2027 CFO Playbook: Understand and Analyze Your Cloud & AI Spend

0:00Greetings, everyone. I'm excited to welcome Ed Barrow, founder and CEO of Cloud Capital. Ed, welcome to the show. Thanks very much for having me, Ben. Great to be chatting with you yet again. I know we're chatting again and looking forward to this episode because I think this is going to be really relevant for CFOs, controllers, founders, tech leaders as we try to tackle cloud and AI spend, which is the topic du jour. But before we dive into some really great topics here, how about a quick intro of yourself and Cloud Capital? Yeah, I'm Ed Barrow, founder and CEO of Cloud Capital. We really help and focus on helping CFOs understand, forecast, and optimize their compute spend.

0:39As you say, for many of us, this is now the second largest nine-nighter on the P &L, and so it's become such an important thing for CFOs to dig into. My background, I built and scaled a software business and sold that to a private equity fund a few years ago. And then I had the joy of leading strategic finance for a much larger software business. I was keen on the spreadsheets and spent a lot of time looking at very large compute spend and managing to really try to dig into, first of all, just understand why we're spending what we're spending, how can we get greater visibility and predictability over that?

1:10And then making multiple commitments to various cloud providers to help drive your efficiency around that spend and have a lot of battle scars of learnings from that kind of procurement process to really understand how do you get the most out of your cloud investments, how do you get the most out of those hyperscaler relationships? Yeah, and it is kind of crazy now we think of, say, a software P &L, and most of the expenses are people and contractors, wages, taxes, benefits, contractors. And now that's shifting with inference spend and cloud spend going up. And as a CFO, I always thought, all right, we get the cloud bill, talk to engineering, what's going on?

1:50Always a black box for me of what's going on and trust your tech team that they're managing this the right way. But obviously things are changing with inference spend, with all the LLMs out there. And as companies plan for 2027, should we be treating cloud spend and AI spend just like payroll, where we have such scrutiny, we look at those variances, we have greater understanding of what's going on? So do you think 2027, do you think cloud and AI spend will be elevated to that payroll level? Yeah, absolutely. I think that's the case. Those were the good old days. I remember when there was just one line item on the P &L.

2:28I think it's probably called hosting. You have to have that in there. And it stayed fairly stable. And we left our engineering team to manage that. And I think, quite frankly, we were a little bit nervous to engage with the engineering team and ask too many questions. The reality is now for many companies, that is not a small or stable line item. It is the one that is now anywhere from 10 to 20 % of revenue. If you're a traditional software business, it could be easily in that range or significantly higher. if you're an AI-native business, if you're investing extensively in AI. And that is at 10 to 20 % of revenue and growing very rapidly.

3:06That is having a very fundamental impact on your business. It is going to be dramatically changing your gross margin, your operating margin, your cash flow, your runway. You have a very different business than you probably thought you had just 12, 18 months ago. And at that level, it is a real financial impact. And therefore, we really should be treating it like we do other major costs, payroll and subscription and travel, and starting to say that we need to understand what are we spending? What's driving that spend? How do we budget appropriately? What do we do to control that when we miss that forecast?

3:40forecast. And there is, as we'll talk about, I'm sure later, there's real financial risk associated with this as well. It's not just the case that it's a line item that comes in and we have to make sure we have some predictability around it. There's actual financial liabilities that sit behind this spend, which really need strong financial management around them. Yeah, I totally agree. It's no longer just here's the AWS bill and I don't know what's going on. I really actually don't. And yeah, we just need more controls and visibility. And you've done some great research on cloud spend. And in one of those surveys, it said more than one-third of tech companies surveyed spent over 30 % of their revenue on compute.

4:21So 30 % of revenue going out the door in COGS, wherever you code that. So this is the problem facing CFOs today is how do we know as we're baking in AI costs and different cloud costs, how do we know that's going to be a healthy investment versus something that's going to be a long-term problem? Or do we just not know until it plays out? At 30 % of revenue, it's something we've really got to tackle right now. It's pointing to just a very different business model that we now encounter. So as you mentioned, we did a bit of market research earlier this year and produced the cost of compute market research support.

5:02So we spoke to over 100 CFOs at growth stage technology businesses. So we focused on the CFO and on that kind of growth stage, technology, cloud native and AI native businesses and said, okay, let's understand more about what you're spending today, how that's evolved over the last few months and how we cluster that segment that by your business model. To really dig in, because I think a lot of CFOs are lost. They're at sea. They're looking at a number and have no frame of reference to understand, is this good? It's very different than we saw just a few years ago are we overperforming underperforming our peers so we found that to put this cost of compute report together was really helpful to give people a great starting point and you're absolutely right we looked at the businesses that that kind of exist and broadly clustered them into what i class as ai transforming right these are businesses that were more traditional traditional software businesses that have been investing in adding ai capabilities whether that using AI within their company to transform their productivity, their workforce using AI, or actually now increasingly obviously embedding AI into their own software product.

6:10So those are AI transforming, and they're typically spending somewhere between 13 to 20 % of revenue on compute. And yeah, that is up enormously. It was pretty common to see people spending maybe five to seven and a half percent that week at an earlier stage in their evolution, that same breath. So they roughly we're saying it's more than double what it was before then we look at the second cluster who are what you call ai native businesses have probably started the last two three years didn't have that existing software infrastructure in place and they built ai first in in both the way they developed their products and then how they'd embedded ai into their into the very core their value proposition and most of those businesses are now spending somewhere between 30 to 40 percent of revenue on compute on on cloud on on and inference and tokens and that is just a very fundamentally different business i certainly remember the era of doing sas businesses was we're going to spend a bunch of money on sales marketing i mean let's spend a reasonable amount of money building the software but ultimately what that creates is a really nice highly profitable annuity stream we have this recurring revenue stream out into the future 85 80 85 percent margin and actually that meant first of all we were creating the nice profitable cash flow but it also meant we were really clear on how much we could invest in sales and marketing and support to actually generate those those revenue streams if we're now talking about the an era where actually that's not 80 85 percent margin that we're benchmark and 40 50 percent of 40-50 % gross margins, when obviously it's not just compute that goes into your gross margin, there's a bunch of other costs associated.

7:53So if you're spending 30 or 40 % on compute, you're comfortably in that sort of 50 % gross margin range. First of all, it's feeling that there's a much larger tax on every dollar of revenue going out of the door before you can have that impact your bottom line. But also it means you have fundamentally different unit economics when you're looking at what you can invest in sales and marketing and how you have to think about things like pricing as well. It's a completely different business model to build around. Yeah. And there are a couple of things that are important there. And I think one, and I'll put the link in the show notes, it's a great report for founders and CFOs to download the research that you just completed.

8:29But we're in uncomfortable territory for CFOs because before we knew, all right, 78 % gross profit, subscription margins could be 80 to 90%. But it seems like now it's for running 50, 60 % gross profit overall. It seems to me, Ed, more like, all right, we better understand why we're there? Have we optimized? And where is our journey going forward? Maybe it's more relative versus the old benchmarks. Now, I've seen a lot of AI native founders who are like, hey, I'm still shooting for 80 % or I'm at 80 % and it's great, but others are not like that. So do you think it's our margin targets maybe now are more relative and it's more about understanding that AI and compute spend?

9:10Absolutely. And what is the right journey? And it may be completely fine at this stage that you are investing in the latest models. You're started at the premium end of the spectrum with frontier models. And you're trying to really deliver as much value as possible to the end customer. And you are well-funded from a capital perspective. You can afford to operate at low gross margins. Today, as long as I think really the crucial thing is what is your strategy to improve that margin as you scale? because as you get towards being a larger business, you need to be able to get to a profitable position long-term.

9:48That's what you're going to be aiming towards. And so having that strategy of we're prepared to operate at low gross margins. Today, we have the capital. We have the business that can support that and sustain that. But how are we going to transform as we move towards the next phase of business to have a positive margin? And what I would say, and what we've seen a lot, is growth stage, later stage investors are focusing a lot more on gross margin than they ever were before. And see, again, it used to be, let's just assume this business is going to have 80 % margins. If it hasn't, it's going to get there very easily.

10:24And so investors refocused on other metrics like growth and other key operating metrics. Now, scrutiny is much, much more on that gross margin. And whether you're a sensible margin today or not, what is your strategy what is your time frame what are you what are the levers that you go pull over the next 12 to 18 months to transform that margin as well as obviously growing your business at the same time and we know the token bill is obvious the inference spend is obvious but i've been talking a lot about our chart of accounts has to change there's more than just that inference spend we have other infrastructure costs with delivering an ai native product line So what are some of those other big buckets that CFOs need to be aware of that go along with token spend and that cost to actually deliver an AI product?

11:14Absolutely. And this is really important to understand the total cost of AI and the total cost of the investments that you're making. It's the bill beneath the bill that is the one that often gets missed and those arrive. I think there's, first of all, a couple of really important things, which is there are costs associated with developing your products and services. Most of the people listening to this podcast will be reading software businesses. And so they have a process of developing new capabilities in the software. And then they have a separate set of costs associated with delivering that product out to their end customers.

11:50And those sit and should sit in very different places in your P &L. and have very different impact on these kind of core metrics. And in times past when the hosting bill was 1%, 2 % of revenue, sticking it all in cogs and not bothering to split that out probably didn't have a really big impact on how you thought about your business. But now we're talking about this being a very significant portion of your revenue. Being able to break these costs out and understand which of them are actually associated with development, reduction is incredibly important. And this applies to token costs. And then it also applies to your underlying compute costs.

12:28And we'll talk about that a bit more in a second. But first of all, at token costs, there's a bunch of people in your organization, I'm sure today, using ChatJKT, using Claude, whether that's obviously your engineering team, probably spending a lot on those tools to divert your new product capabilities, but also your sales, your marketing, your finance teams will all be using those tools as well. Those are essentially what we're talking about as workforce AI. Those are tools that are helping your team be more productive, deliver more faster. That should be in OPEX. We're appropriately allocated with an R &D or G &A and so on.

13:03And then separately, if you're then also developing a product that then embeds AI capabilities, embeds large language model calls into the actual product, those are the token costs that go into your cost of good software. So thank you. We've got to be able to separate the tokens and separate those. and not just from a chart of accounts perspective, but also because they're just driven in a very, very different way. Workforce costs ultimately are a factor of headcount. The more people you have and the more they use those tools, the more you're going to spend. Production costs in the cost of goods sold, those are a factor of how many people use your wallet and how big enough.

13:39I think in 2026, we were all spending a lot on building products. 2027, 2028, those products are going to get shipped and they're going to scale in usage and you're going to see, unfortunately, that token bill that was really expensive from your engineering team is going to be completely eclipsed by the bill that comes through with all your customers using your product. So the virtual customer. On a token side, it's important. But the other point is exactly the same thing applies to your underlying compute, underlying cloud costs. So you will be incurring development costs on building, testing, developing, staging environments associated with compute.

14:15And again, that should go into your R &D or PX and breaking that out. And that's typically somewhere between 30 % to 40 % of your bill from Amazon or Google, whoever you're using or compute. 30 % to 40 % of that bill is non-production costs, should be a box. Splitting that out, now that bill is so large, splitting that out really moves the needle on gross margin metrics just by correctly allocating that. And then again, as you deploy new capabilities, the other part of your cloud bill is in production costs and that will scale. And what we've seen in the last 12 months is a lot of people are embedding large language models into their software product.

14:54And they're seeing token costs grow. But we're actually seeing compute costs grow faster. And ultimately, that's because, yes, you're going to make a call out to one of these models. You're going to send it some information. You're going to get some information back from it. But you need to then process that data. You need to store that data. You need to present that to the customer, gather whatever their interaction is with your product, feed that back to the model. So one call out to a large language model is surrounded by mod cost from compute CPUs, as we used to talk about. It is data storage, network, egress costs.

15:31And actually, in the cost of compute reset support that we ran, it was actually those costs which went over budget more than token costs. For the CFOs that we did the survey with, we said, what went over budget this year? And it was actually those underlying compute costs, which were ones where I think we're all very distracted on that token deal and didn't think that actually those costs would be growing a lot as well. Yeah, I could definitely see some misses there because of CFOs we didn't even know that existed. In talking about the cloud bill, and say we get this long PDF, 50, 60 pages, and if I'm a CFO or founder looking at this or non-technical founder, where should my eye go?

16:12Are there certain categories that I should be looking at to generally understand this cloud bill? Again, I know your platform deciphers it all, but if I'm looking at that, where should my eye go first to understand the invoice? Yes, we all have seen those wonderful, huge, long PDF invoices from Amazon or Google. when you've seen enough of them you start to wonder whether they intentionally designed and the reality is that's very granular actually often far too complex and naming all of the different technical services and things that are being used so the first and most important thing is actually you should skip the pdf what you need to do is actually get to the cost data underneath it and all of the cloud providers provide reports and access eventually allow you to get to that underlying cost data cloud capital's product we import that automatically to do all the work that we do because that allows you to filter that data and segment that data and slice it and allocate that so the pdf is really not going to get you very far bringing that data whether you're exporting that as a good old csv into a spreadsheet and trying to manually do this yourself or if you're using a tool like cloud capital that automates this whole process what you're looking for then is to segment that by a couple of different things.

17:29So first of all, environment. So most of your costs will be tagged or grouped by the environment that they run in. So is it a development environment? Is it a testing environment? Is it a production environment? So that's a simple fix to separate out that production, non-production, or OPEX and COGS costs. You can also look often that separated by product, by team, by customer. and so further slicing within that to see okay within production here's product a or product b and then finally actually the underlying services and this is where you do need to skill up a little bit if you're going to do this manually because you need to understand what is amazon ec2 versus amazon rds these are different services one's compute one's a database and importantly that means they're going to scale differently right they're going to someone's going to stay up scale with how much use is i was guess that scale with how much data that you're storing and so on And so just having that kind of understanding of environments, splits, kind of OPEX versus COGS, products can be tagged that way, and then the underlying services.

18:31The reality is that's quite a lot of slicing and dicing, and often the tagging isn't exactly perfect. Someone started something and didn't actually allocate it to a particular environment or whereabouts. What we help with is pulling that data in, bringing that in real time into our forecasting platform, and then allocating that on an automated basis to essentially your chart accounts that enable you to understand, actually, I don't need to look at a 70-page PEF. I need to understand what's in COGS, what's in OPEX, what's within that, what's product A versus B, and how is this going to scale? If I add more customers, if I drive usage, if I drive revenue, what's going to happen to this spend over time?

19:09We provide that as a free platform and built really based off what we know of good FP &A, right? So we've all used great FP &A planning tools. We've gone beyond the spreadsheet into those kind of tools. And we took the best practice out of those, but actually applied it to the granularity you need to manage these compute and these token costs. Yeah, interesting. Let's talk forecasting because I guarantee you, and I've done this, all the FP &A forecast models, we're just forecasting hosting as a percent of revenue. Or maybe if we're enterprise, maybe we do it on a customer basis to scale that. And do you think that's going to continue to work in 2027, especially launching, say, AI-native product lines?

19:50Is that old way of forecasting broken, and do we need a better way for 2027? Let's drag right on the spreadsheet. It's probably not going to work for 2027. It probably didn't work for 2026. Yeah, probably not. We probably found ourselves over budget and confused in the first one. So yeah, the easy way has been, let's take our hosting bill and it's X percent of revenue. Let's assume it stays that same percentage or it's been growing 1 % a month. Let's assume it keeps growing 1 % a month. The reality is actually we need to do a better job. And a few months ago, we launched the cloud infrastructure accounting standard.

20:29So we really dug into this topic and said, we need a set of standards for how do we allocate these costs correctly? What should be in cost of goods sold? What should be in OPEX? There are nuances to that. You think about security costs, observability, data that allows us to monitor what's happening. How do we allocate those? We can share the report as well to give people a real standard you can work to that when you think about auditing these costs, you can stand by that and say this is how we should correctly allocate these costs and i think the other part that's crucial therefore is then that allows us to budget more accurately because ultimately what we know isn't going to be true is the bill we got last month that's not going to tell us much about the bill that's going to be coming through next month there's a lot of things that are going to change over the next 12 months for this 2027 budget but if we can split the costs correctly, then we can apply the right drivers and the right context to that.

21:32So production costs, they're going to scale some parallel, but over time with how the business scales. So you can say, okay, as we add revenue, as we add customers, as people use our product more, those costs will scale. The relationship between those is not easy to do. And you can make some broad assumptions, but our platform actually can really work out the relationship between usage and cost or revenue and costs and maybe not policy. Yeah. And that's interesting. Yeah. I think, yeah, we're going to need a few more levers for forecasting. And historically, CFOs have spent a lot of time with the go-to-market team, right?

22:09So much investment in go-to-market, understanding CAC and CAC payback cost of AR. But do you think that's changing where CFOs need to spend a lot more time with the CTO, VP of development, whoever that might be, really to understand? And because, yeah, we spent a lot in sales marketing, now we're spending a lot of cloud and AI that now we need. Do you think we need monthly formal reviews with CFOs sitting down with the CTO? Absolutely. And it's really, again, driven by the fact that this is now a serious line item on the P &L. And it can move rapidly from month to month. And understanding, working with engineering, we can get a much better understanding of what's driving those changes.

22:51And actually, there are two things that change your cloud spend or your compute spend. One is how your customers interact with your products, and that will determine that. And your engineering team will have a much better understanding of it as we add a customer. What really scales and what does? And that's a great conversation to have. But then the second thing that we've had that spend is what are your engineering team actually doing? They will go off and do things. They've got a nice big roadmap over the next 12 months of things that they're going to do. They're going to launch new products.

23:18They're going to update parts of the infrastructure, migrate from one database technology to another. They may actually be sat there and doing some optimizations. Maybe there's some things they think they can probably turn off or adjust and tweak that will save money. Those things will dramatically change your spend. And we often see that's the biggest factor that causes people to go have these nasty surprises is one month the next, the bill comes in and a particular database was twice the cost. and we didn't add twice the number of customers. So we weren't expecting that. That didn't follow with our revenue.

23:54But it's because engineering went off and did something that they'd planned to do. And they found a roadmap and they had a plan to do it. But we just had no understanding of the impact that was going to have on our spend. And so those are the real conversations to happen. Yeah, as we add a customer, what actually changes in our infrastructure? Help us understand that. And when we execute on this roadmap, when we do these upgrades, these migrations, these new in the product innovations, what do we think is going to happen the problem that finance has is they don't understand what those things are problem engineering have is they don't know how to quantify the financial impact so from our platform we help turn those technical initiatives of what your engineering team is saying they're going to do into an actual financial what's that going to cost and when and allow you to collaborate it is the rosetta stone for you to help you to understand that yeah in yeah definitely Definitely, I think CFOs need to spend more time with the tech teams.

24:49But one thing that's, I think, kind of unfamiliar territory for CFOs are cloud commitments. And I remember years ago, you'd see these like, come with our, sign up with us and we'll get your cloud bill down 10%, which I think you told me that no longer works. But tell us just in 2027, now, how do cloud commitments work? Just a little education for CFOs. And is this something that we should be looking at? And when is this applicable? because we're afraid to overcommit, obviously. But what are the do's and don'ts of cloud commitments today? Yeah, it's a very important topic, particularly if you want to spend less on compute.

25:25And I would guess everyone is in that camp. I don't think there's many people who want to spend more if they can avoid it. The way that cloud, the way that compute works is actually fairly simple. Most of us start on a pay-as-you-go model. So we pay by the hour for the servers that we use, the databases that we use, we pay by the token or the inference that we consume. And that's great, right? Provides you a nice level of flexibility. You can try out different technologies very quickly. And it was designed for exactly that. It was about making it as easy as possible to adopt these new technologies and to get those cloud providers embedded within your business and your product.

26:04But all of the cloud providers actually offer really compelling opportunity to save money if you're prepared to switch from paying monthly in arrears on a pay-as-you-go basis to making a long-term commitment that might be a one-year commitment or a three-year commitment it might be to just a specific server or specific database or some overarching commitment to your entire spend and if you're prepared to make that commitment that's great and you can actually get anywhere from 10 to up to 72 percent discount on that kind of pay-as-you-go price point so it's incredibly attractive as an opportunity to make those commitments and get that kind of savings.

26:44And that savings is great, right? Because it doesn't require engineering effort. They don't have to go off and do a bunch of work. It is just purely, I'm making a financial commitment. I'm looking in a certain amount of usage for particular services at the next 12 months or 36 months, and I immediately get a discount on that price. And that really is a financial decision as well. Think about it. This is a financial investment that we're making. That really should be something that finance are very involved with oh yeah the problem of course with these commitments are they are commitments they are financial liabilities at the moment we don't have to report them on our balance sheet as such but they are real financial liabilities and i can't easily get out of them right so i if i make a commitment to a server for the next year or three years and then things change as they inevitably do engineering go off is actually we need to use a different database now this one's better it's going to scale it's going to be more effective for us or actually there's this new technology that's that that's we really want to try out or actually turn out that part of the product didn't go so well we haven't got quite as many customers that we thought we'd have and we're moving the business it's very hard for us all to think about what the business the world is going to be like in 12 months or 36 months and so the big risk is you make a commitment and then you don't need it and that's really painful that's partially painful the business and two ways.

28:06Why is you could just end up on the hook preparing for something you don't use. You're just money going out the door every month for a bunch of servers that you committed to that you don't need anymore. Or use a CFO that has to turn around and say to your engineering team, sorry, I know you'd like to go and use this different technology. I know you'd like to improve the platform, but you can't because we committed to this old database. And so we have to keep using that one. Yeah. Those are really difficult decisions and really difficult. Yeah. Commitments can't be transferred. what can't be transferred now is the risk and so that's what cloud capital is all about so what we do is we come and we underwrite these commitments so we said okay we are going to look at your forecast figure out what you need we are going to make these commitments for you and we are going to take on the financial risk and we're going to guarantee you a saving whether you use it you get a certain discount if you don't use it you're not on the hook then for the part that you're not using so you get the benefit of moving from that page you go to the committed pricing model without the downside without the financial cost of an unused commitment or without the engineering lock-in that would hold you back on innovation and scale and growth and so on and i think as i said this is so critical for finance teams to understand because these are real financial decisions these are not engineering these are not technical decisions these are real financial decisions and come with real financial risk.

29:31And I think a lot of that sits below the surface of that big 70-page PDF. Yeah, and definitely scary for CFOs, but that's where it's like you just, yeah, you just got to understand what we're spending, where the tech team is going, all that stuff. But this next topic is definitely a hot topic with CFOs. If we're delivering, say, LLM-based or token-based product lines, an AI feature or AI product line, how and we need to know now customer usage we didn't really need to know that before but now how much how many tokens are customers using in these features and product lines so can we do that today can we tie product line a customer a to how much they're using so the big thing is so we can understand the margins that we're getting from that customer absolutely absolutely that's a real challenge because you might be making calls to Anthropic Groupus 5.5 for many different reasons, right?

30:24You can be using that for multiple different customers, for different departments and different teams and certainly different features. The great thing is that you can tag those requests. There are simple ways of using different API keys for different teams. We have blunt tools, if you like, of accessing in a different way or running them in a different account so you have the same system same products but in different cloud accounts or different environments that's a crew way to split the bill because we're actually calling in a different way that's helpful if you just do very high level allocation this is my finance team and how about they're using this is our product and what it's using we obviously are going to want to go more granular than that we don't want to know what feature is really driving and what customer is really driving and particularly if we think about pricing models and things like that what's great is you can tag every single request.

31:14So the cloud providers offer what they call kind of inference platforms, Amazon Bedrock, Google Vertex, and they allow precise tagging. You can use a router, so open router or light LLM are the two big ones. And those act as sort of gateway wrappers around those models. And in particular, allow you to say, okay, every time we make a request, we can add a little tag to that to say it's for customer A or for feature A. and that allows you to then tab all that information obviously then you've got to pull it into something that you can allocate those costs and split those costs and that's the platform that we provide but it is possible to do that and it's incredibly powerful for pricing for putting usage caps on for modeling what if actually for this feature we didn't use this model we used a slightly chiquer model so separating those costs out and thinking about potential optimizations potential changes that's what's really powerful so I think a lot of us have seen these kind of big things like open routers being very famous recently and light nlm who talked to your engineering team they will have all heard about that these are really good for kind of monitoring and tagging and then allow you to build a much better allocation and forecasting for everyone yeah yeah and definitely that's expectation day i just want to make sure that was clear that you said that it's yeah we can track at that level we need to track at that level if we have infraspace product lines and also interviewed the founder of trustedrouter.com which another option there too.

32:38And appreciate your insight today. And one final question here for founders, CFOs listening, and they haven't really paid too much attention to their cloud bill other than just seeing a post to QBO. What should their 30, 60, 90 day plan be to be more proactive around cloud and AI spend? I think first of all, we've got to set the right timeline. We've got to all put together a budget for 2027. So we've probably got more like 60 days given the holidays to figure this out i think the first thing to do and let's try and get this done before the middle of november is make sure we have visibility on that total cost of ai so we want to make sure we've got the token spend and the compute spend coming in and that we've got that tagged so the compute spend will already be tagged but we as we're just talking about we need to make sure we've got our token cost tagged and then get that allocated so instead of that being one line in in qbo that we've split that into the production, the non-production costs.

33:38We've made that split so we have the right COGS, we have the right OPEX split. And then I think the second part is don't have the conversation with engineering. Let's think about what's going to drive those COGS, what's going to drive the OPEX. Let's have and start to engage with engineering on a regular writful for reviewing that spend and understanding that spend and have a better plan for 2027 about how to scale. Let's not just drag right on the spreadsheet, but let's actually say, okay, we know what we want to drive the OPEX costs and the COGS. And then the third thing is start to now look at what commitment options are available for you so that you can lock in a better price starting in 2027.

34:19What we really want to do is have that New Year's resolution of let's trim the price, Let's make sure for the areas which are stable enough for us to commit to, that we can make those right commitments and optimize how we start the new year, knowing that those costs are obviously going to start to scale and change as we grow. So first step, visibility. Make sure we have those costs in our hand and we can at least split them between what's going to grow in COGS and what's going to sit in OPEX. And then we can start to apply those drivers and make the best investment decisions that we can. Yeah, that makes a ton of sense.

34:56I'd appreciate your time today. It's clear that what worked before in FP &A and finance is not going to work today with cloud and AI spend. We need better visibility, better forecasting, additional chart of accounts. Have to consider commitments now. Tie AI usage to that customer. So it is a new world for CFOs. So hopefully spending some more time with those tech teams to learn. So, Ed, if listeners would like to learn more about what you're doing at Cloud Capital, where should we send them online? Yeah, absolutely. I suggest the best place is to follow me on LinkedIn. So I'll be there. I'll be sharing a lot more of this benchmark data.

35:30And we'll be talking a lot more about those 2027 budgeting and planning steps and helping people with those. So if you're going into budget season, as we all are, trying to understand what's going to happen to that spend, follow me on LinkedIn. I'll be sharing a lot more research and advice right there. All right, perfect. And we'll put Dear Ural in there so people can follow and follow that journey. And it's so timely. Like you said, Q4 planning for 2027. We've got to get that. I see that FP &A infrastructure in place so we can really forecast this correctly. And thanks. Yeah. Thanks for sharing the insights today.

36:00And thanks for coming on the show. Thanks very much, man. Really appreciate it. Thanks.