ICONIQ State of AI: Bi-Annual Snapshot Report

Live from Schenectady, New York, it's the Metrics Brothers, Growth and CAC.

And I'm Growth, better known as Ray Reich, founder and CEO of Benchmarket.

And I'm CAC, better known as Dave Kellogg, EIR at Balderton Capital, independent consultant, and the author of Kelblog.

And together, we are the Metrix Brothers.

And we go together like tokens in arcades?

Ah, reminds me of going to the arcade when I was a kid, and my dad would always get mad at me because I spent all my cash on tokens. But wait a minute, what do arcades have to do with software?

I don't know, Ray. I don't know. You were supposed to say something about quarters, and I was going to talk about nickels, but that's all right. Um, Let's restart.

Oh, wait a minute. You're saying that when I was a kid, arcade games only cost a nickel, Dave?

That's what I was saying, Ray. I was trying to get a little dig there. But in any case, doesn't matter. That's not the kind of tokens we're talking about today. We'll be talking about the ones whose costs eat into the gross margins of software companies as they integrate more AI into their products.

Well, based upon that topic and theme, it seems like we're going to talk about some of the recent findings in the iconic State of AI, their biannual snapshot report.

Well, however did you know, Ray? That is in fact the report we'll be talking about today, and it's a fine one. But before we dig in, let's hear a word from our sponsor.

The Metrics Brothers is proudly sponsored by BenchmarkIt!, the B2B AI and software industry's largest, broadest, and most contextualized benchmarking index. Dave Kellogg and Ray Groth Wright, have put the Metrics Brothers podcast on hiatus for remainder of 2026. All episodes we have ever recorded are still available on your favorite podcasting app. You can reach Dave at Kellogg and Ray on LinkedIn at Ray Reich. Thank you so much for listening to the Metrics Brothers podcast. Okay, let's do a little bit of an overview. Now, we've covered iconic reports before, and Personally, I'm a big fan of these benchmark reports that ICONIQ actually publishes. This one's a 44-page report they published in January. About 300 software executives across areas of responsibility, including CEOs, CFOs, heads of R&D, heads of AI, heads of product, and even CROs. Don't know how many of each, but it was a cross-section of executives. Pretty good distribution. With the majority of companies being in that $10 million to $99 million range, 85% were North American. And one of the things I liked that they did with this report was they took the information from their Q2 '25 AI Builder Playbook, and then they asked a lot of the same questions and showed some of the trends between Q2 and Q4. So I liked that. And Dave, they had a big statement on page 2 of the report. And it says, we believe the AI market has entered a new phase of maturity. What started as a race to experiment with AI models and launch early AI features has evolved into the challenge of scaling AI into durable and economically sound products. Sounds like a little bit of product maturity to me.

Yeah, I think it's a little bit new in the game to be calling anything mature in AI, but, but I understand what they're saying. We're moving from pure kind of platform foundations into building apps. Which I agree with. My take, my less granular take than yours on the respondents is roughly half below $100 million, roughly half above $100 million. So, just to think about who's answering. I can't remember if you mentioned the growth rates, but what's some 60-something percent are 29% or less growth. So, some 38% are 30% or more. So, it's not quite as growth-oriented as you expect to see. And then, the one that got me, Ray, was 85% North America. I don't know if you mentioned that, but I do a lot of work in Europe and this is very much a— North American report. And back to your prior point, they're going to argue that we're, to me, we're now really kind of getting to the application layer. And that in this case means largely vertical, but not exclusively.

Yeah. In fact, one of the things they did on their transition slides, and that's slide 4, they actually provide a little bit of a commentary on slide 4 here. The commentary is that I don't know why they said nearly 70%, because the next slide 69% of companies are building vertical AI applications, and that 49% of the respondents cite that application layer innovation is their primary differentiator versus the model. And we'll talk more about that in just a minute.

Yeah, almost surprised that's so low, but I hear you. These, as a critique of the report, which we do a little of, but not too much, these are actually pretty verbose. These these slides in between sections, but I almost prefer the data.

I actually prefer the data also, because in slide 4, it's like the number of models that these application software companies are using has increased from 2.8 models 6 months ago to 3.1. So they kind of threw that in there and it's an increase, but I'm not sure what it meant to me. What did mean something to me was slide 5. Where they show that vertical AI applications is 69% of these respondents. That's up from 59%. And then on the same slide, only 33% believe that their primary AI product is more about being a core AI model than applications. Now, once again, that shouldn't surprise me. We know that applications are a much larger target addressable market, but That core in this survey was only 33%.

Yeah, I'm not sure it gets bigger, Tamara. I just think there's more room for diversity the higher up you go on the platform. You know, like all life is carbon-based at some level, right? At the foundation level, we all want to be working with the same tools, right? Like every app runs on, well, these until recently, like an Intel chip, or it runs on a Microsoft operating system or Linux operating system, right? The way tech stacks work is there's room for little entropy at the bottom. So you end up with some very big players and you may have some very big TAMs down there, but what you don't have is a lot of diversity, right? You don't have tons and tons of vendors all doing very interesting and unique things. Those flowers tend to blossom at the vertical AI application layer and at the horizontal AI application layer. So that's what I think is going on. I think we're seeing the Shakeout/consolidation at the lower levels of the stack. And because you use the word TAM in specific, and I'm not sure it's about TAM, because, you know, database companies got pretty big, right? But so they had a plenty big TAM, but it's really about industry structure and kind of profit pools in my mind, but also just about diversity. And you're going to find diversity at the higher levels in the stack.

Yeah, totally agreed. So the next section that I wanted to kind of move to was all about gross margins, because we've heard a lot of buzz in the industry about Hey, AI-native companies have a much lower gross margin. What are we going to do about it? So there's some nice data here. So it is page 17. It's that transition page. They say that companies are spending— companies in this survey— 50%, 57% of their R&D budgets being allocated to AI development. And that's up from 38% just 2 quarters ago. So what this sounds like to me, Dave, and I find interesting is all those legacy SaaS companies are investing more and more in AI products. Their R&D budget now is aligning to that with a 19% increase over the last 2 quarters. Then the second part of that is they're talking about gross margins and that, and we'll go into a slide in just a minute that shows this in '24 versus '25 versus '26. But gross margins are expected to increase all the way to 52% in 2026. When I first read that, it's like, well, I know that means it's increasing for AI products, but most CFOs at SaaS companies aren't going to be happy with a 52% gross margin.

Yeah. Well, whether they're happy or not, it's what's happening. So, I think a couple of things here, Ray. One, for a data guy, you keep picking the slides with the words on them. Don't know what's up with that today, but I like my numbers and tables. Thanks very much. in text, as we see on slide 17. But to your point, you know, gross margins of 52%, it is an improvement. To your point, for AI, it's not bad, right? But relative to SaaS, which is, you know, 80-85%, it's bad. And if we want to really put on our good old days hats, Ray, relative to the halcyon years of on-prem, it was 97-98%, right? So, gross margins have been dropping, but I think, I think, look, my belief, Ray, is you got to look at gross profit dollars. Gross margin, right, it's just gross profit divided by revenue. But gross margin to me is roughly your highest level value add, right? Like if I build a chair and I buy $200 of wood and I sell the chair for $1,000, then I have 1,000 units of revenue and $200 worth of cost of goods sold plus my labor. And that'll produce my gross profit on the chair, right? So, and the more expensive the wood I'm using, right? Just say I have $2,000 wood and I sell a chair for $2,200. Well, gosh, all the value is coming from the wood. You may as well almost take the chair apart and build something yourself, right? So, but so gross margin to me is an approximation of your value add. And the thing, the reason I think, I mean, I have a very short answer to your question, which is just build the rest of the P&L off the gross profit dollars. Just treat gross profit like revenue. And if sales wants to be 30% of revenue, Now, it's got to be 30% of gross profit, right? So, and put concretely, my favorite example, Ray, is Costco. You know what Costco's gross margins are?

I'm guessing 30%?

12%, right? But you know what their gross profit dollars are?

No idea.

$32 billion, right? So, they're doing only 12% margin on $250 billion of sales, right? So, Their value-add is coming from operations in their specific case. But if you look at the gross margins and go, oh, that's a crappy business, you're looking at the wrong thing. You need to look at the gross profit dollars, not simply the percent of revenue.

Yeah. Well, it's interesting, Dave. This is for another episode, but I've been doing a lot of both kind of interviews, some research that COGS is the new CAC because I believe that you're going to have to significantly reduce the amount of money being spent on sales and marketing to justify these new gross margins, i.e., it's going to be a percentage of gross profits, exactly what you said.

So totally in alignment. Yeah, the math is actually quite easy. You just aim at a different line in the P&L, right? Because basically when gross margin approximates 100, you could just say everything is a percent of revenue. Right? Because, well, it's only— in the old days, it only used to be off by 2% or 3%. Then, it was off by 15%. Who cares? But now, it's off by half. And you need to say, hey, if sales and marketing wants to be 30% of something, what's something? And the answer is going to be gross profit dollars. And you're going to do as best you can to reduce COGS. But your P&L is going to have to be built, in my mind, off gross profit, not off revenue, right? Revenue is a luxury you have When gross margins are assumed to be high to super high.

Right. Now, one of the things you said, and I do love charts almost more than anyone I know. So slide 19 has the chart on gross margin and it shows the '24 to '25 to '26 increasing from 41 to 45 to 52%. But then they break it down by that one of the early questions, what's your key innovation lever, application layer. In '26, the application layer innovators say they're going to be at 45% gross margin, and the proprietary model developers are going to be at 59%. But the people who are doing a combo, some differentiation with that UX and intelligence labor into workflows, they're going to be driving 53%. So I thought that was a really interesting slide, Dave.

Yeah, and the segmentation Ray's talking about there, lest it not be clear, Is what is your primary differentiator? Is it A, application layer innovation, i.e., we've got the best app, B, proprietary model development, the opposite, we've got the best model underneath the app, or C, a balance of the two? And Ray talked about that gross margins in all 3 categories have basically increased, and I'd say roughly at the same rate. Over the last 3 years. So I think it's an interesting segmentation. Uh, I didn't get anything startling from the data unless you did, right?

No, nothing was startling other than maybe we need to think about a new technology, software, at least software world where 50% gross margins up to 60%, maybe that is a new reality and it changes the thought as you've already talked about.

Yeah. And by the way, Ray, there is certainly a possibility that, a strong possibility that the foundation models get more efficient over time and that COGS goes down and/or that that model, that market ends up looking more competitive with 3 or 4 different strong players than oligopolistic, easy for me to say, like it would with databases today. I always use databases today because there's only 3 of them. And they're all very profitable because there's only 3 of them. And each actually kind of has its own corner of the market, right? Like SQL Server at the low end, Oracle at the high end. Geez, what's the other one? Then you have all the NoSQL databases, but like, I don't know, MySQL or Postgres on the open source side, right? There just aren't that many alternatives. You're either going to go open source, you're going to use Oracle, which is definitely a mono/oligopolist with the 40% to 50% operating margins to prove it, or you're going to go to the cheap and cheerful Microsoft, but their margins are also 40%. They just do a lot of volume off that at lower price. So, in any case, to me, you're assuming that life doesn't get better in terms of COGS for the use of foundation models. And I'm actually not assuming that. I think we're in the brute force era, and I think the costs are going to go down. Now, usage will also go up, right? Jevons paradox. As the costs go down, usage go up. But I am an optimist that we're going to end up at 60% or 65% gross margins over time. I don't think we're going to get back to the 85%, but I think 65% will be the new 85%. I don't think 50% will be the new 85%, but I don't have a lot of proof for that.

Well, good news is we've got it here in a recorded podcast, Dave.

That's right.

So, hey, let's go to the next section. And it's really about how AI products are forcing a fundamental rethink and transformation of our go-to-market strategy. And I am very pliable when my co-host tells me I need to go to the charts. So I'm going to go to slide 23, Dave. And this is where they're talking about AI pricing strategies. And 2 things I found really interesting about this chart. It shows that 58% of companies are using subscription or platform, kind of a fixed fee. Now, by the way, this question was multi-choice. So it's going to add to more than 100. 35% of the respondents are using some level of consumption-based. But then the third bar, it says 23% are seat-based. So I really like the fact that they actually broke out subscription or a fixed fee for the platform versus it's a seat-based subscription. And then the fourth bar shows outcome-based, and 18% are using some type of outcome-based. Now, I did some pricing benchmarking myself. about 12 months ago. And so did Kyle Poyer. And this was about 8%. So we're seeing a significant increase in outcome-based usage. And I really liked the fact that they then broke outcome-based into what was the outcomes that's tied to pricing. And the number one segment by far was cost savings. And that was 36%. So that's why I really liked this slide.

Yeah, I like this slide too, Ray. I mean, I think I like the fact, just to be contrarian, that 23% are still on seat-based because the VCs and the tech media often like, seat-based is dead, long live outcome-based, right? If you just read TechCrunch, you'd think it would be 0 and 100, right? Like nobody would be seat-based anymore and everybody would be outcome-based. So, the fact of the matter is these things take time to change. I think subscription platform and seat-based, to me, the magic's in the waiting. Is it $1 million to the platform plus $1 a seat? So, it's effectively a containment mechanism. Or is it a nominal platform fee of $1,000 per what? Per entity, plus $20,000 a seat as you might find like a Bloomberg terminal or something, right? So, the question is, what's the weighting? But I still think it's a very useful slide. These transitions happen slowly. It is interesting to see the number one. I mean, basically, costs are more certain than revenues, right? In general, just like VC boards and PE boards.

Right.

Like, if you say you want to increase your Rule of 40 and you say you want to do it by growing faster, they're like, I don't know. If you say you want to do it by cutting price or cutting costs, they're like, yay, right? There's this notion that costs are more certain than revenue. So, it makes sense to me that even in pricing, when you're trying to sell value-add, that costs are more certain than revenue. Am I actually going to increase your sales rate? I don't know. We hope so. But am I going to save you? You can hire 2 SDRs instead of 4. Yeah, that, that we're pretty sure of. So makes sense.

You know, the one thing I'll also share on this slide, slide 23, was the lowest level outcome tied to pricing. It says buyer CSAT 9%. And in my mind, Dave, I'm like, how does a software vendor ensure their CSAT goes down? Tie their pricing to it.

Yeah, that's a weird thing to make an outcome and outcome-based pricing. I'm going to like assume that was a A typo or a brain fart or something like, I don't even understand, like pay me based on how happy you are. Yeah. I'd have to think of a category, maybe for some category like CSAT surveys that works, but in general, I don't know. Yeah.

And then on slide 25, it looks like they asked a question. They didn't provide the question here, but it looks like it was, do you have plans to change your AI pricing in the next 12 months? 37% said yes, they plan to change their pricing. 40% said, I'm not sure, it's too early to tell. But of those 37% who are going to change their pricing, the number one thing is they're going to switch to consumption-based pricing. Now, we don't know if that's partial in a hybrid model or fully, but it was by far the highest. And then the second one, which I didn't quite understand, was at 21%, and they're going to refine their pricing based upon adoption. And I don't know if that one makes sense to you, but I was confused by that one, Dave.

Uh, but I suspect— so look, to me the takeaway is slightly different, Ray. I think 80% of the people are either sure or maybe going to change their pricing, and only 20% are sure they aren't, right, is the way I read this. So for most people, pricing's up in the air, right? And for those who it's up in the air, half of them are saying yes for sure, and half are saying don't know, but still up in the air. Not so not up in the air that we said no, right? So, 80% of people are thinking about changing their pricing model. Makes sense to me. For those who are sure, you've got a third of them saying, a little less than a third saying they're going to go to consumption-based. Then the next answer to that is a fifth of them saying they're going to refine based on adoption. And I have a feeling what that means is we're selling on some other model today and we're going to see who's using it to do what. So, based upon kind of empirical data we get from adoption, We're going to figure out what to change the pricing to. And believe it or not, while it's verbose, that makes a heck of a lot of sense to me because I think it was somewhere else in the survey that, you know, you only want to switch to outcomes if you're sure people want to pay for them. And if you're quite sure you could deliver them, and if you're sure what, you know, what they're worth. And the only way to do that is experimentation right now, right? Say you're selling an AISDR. I think I can deliver this many MQLs. I think they're worth that much. But let's go put this thing out in the field and see what it's actually good at. And maybe it's really good at outbound, just to make up an example. So, we're going to charge a lot for an outbound outcome because it's worth more and harder to get than an inbound outcome where it really is commodity and there's 10 other products fighting me. So, to me, when I see refined based on adoption, I almost translate it perhaps incorrectly to kind of refined based on emerging reality.

Man, that's pretty intuitive. It's like whatever features or whatever value the customer is getting from it, that's what we're going to refine the pricing on. I like that idea.

Yeah, it certainly wouldn't be the first time that you've set out in software, setting out to do one thing, setting out for the product to be used in one way, and it ends up being used in another or being better at some alternative use case.

Yeah. So here we are, about 20-some minutes into it, and we're on slide 27. I only got a couple more I'd love to really highlight, Dave. The next one is slide 27, because I've been trying to really track the trend of how AI products are being sold to kind of reduce their friction. And the question they asked was, do your company use free trials or a proof of concept when selling? And they have like 6 different categories here. One at 30% was we offer a free trial with the usage or cap limits. Then they go into, hey, we offer a proof of concept that is paid. So the customer pays it, that's 29%. Or we offer a proof of concept that is unpaid, i.e., the vendor basically supports it, that's 28%. And then they also have another potential answer, which is we offer free trials, but with time limits, and that was 27%. So it was very interesting to see how companies are charging for these proof of concepts. The one thing I would have liked to see, because I think free trials sometimes are more individual productivity tools and proof of concepts are more enterprise kind of platform solutions. So I'd have liked to see it segmented with something like that.

Yeah. Look, the reason I think they wrote the question as is, is because they were trying to make a question that was answerable by $1 million companies and by $1 billion companies, because that range is supported in their sample. Right? That's who they're surveying. So, I think POC is very much an enterprise notion, right? And free trial, it's not only enterprise, but certainly— sorry, not only consumer, but certainly much more popular in consumer land like Calendly and such than it is in enterprise land. That said, many enterprise solutions do offer free today. So, it's a little bit unclear. So, I agree with your request. But I think they've actually introduced some noise into this question because they're trying to say, I think, if you're really big, do you do free POCs? And either if you're not really big, big and/or if applicable, do you do free trials? So it makes it kind of hard to interpret the data. But my interpretation is, as always in my coarse Kelloggian sense, yeah, about a third of the people do free somethings. And it's low, to be honest. And I don't know if I should add these at all. Again, we're back to data interpretation. Like, you wouldn't say we offer a free trial with usage cap limits and we offer a free trial with time limits. I guess one company could click both of those boxes. But I don't think so. So, I think that would make them kind of additive to the one that says, we do not do free trials or POCs, right?

Right.

And again, this is a select all that could apply. And if you have both an enterprise and a consumer product, right, you could theoretically do free trials on the consumer side with paid POCs on the enterprise side, right? So, you'd check those first 2 boxes. So, there's a survey design question here, Ray. But what I'm getting, the gestalt, Is a third to maybe a half do some sort of free something as applicable to their space. And look, because I'm an enterprise guy, the whole free POC versus paid POC in the enterprise, that is a first religious discussion. It's often used as a sales qualification mechanism. So, it's a sales process discussion. And often, those paid POCs are credited back against an eventual purchase.

Right.

So I would still call it a paid POC, but it's not quite the same thing as a paid POC where it's not credited back against an eventual purchase. So I don't know. I think people are doing these things, but certainly not everybody's doing it. I'd say maybe a third. Yeah.

And then I'm going to wrap up with the— and they on page 30, they go into internal AI usage to increase operational efficiency, productivity. And specifically slide 35 breaks down how are you measuring the impact of your internal AI? And productivity gains was number one. It's actually 83% of companies saying we're measuring it by internal productivity. That's up from 75% 2 quarters ago. And the second one, which really made me feel good, because I love to see ROI measured in increased revenue or decreased expenses, 59% are using cost savings as a key ROI variable, and 39% are using revenue uplift. I guess if you're deploying in marketing or sales, but really good chart of how people are measuring the return on investment for their AI investments, Dave.

Yeah, I think it's a good chart. I think we could have a religious debate about productivity gains, right? Because I think it's a fair thing to measure. And look, the purpose of all productivity gains theoretically convert into either cost savings or revenue uplift, right? That you can't be more productive and either not get more work done or the same work done at less cost or sell more at the same cost. So, it's just a question of what the company chooses to measure, right? So, if we put in an employee break room where people could have lunch so they didn't have to walk 5 blocks to the deli to buy a sandwich, right? That would arguably be a productivity gain. But would it reduce costs potentially because we'd have to hire one less engineer because we get you know, an extra half hour per day out of somebody. And the question to me, for guys like you, Ray, who are like, I want to see the hard numbers, my answer is always, well, I can get you hard numbers, but they're kind of going to be contrived and/or bullshit because I'm going to be multiplying a lunchtime walk times 100 engineers times 4 days a week in the office, right? And I can get you a number. The question is, how real is it? So that's why personally, I'd rather call that employee lunchroom a productivity gain than a revenue uplift or cost savings and not force the mapping. To me, it's just being more honest about what we got. Hey, we think we're saving a half hour a day per engineer because they could eat lunch at their desk as opposed to having to walk to the nearby deli. I don't know.

Dave, I can tell that you were in marketing and I was in sales because they only wanted to see one number from me, sir. And that was, how much revenue did you close?

Well, I'm really good at numbers. I can get you a number. The reason I'm so certain of my viewpoint is I know exactly how to make the number to show you the productivity of the break room. I just don't believe the number I made you.

Yeah. Well, I'm going to wrap up here. In summary, you know, and I sometimes, I really have to protect myself not being kind of almost subjective bias because I do like ICONIQ's reports, but I thought overall this was a great report. I like the fact that they're doing it biannually so we start seeing trend lines. Around some of these things like gross margins, the fact that vertical AI is the number one kind of innovation and really at the business logic layer and that intelligence layer. I like that. And the final thing that I really enjoyed was they had the internal AI usage and productivity. And then it begged the question to me, and we talked about this, COGS versus CAC. Are we going to look at OpEx now as a percentage of gross profit versus of revenue? I really like that. So I like the report, Dave.

Yeah, me too. ICONIQ does good work. I want to just close by talking about slide 40 because we are recording this on Citrini-pocalypse day. So the SaaS-pocalypse has been replaced by the Citrini-pocalypse with a new report they just wrote. Kind of, it was a memo from the future. I think it was 2028 talking about what AI had done to the economy. And I'm going to say in slide 40, it undermines that because slide 40 talks about internal headcount planning and how much AI has impacted your headcount plans. 42% of companies saying no significant impact, 35% saying slight decrease, and 15% saying yes, well, increase in headcount to staff up for AI. Interesting. And then only 7%, yes, significant decrease. So, all these people who are feeding this basic narrative that AI is going to make us so productive, that we're going to fire 3 quarters of our employees, then nobody's going to have jobs. So nobody's going to have purchasing power. Purchasing power. So there's going to be a massive recession. I don't buy into that meme. And at least this one slide in this report doesn't support it either. Certainly not soon. Maybe by the Citrini timeframe, but not in 2027.

Yeah, that's it. We should probably cover that Citrini report because the first line is, oh, welcome to June 2028. We're at 10.2% unemployment rate. It's like, oh, Hey Dave, thank you.

Yeah, it's a great piece of rhetoric. We probably should do an episode on it.

Hey, okay, talk to you later, Dave.

See ya, take care. The Metrics Brothers AI, SaaS, and More is a production of The Metrics Brothers Growth and CAC and a member of the Benchmarkit Podcast Network. By accessing this podcast, you acknowledge that The Metrics Brothers make no warranty, guarantee, or representation as to the accuracy or sufficiency of the information provided or the humor content of the jokes told.

Ray?

Information, opinions, and recommendations presented are, according to our spouses, probably wrong and provided for information purposes only. This podcast should not be considered professional or, for that matter, unprofessional advice. We disclaim any and all liability for any direct, indirect, indirect, misdirect, incidental, special, ordinary, consequential, inconsequential, or other damages arising out of any use of or God help you reliance upon the information presented here. Ray Growthreich is based in New York City and available on X at Ray Reich. Dave Kellogg is based in Silicon Valley and available on X at Kelblog. Schedectity, which is French for unspellable, is not our actual production location. Thanks for listening.