Ramp's economist: 35.5% of US businesses are using AI — four times higher than the census estimate — and vibe coding platforms are growing faster than model companies

around 4. 1 because we're at 4. 5. Anyway, uh let's bring in our next economist. We're doing a little econ day here. We got Arazzian from RAMP. You might have heard of him when we did our personnel news segment. Look in the suit. In the suit. Looking great. Uh great to talk to you. Welcome to the show. How you doing?

Thank you guys. I'm a big fan of the show. You guys have done such done such a good job. I love the announcement. So happy to jump into it. Yeah. Yeah. The idea with those is uh you know, very low TAM, but we want to set the Slack account on fire in whatever company uh get gets one of those messages.

Uh but thanks for joining this the show today. Uh we want to talk about the release today. Can you break down uh the announcement, how regularly the these spending reports are, and then we'll go into some of the top takeaways.

Yeah, I mean I'll start by talking about the main question that I get asked a lot is that, you know, what is my job? Why do I do it? Why does RAMP even have an economist? I think it confuses a lot of people. Uh, but I think it really does come from RAMP's product, right?

It's a spend platform, but what RAMP, I think, figured out very effectively is that businesses aren't just looking to spend money. They're looking for help on how to make decisions, uh, how to spend their money most effectively, what the best businesses are doing. Mhm.

And so features like that are always integrated into RAM's product to help businesses make, you know, those kinds of calls. But what, you know, the logic of my role is that we can help a lot more businesses make decisions by making a lot more of that data public.

Data both about how businesses are spending money, but also where these economic trends are going and how your business can can operate uh within them. And so our latest spending report, what we put out this month, gets right at that. We're looking at trends in AI adoption. Our main finding is that uh 35.

5% of American businesses have already adopted uh AI to produce goods and services. That's four and a half times higher than the current US census estimate. Uh and then we also posted a lot of results about general business spend trends, where businesses are spending more money, where they're pulling back.

But the idea here is really to be able to say something really compelling and new about what's going on in the economy. Yeah. and trying to respond to the fact that there's not a lot of public data available about where businesses are spending right now.

Can you uh just jumping right into the first point, can you talk about how you guys came to the 35 and a half% number because I imagine it's pretty hard to triangulate given that you know some you can get AI in a non like AI you know product as an example right like you're using Google Workspace and it it sort of like you know pops up a little like you know magic wand icon and you click it and that's like technically using AI Clippy's coming back.

We've seen Clippy's coming back. We're bullish on Clippy. We're making a PY market for No, but but but then you know different you know differently it's like a company directly spending money on OpenAI or Grock or Anthropic or any of these other things.

So I I'm curious how how you got there because we've been talking we were just talking with Tyler Cowen. What is AI adoption right now? It feels like in in our world it feels like it's 100%. Yep. But, you know, it's also very different to use OpenAI as like a search engine versus using it to develop software. Yeah.

Right. Totally. Yeah. Well, you guys are getting at the key issue with a lot of these conversations about AI is that nobody's really sure where to draw the line. Yeah. About what AI is, if it's going to be integrated into every single tool. Yeah.

And that's similarly, I think, what you get with uh the problem of a lot of AI adoption measurements.

The current US census survey, for example, is just a survey question and asking one person at the company, typically the person who answers the phone, uh whether or not they're they're uh using AI to produce goods and services. And it's kind of a confusing question to answer because you know what does that mean?

Does it mean we're using AI for our manufacturing processes and robots or are we using it to make sure all of our customer service agents have this body of knowledge which was which to answer answer questions? I mean, by the way, no shade to the Census Bureau. They produce a lot of really great work.

Uh, and and shout out to them. They have a lot of really great economists on staff. But I do think we can do a lot better. If we're trying to measure AI adoption, we should probably use data sets that measure actual transaction activity.

Yeah, we have access to the contracts that uh businesses are using to engage with the large model companies.

we can actually see the size of those contracts, how that adoption is changing and particularly how that's changing in different sectors and that's really where you get a lot of the variability right uh I mean you see adoption rates in our measurement for technology are about 65% in the tech sector and that even that sounds kind of low you'd think why is there a tech company that hasn't adopted some kind of AI tool but you know when you look at this actual spend data you also have to remind yourself that businesses are much more diverse than we realize there are large and small companies that are still evaluating whether or not AI makes sense for their business.

There are companies that are still using the free versions. They're c they're employees at many companies that are using their own free that's not going to be captured in our data set either.

And there's other companies that are spending money with Accenture saying like we want to have an AI strategy, but we're going to give you we'll review the deck in Q4 and then decide if AI is right for us. $200 million.

Have you had uh like venture capitalists reach out to you like hey let me get an early read on the on the spend report. I want to know. I want to frontr run it because I feel like every time I see the screenshot go out, I'm like, "Okay, there's the next series. There's the next set of series B's that are going out.

" Yeah, we get that question. I get a message on LinkedIn or Twitter daily. There are probably people listening who are going to message me now or or I haven't responded to. The goal is to make this data public. Yeah, of course. Goal is to make it as available to as many people as possible.

And look, there's the business side of it where we really do want to help businesses make better decisions. Yeah.

But there is also a public service aspect to this kind of work which is look if AI is going to be this transformative technology that's going to change our economy that's going to move over economic indicators that's going to increase labor productivity the very first place we're going to see that is in AI adoption rates and so if you are a policy maker that is you know building policy that's going to try to move these metrics uh we should probably know what that metric is and I just I don't think 8% is the that's the current US census sentiment.

I just don't think that's the right number. Uh but even so, I think we can add something to the public discourse by showing people, well, there's still a lot of room to grow. I mean, we have 15% of, no, 18% of restaurants on our platform are also using AI right now.

And that doesn't mean they're using it to serve people, but they're, you know, when we talk to them, they're using it for a lot of back office tasks, which means 82% of restaurants haven't done that yet. Yeah. Can you re can you restate those high level numbers? census at 8% and then your estimate was what? 30 uh 35.

5% is our estimate across US businesses. That's like significantly higher not growth was higher than previous months. Wow. Saying that we're not at this point yet where AI adoption has plateaued where you know companies have figured out how to implement it yet. Yeah. Yeah.

The other thing I would say is that there what we have observed is there's this sort of learning curve in AI adoption where your first month on an AI platform does not mean that's how you're going to stick to implementing into your operations. You might be evaluating multiple vendors.

We often see in ramp data that businesses are multihoming AI vendors and using open AI but they're also using anthropic and they're using the more verticalized uh services available to them. see a lot of experimentalism where they use one model for one month and they switch to a different vendor for another month.

Uh so all of this stuff is still very early. It's still developing. Yeah. What what is the ramp data showing about the current uh market chaos and tumult around tariffs? Yeah. I mean we it's a really common question that we've gotten.

Uh look, we haven't seen a significant slowdown in if that was the concern, we haven't seen a sort of slowdown in business spend on ramp data yet. That's typically what you'll find in both ramp data, but you'll find that in a lot of public and private data sets as well.

Uh I mean last week we did go out and actually talk to a bunch of businesses. We interviewed 30 businesses in retail and construction and manufacturing to try to figure out how they did feel about tariffs. Uh, and we found a few main points in our findings.

I think we found first of all, there's a lot of policy uncertainty that businesses are concerned about. They're not sure how to respond to the tariffs, how to adjust their operations quite yet.

A lot of them are in this sort of wait andsee mode where they're trying to wait for tariff policy to settle before they make significant changes in their spending decisions. So, that's part of why I think we haven't seen that in our data yet.

though we did talk to several businesses who told us that they were pausing all capital expenditures until they see uh the tariff policies clearly settled. How that's going to affect our data, I do think we're going to need a couple more months to really find it.

But so far, we haven't seen aggregate changes in business spend in both ramp data and in private data sets, public data sets. Is there another leading indicator?

I mean AI might be something I know it's like the most important segment to follow in the spending report but at the same time I imagine that a company that's going into a tumultuous financial market that might not be the first place that they cut. They might actually double down on AI uh and cut back.

But in terms of the type of expenditures that go through uh corporate cards, what's kind of the canary in the coal mine for a recession? you know, in the in in like the Fed data, we usually look at like credit card delinquencies or uh car loan repayments or something like that.

But are there are there specific categories within corporate spend at American companies that you would expect to track more closely as a leading indicator of recession? Yeah. And it's definitely going to depend on the sector. Sure. So, we're going to have some sectors that are a little bit more exposed to tariffs.

They're going to be making those changes a little bit sooner than others. Sure. Uh I mean I generally start by looking at the largest categories of spend by business and thinking about where you're going to see budgeting or movement there. So advertising is one.

Digital advertising for I think is a really great way of measuring how businesses feel about the economy. Yeah. And you know we we rely a lot in the public sphere on these CFO surveys or CEO surveys about business expectations.

Uh but if you really want to know how a business feels about their prospects, you should look at their digital ad spend. Yeah, that makes sense. That is what's going to capture whether or not they see the value in putting ad dollars toward trying to get new customers or increase their revenues.

Uh I mean, we're seeing some changes there. So, we're seeing as last month, year-over-year, 55% of large businesses uh reduced their ad spending or kept it the same. Small businesses, however, but that's going to include startups are increasing their ad spend year-over-year. That's going to be a harder sector to track.

Startups don't always track the overall economy. They can be a little uh a little skewed in that way. But it is worth noting that we are seeing some declines in advertising spend which could suggest some some kind of lack. Yeah, it's just an easy thing to react.

You can go into Facebook for Google and just immediately reduce your budgets and and it's a way to just be a bit more, you know, pragmatic. I think secretly every entrepreneur thinks, am I really getting my money's worth on those Google brand keyword ads?

Like maybe we should do that hold back test and cut those back this month and then uh see what happens to revenue and then eventually they sneak back in and you're like I want to own my keyword. Can you can you talk about spend across the different model providers even at a high level?

I know there had been some interesting data on on deepseek specifically um you know earlier this year but can can you dig in there? Yeah. So open eye is the clear leader in the model companies across business adoption. More than a quarter of businesses on our platform have an active contract with open AI. That's crazy.

Yeah. No. And they were the first mover and it just shows you sort of if they're sort of become synonymous with AI products and companies. Totally. And that's not just coming from the chat chat subscriptions. it's coming from their sort of a API integrations uh down the line too.

Enthropic is a is a is second uh and they are still growing very quickly. Uh but open is a clear leader in all of the markets that we're tracking. Uh though we are still seeing a lot of growth in some of the newcomers to the markets too.

XAI has grown very quickly despite only being a couple months old really you know operating. Uh Gro 3 pushed a lot of their adoption up. uh you're still seeing some amount of experimentalism, some amount of that learning curve from businesses.

Deepseek had a really big spike in January when it really came to fruition that slowed down a lot the following month. Uh and uh and XAI has now sort of leaprogged it, but they're still relatively small players. Yeah.

Is it hard to uh understand AI spend with the hyperscalers because you can go to Microsoft and vend you know GPT4 API into your organization but that's probably just going to show up as Azure on your on your ramp bill if you're even billing it to your corporate card.

You might have you'd be surprised I I really have such a lucky job. I s as for sort of a private sector economist I sit on such a good data set because a lot of those bills they do actually they they produce the line items for API credits and Oh yeah because you get the receipts too it's not just the charge.

Yeah, for deepseek that was originally a challenge because how do you pay deepseek particularly when a lot of the business implementation of deepseek was not necessarily going to go through deepseek the company totally because of the security concerns with using deepseek and a lot of it was going through sort of these hyperscalers or a lot of the other sort of companies that allow you to integrate an AI model that's that's working on their own local servers.

Yeah. But even then we'll see deepseek as the line item. And so we started putting that data out try to get at you know we put an article out last month about whether or not businesses are actually adopting deepseek and high high rates uh so that business can really figure out you know where is this market trending.

I think it's extremely hard to figure out without some data given how quickly uh everything's moving in this industry. What is your tech stack look like for some of these analysis? Are you IPython notebooks? Are you using Julia or Julius?

two different people people can make fun of me, but I do a lot of work in a notebook. Yeah. Uh it's great. I mean, it's it's it's uh skilled skilled very effectively. But the real goal is to put things out quickly. Yeah. And so, you know, we have a lot of notebooks that are just outputting directly to our website.

Oh, very. And uh and similarly with the benchmark report that we just released. Yeah. Uh the goal, I mean, you'll note that it goes all the way through March data.

And so the goal is to really provide people with the earliest available data as soon as possible because particularly in this news environment right now, yeah, things can just become immediately outdated. Yeah, Jordy, um, how what what is act what is AI adoption look like across different teams?

because I can imagine at some of these larger companies you have an engineering team that's using you know uh Devon or Anthropic or Cursor and then you have you know the marketing team which is using some like app layer company or they're just using open AI to generate images um how how are you seeing it kind of roll out well AI adoption is generally going to be higher at the large companies than the small companies and we think it's for that specific reason large companies have many more teams many of them using AI products and services, many of them aren't.

And then those large teams, those teams at those large companies will then evangelize the products they're using throughout the company and increase the adoption rate that way.

Um, I mean, the model companies are still growing at a very fast rate, but one thing we saw this past month was that the sort of coding platforms, the quote unquote vibe coding platforms, the cursors of the world, the lovable, uh, those are currently growing faster than the model companies themselves.

um seeing growth rates in the 40s right now. And so uh you know people talk about you know whether or not you know any of this growth is sustainable. For what it's worth businesses are clearly acquiring value from them. Yeah.

And uh and and the the way these products are developed I think is it's going to be really interesting to see how they get integrated. I mean canvas also Canva just announced its version of it too. Yeah. Uh that was also one of our fastest growing vendors last month, but it wasn't because of an AI coding thing.

It was just because it's a really fast growing vendor that's implementing AI across all of its features. Yeah. Have you gotten push back from any of these companies saying like, "Hey, maybe don't leak our financials or like uh you know, it's like you're sharing too much insight. " Actually, completely anonymous.

So, I don't think it's Yeah, it's anonymous, but again, you're naming like a specific company. And for most of the companies, it looks amazing and it's like, "Wow, they're growing so fast. " But uh you mean on the you mean on the customer side? Yeah. Yeah. Yeah. Yeah.

Like like you get a you get an email from OpenAI saying like hey like because of secondary is being repric know that we're growing at 30% instead of 40%. Or this is off and we're actually growing bigger and non-ramp customers and so your data is low. Yeah.

But what's interesting about it is that we've had customers that we reported on reach out to us. Yeah. For more data on what we're reporting on. You can imagine that.

What's really great about having this data set at RAMP is that in in some ways is you know the most comprehensive data set you can find about overall spending by businesses and so you can say large AI company for example knows how much spending is happening on its platform but they don't necessarily know how much spending is happening with the other AI model companies.

Yep. Desire is really to have some kind of data that estimates that sort of total market view. Yeah. And the the logic of this role again is that businesses, people, but everyone businesses really make better decisions when they have access to better information.

And and I see my role as trying to put that kind of information out so that companies can make decisions that are right for them and it's available to everybody for that reason. Are there are there CFOs on RAMP who come to you uh in a less positive light?

Instead of saying, "Oh, what AI model should I be adopting or paying up for? Instead, they're like, give me the what should I be cutting? Like, where are the biggest places to save money instead? " Or, yeah, they're like they're looking at like declines and then they're saying, "Oh, I'm going to target that.

" If everyone's if everyone's unsubscribing from, I don't know, Netflix in their organization or something, uh, they're going to try and do that internally to save money. They do.

And that's the fundamental goal of this kind of work is not just to show you what to buy, but it's to show you first of all what to cut, what other businesses are cutting, how can you benchmark your business relative to how others are performing. Yeah.

The goal is that if you have access to this kind of data through RAMP's product, but also through the kind of research that we put out, you might get a better picture about, you know, what it really means to be a high performing business, how to run most efficiently, uh, which vendors to use and select, and which vendors other people in your industry are using, and then which ones they're not using.

Yeah. One thing you pulled out was that Tik Tok's ad platform leads in year-over-year spend growth, signaling advertisers have not been deferred by the prospect of a ban. My question when I saw this was, is it possible that people are like, "Hey, we think this platform could get banned.

Let's basically try to acquire every poss. " Yeah. And I think what's showing you is why there's going to be such a big bidding war for this kind of platform. Sure. Right. The ad spend just continues to grow. Their user base, I imagine, still continues to grow. Yeah.

uh it is really one of those unique platforms that is able to both attract a lot of visitors but also keep them on the platform for a long time. It's made ads on short form video kind of work well and it's still growing a lot.

I mean I I I'm I'm really interested in seeing how different how how this sort of Tik Tok Tik Tok fight plays out in the business side. Yeah, makes a lot of sense. Poly market has the recession at 51%. What is uh what does your gut tell you as an economist? How we looking?

My real goal here is just to put out data to help make their appropriate guesses. I mean, I do think there are a lot of headwinds what we're hearing from businesses. So, I hope those get resolved. Yeah.

But, um hopefully I I I hope people are able to make the decisions that make sense for their businesses based on what we're putting out. Yeah. Makes sense. Do you have any do you have any uh thoughts on AI can has the potential to be deflationary in some ways?

Let's take like an e-commerce business who, you know, might spend $50,000 a month producing content to make ads.

And then if if suddenly they only need to spend like $5,000 a month because they're able to do one shoot and then just like generate iterations of it or or maybe they're eventually able to, you know, just generate all the advertising content that they need.

in theory they would just take that budget and spend it to acquire more customers or spend it to launch new markets or things like that.

How do you think about um how do you think like long term about you know as AI drives efficiency where those dollars flow right because budget if if budget is being you know sort of reallocated because if AI can do something a lot more you know inexpensively the budget will naturally flow to where it can sort of generate the highest possible return and that could be as simple as like oh you know we we spend $20,000 a month for this CPA firm with AI we now only need to spend $5,000 a month because they don't need as many people, you know, working on on it or something like that.

But then again, you're going to want to sort of reallocate that spend to continue to grow the business. Do you have any do you have any sort of highlevel thoughts on if it's again if it's going to okay, let's just run more ads or or or things like that?

Well, I think that re real reallocation will happen and I think that's the reason why we probably won't see deflationary pressure. Uh I look I'm of the school of thought that AI will be a very transformative technology. I think it's going to change a lot of knowledge work.

I think it's going to reduce costs for a lot of businesses. But at the same time I am also fairly realistic about the fact that many parts of our economy just will never be touched by something like AI.

I mean it will be touched in some ways right like me going to a restaurant I imagine that restaurant will probably have much easier back office automation because of something like AI.

But, you know, me sitting down across from someone at a restaurant is is not a which is a lot of most people's day-to-day spending, right? It's just going to businesses on your small business street. You know, most of that is not going to be touched or moved.

And uh so I think we'll see a lot of parts of a lot of parts of the economy continue to grow. I think knowledge work is going to change a lot. Uh and I'm bullish about the impact of AI. Uh but uh you know I I do think much of our day-to-day spending is going to stay fairly similar as consumers at least. Yeah. Yeah.

Uh have you thought about uh any like more niche industry benchmarking?

And I remember uh it was always useful to know how other companies in the same sector, how much are they spending on their legal bills or how much are they spending on e-commerce development and kind of have a rules of the road and when you I I I did some back of the envelope with some rival companies at one point and found that some of the companies were way way overspending.

Um so we should be clearly under that. I could imagine that type of uh benchmarking being useful just to kind of allocate budgets to say, hey, realistically, you know, the IT budget at a company of this size in this industry really shouldn't be more than like 5% of our costs or operating costs, anything like that.

Have you been able to produce anything like that or are you thinking about that? We're working on that right now. I mean, the goal is to be able to see whether or not you're a small construction firm in can Kansas City. Yeah. And your current cost structure can compare that to other construction firms in Kansas. Yeah.

And and see how you're sort of tracking relative them. Uh I mean I think that's going to be the real frontier that helps businesses make those kinds of decisions for them. Totally.

Find out where they're overindexed and then when they're under indexed and then to try to track their performance following decisions and just make like a choice.

Like there are times when you want to invest more than your competitors in a certain area, but there's other times when you want to be below them and maybe that's your competitive edge. Uh well, thanks so much for joining. This was a fantastic conversation. Let's make it a quarterly thing. Yeah.

I can't wait for the next one. And thank you for coming uh coming prepared. Yeah, you look fantastic. I know people people always think it's weird to everyone at work is commenting like, "Oh, I can't believe you wore a suit today. " It's like this has been the normal component of men's wear for centuries.

It's just only weird. You work in you work in finance finance. Finance tech. So, dress the part. Great to see you. We'll talk soon. Bye. Next up, we have And I'm still I'm still thinking about how um Tyler Cowan just came in. He's like, "AGI is here in two days.

" And uh not super bullish on robots, but uh yeah, AGI is here. AGI is here, but it's not necessarily going to grow the economy as fast as we want, but still here. It's I mean, it's a good take because it's it's the it's the same as the touring test. Like, we passed the touring test and it was kind of a nothing burger.

It was like, "Okay, yeah, you can chat now, but you probably still want to watch a live stream, especially when you have the founder of California Forever in the building.