverage. uh we have a pretty good understanding as to how to connect the data, how to normalize the data. We talked about uh attributes or contextualization so that we know that uh whose data it is and where the impacts are coming from. And then the last part is now that we have all this, how do you govern it and how do you really drive forward with it? So that's kind of where we come in and we talk about our brand new module AI cost management.

And I think Victoria, do you want me to just click next and you run the demo?

Demo (39:18 – 42:51)

AI adoption is moving fast. New tools, models, agents, and services are showing up across the enterprise, and each one measures consumption differently. Tokens, credits, agent runs, executions.

When every vendor speaks a different language, nobody can answer a simple question. What is AI costing us and who is driving it? Here's where Flexera AI cost management is heading and how it answers that question in four steps.

Connect, normalize, attribute, and govern. It starts with one view. Coming in Q4, the overview brings AI spend together across tools, business units, and consumers. At a glance, you see how much is fully attributed, how much is assigned only to a cost center, and how much is still unmapped. That gap is your starting point.

Step one is to connect. Flexera links to the platforms where AI consumption actually happens. The first connectors include tools like Cursor and Salesforce Agent Force with OpenAI Enterprise, Microsoft 365, and Google Vertex planned for Q4. Each connection keeps data current so the numbers you act on are numbers you can trust.

Step two is normalization. Every vendor bills differently. Seats here, tokens there, credits somewhere else. Flexera keeps each vendor's native unit so nothing gets lost in translation and then puts them on a common footing. Now you can compare cost per active user across tools that never fit in the same spreadsheet.

Step three is attribute. Visibility only matters when someone owns the spend. Every row here is a consumer. Some are people. Many are not. This production API key accounts for $8,400 this month, up 22%. This sales coaching agent, $4,200. Service accounts, agents, projects, and workspaces all get a name alongside the users you would expect. For the first time, every AI dollar has an owner from individual consumers. Flexera rolls up spend to the organization. Coming in Q4, hierarchy mapping and team showback let finance structure the report the way it already works by cost center, then team, then tool, and any spend that can't be traced yet is called out so the gap gets smaller every month.

Step four is governance. Visibility becomes control when someone is watching. Also coming in Q4, budget and spend alerts flag problems while there is still time to act. Here, the OpenAI engineering budget is the critical alert. And every alert comes with a clear next step, not just a red light.

Forecasting closes the loop. Instead of discovering an overrun when the invoice arrives, you see where spend is heading before the month closes. The dash line is the projection. Break it down by tool and the source is clear. Open AI alone is projected to finish about $12,000 over its monthly budget. Looking further ahead, the real question isn't just what AI costs, it's what AI delivers.

Planned for 2027, Flexera adds cost per outcome and return on investment with allocation by application, agent, and model. That's when AI spending stops being a line item and becomes a measurable business investment. Connect, normalize, attribute, govern. AI cost management is more than another dashboard. It's an operating model for scaling AI responsibly. With Flexera, organizations can move from simply paying for AI to understanding, owning, and managing the value it delivers. Take control of AI consumption before it controls the budget.

Mark Bradley (42:52 - 44:42)

I'd be curious uh I'd be curious of your thoughts uh you know populate the chat there if uh this brought some interest to you in terms of uh the way you're looking at it, the way you're trying to manage it uh the way you want to manage it. And um my hope is that you walk away with that there's one that Flexera has been built to help address AI. We've been doing this for a very long time and that this is the next wave. Um and then as this next wave begins to unfold, it's going to be a very interesting journey. And I believe we have uh our ability after this is to uh have you sign up for uh to be able to do a cost management and maturity assessment.

And so the idea behind this is to uh is to partner and explore where you're at, what you're doing, and then seeing how either :

a. we can help or,

b. where it sits in our roadmap journey to help address that.

And I think uh I think that should bring some interesting uh dialogue for both of us. So, with that, I think um I think we get right towards the end here where we uh take questions. I think we're open for questions for a little bit and um take that from there. All right, go ahead and populate. Yeah, the whether it's in the chat, the messages, uh Q&A, we could take a look see if there's any uh any questions here. Um Victoria, you want me to go ahead and just jump right in there? There you are.

Victoria Anderson (44:43 - 45:02)

Hello, I'm here. Okay, we have two new questions that just came in.

Um, from Maria, does this model require enterprise on AI tool side ChatGPT, Claude. Currently, we have a business plan paid by card. What will it work in this scenario?.

Mark Bradley (45:03 - 45:35)

Good question. So, chances are um when you're doing via credit card, we'll probably catch that

through FOSM or Flexera One SaaS Management because if we grab your cards, we'll see that you're using ChatGPT from there. And then the question becomes, do you allow that shadow IT? And if not, can we put them on an internal uh subscription? And then once it's on an internal subscription, then we can grab it through the API.

Victoria Anderson (45:36 – 45:41)

Chris, do you have anything to add to that question? We can move to the next one.

Chris Hayes (45:42 - 45:44)

No, I think we can go on to the next one that's comprehensive.

Victoria Anderson (45:45 - 46:02)

Okay. Uh I think this is like a two-part here. Um what uh what percentage of AI initiatives fail? AI reports 85 of intuitive fail. And then this panel does not discuss failed waste in dollars.

Chris Hayes (46:03 – 47:41)

Um yeah I I would say you know to that to that kind of comment um what I was talking about before is very correct right I think and it's not quite paro at this point right it's not quite that 80/20 rule but really the way to have a healthy management perspective and governance perspective in AI until there is more robust governance is the instances that are failing and it could per industry, per business, right? Some might have a little bit better hit rate and you know it's not going to be 85% etc. The instances of failing quickly you one needs to be able to recognize and needs to be able to get out of the business. Right? So you remember that slide where we had the technology and business outcome on the top that pyramid slide. If you determine this is a divest this is a get rid of this and it's not driving value based on your visibility based on your attribution etc.

That is an okay thing to say look we are all of us are a little bit in the sandbox these days saying hey is this going to work is this going to drive value oh it doesn't fail fast would be a key governance message to say it's okay to remove the items that are not delivering value you will get a low percentage of what's driving value now again until the industry at large gets a little bit more mature you start understanding that value case you start implementing these things a little bit smarter we're all learning So I think it's okay that that is kind of a you know learning as we go from our side as well. Get rid of the stuff that's failing. It is okay to fail fast and divest that and then focus on that value.

Mark, I don't know if you want to add anything.

Mark Bradley (47:42 – 48:57)

No, that's uh you know it takes me back to when cloud was spinning up and um you know you would spin up a server, you'd load up some software, you run through your use cases, sometimes it worked, sometimes you even took it and moved it right to production and then before you know it crashed and burned and then yeah, you had to fail it and start over. I think what's unique about this situation is that we're seeing it from the start. We've seen this before and because of this awareness we know what to look for because history is repeating itself. This time it's AI. 

So um there's going to be many uh failures as we go through. We kind of expect that it's growing uh AI the whole aspect of AI from all the different you know we saw the six different panels. It's growing like crazy. So what worked yesterday may not work tomorrow and tomorrow may be a better way of doing it than what you attempted today. I know it sounds hiberally and crazy but uh it's that fast at the moment. So I think we should expect failure to happen recognize it and then see where we uh shut that off and move on to the next idea.

Chris Hayes (48:58 - 48:59)

Cool.

Victoria Anderson (49:00

Right. We don't have any other questions. probably 30 more seconds if anyone has another one.

Mark Bradley (49:08 - 50:12)

One Q&A. Um you mentioned that Flexera One can optimize AI cost. However, the cost of an LLM interaction. So there's a couple things here. Um I can tell you from the cloud bills, I can see the LLM. I can see the family, the class, the model, and the utilization and the tokens. Um and we're able to see that as we go across the landscape. Um you know, if we're able to plug into your APIs, you know, right into Claude's API or OpenAI's APIs, there's some information that we're able to pull to help answer that. But we're also seeing new um agents in the market to help us um dissect, if you will, uh those transactions and how they're being measured, monitored. Um and I would say to your question is um wait and see. It's still unfolding and we're still evolving the technology stacks to help address that. Cool.

Victoria Anderson (50:13 - 50:20)

It looks like maybe add on to that other question. um was just above that mark that came in.

Mark Bradley (50:21 - 50:21)

Oh, okay.

Victoria Anderson (50:22 – 50:22)

Four minutes later.

Mark Bradley (50:25 - 51:54)

If I can add another question. How does Flexera One calculate and optimize the true end to end?

So, that's kind of a that's a really good question. It's a loaded question. Um my assumption is you're asking end to end in terms of cloud and the answer is because the like as I was mentioning before you saw each of the pillars of ITAM uh SaaS management cloud and now this AI um we've had I consider a fullcourt press from the start and we saw a lot of the cloud transactions that were happening and then it made sense to go make this new section AI Cloud management. And as we're digging into those use cases, uh we're able to pull information from the first three areas.

We're able to pull information based on direct access and the new uh agents that we're creating and building to help us track that. But more so, um as it continues to evolve, we're going to continue to capture it. And I think if you put this you know mark this day on a calendar it's early in the discussion and I think uh a year from now when we come back to this discussion that becomes more clear as well as how the market is addressing it. So I that's more of a stay tuned more to come discussion.

Victoria Anderson (51:55 - 52:03)

Oh, to your comment on the last question, uh, sorry, it's not in terms of cloud. This is a true end to end calculus.

Mark Bradley (52:04 - 53:31)

Okay. Okay. So hopefully as I was going through the various layers of our technology stack uh you saw the ecosystem that we play in. You saw the different discovery methods that we have. You saw that our ability to normalize and enrich that data and then add your contextualization to it. And

then as we went through there it begins to now how do we present it to you and then how do we automate it. So from an end to end we like to look at it from what we call the hybrid estate.

We know that you're going to have technology both you know at the end user laptops monitors we know that in the data center you're going to have technology stacks and then we know in the cloud based on your subscription and then of course now as uh for SaaS so we see your subscriptions there in all those key areas. Um, and I think we had a slide I might have taken out that really talks about the different agents.

Do we want to see if we can go through that if I put that in here? Um, I think it's I took it out. So, I had one that had all the different agents. There was like seven or eight different ways that we uh we addressed that. Um, but if you need more, uh, ping me offline and I'm happy to show and share the slide in more depth for you.

Victoria Anderson (53:32 - 53:47)

Okay, next question. Uh, is it possible to tell what the tokens are used for? Uh, example is like user is using X amount of tokens on Open AI in spreadsheets.

Mark Bradley (53:48 - 54:43)

Um there is there is some um there is some info that we're seeing more and more. Um there's also a concept called token types whether it's a cash token, a fresh token or uh there was there was a

third option there. Um and one of the number one questions to that exact question is can you interrogate the prompt? Can you interrogate what's happening? Um, and I know that there's some there's some legal aspects that we have to consider in different parts of the world in terms of at what layer and level you can get to. Um, and so we can see some things uh more to come and we also have to be aware of the legal aspects of what we can and need to show. So it's a complicated question to a simple question if that makes sense.

Victoria Anderson (54:44 - 54:50)

Yep. Okay, I think that wraps it up.

Mark Bradley (54:51 - 55:07)

Awesome. Awesome. Well, then uh thank you. Thank you everyone. Thank you Chris. Thank you Victoria. Uh great session. We're excited about uh what's coming next for AI where hopefully you're excited. And if you have questions, you want to follow up with us, please do. We are excited to show you where we're at and we'll take it from there.

Victoria Anderson (55:07)

Thank you.

Chris Hayes (55:07)

Thank you very much. Cheers.