Aaru predicts human behavior for Fortune 500s using ground-truth behavioral data — not surveys

platform for payroll benefits and HR built to evolve with modern small and mediumsiz businesses. I nailed it. I didn't say smart. I didn't say modern. I said modern small and mediumsiz businesses. Our next guest is in the reream waiting room. We have Cam Fin, the co-founder and CEO of Arun. Welcome to the show.

What's going on?

How you doing?

Hey, thanks. Uh, thank you guys so much for having me on. You know, it's been a dream since the day I was born to be a TBP. So,

incredible.

I know you're young.

You're the youngest ever guest. You're 11 years old.

Yes.

No.

Coming soon.

Uh, but since it is your first time on the show, introduce yourself and the company.

Yeah. I'm Cam. I'm the co-founder and CEO of ARU. We're a business that predicts human behavior for

uh almost every type of business on the globe. So, we tell people who's going to win elections, what products you're going to purchase. We help people predict the outcome of marketing campaigns. Uh no matter who you are or what type of business you run.

Okay. Predicting elections is interesting because we just went through a huge uh prediction market boom. That financial instrument was one way to harness the wisdom of the crowd. You're sort of doing that through AI and data that you collect or is it all simulated? Walk me through like how do you actually get to a better prediction than what what the state-of-the-art is?

Yeah, I mean it all starts with our idea of rather than training off of what humans say they they do or who they say they are, right? Like we all know polls, focus groups, surveys are fundamentally wrong. There's survey bias, sampling bias, incentive bias, let alone the fact that people lie, right? We train on ground truth behavioral data only. So we're looking at things like credit card purchase history. We're looking at real marketing campaign click-through rates. We're looking at, you know, health insurance information. We're looking at real election results. That is all way more indicative of the actual decisions that people make and thus, you know, far likelier to predict elections better than uh anything else. And and so so is that like if somebody buys at REI, they are more likely to vote a certain way and that factors it like like how how many different ways are you trying to like triangulate and then and then help me out in understanding like how the actual platform works? Is this like effectively you have

uh your own set of data and then you're spinning up a bunch of agents and it's you're basically prompting it to say like pretend you're this person and then this event happens like what is your response like how does it how does it work? Explain it to me like I'm a podcaster. I mean 100% when you start by asking right how how is it working in terms of like what sort of different data are we including your REI example it's like that but at massive scale right we can understand how the differences in the price of eggs in someone's zip code is going to change their likelihood to vote for one candidate or another or to care about some different marketing campaign or another right and then as far as it comes to an individual simulation one simulation is can you know composed of tens of thousands of agents for any audience on the globe, right? So remember, because we're not constricted by what you can reach in a traditional survey and then going and trying to train a model on top of survey responses, we can generate any audience we want, right? So we can generate maybe an audience of social media influencers. We can even generate an audience of podcasters. We have a client in the podcasting business.

Uh probably simulated you, John and Jordy, somewhere in there.

But because of that, and then each agent gets given to a model that we build in house, right? So, it's our core model, our foundation, and then that model is able to take on that profile for all 10, 20, 50, 100,000 members of an audience and simulate their behavior more accurately than anything else on the globe.

Uh, how do you how are you working to build sort of confidence within your customer base? Like this feels like the kind of thing that I like a lot of businesses would be down to try. Uh, and then but how do you how do you actually prove accuracy over time?

The person that they predicted gets elected.

Great question. I mean

I mean that that's a lot of what happened in prediction markets. It's like they got it right and then everyone was like, "Okay, I guess it works."

Well, and well, they get it right most of the time, not all the time. Uh, you know, take the Virginia Attorney General's election uh as a good example. But

um, as you look at us, right, like we've been around for 718 days, right? It's almost two years of ARO. If we didn't work, right, then we wouldn't exist as a business anymore if we weren't able to like predict behavior accurately. Uh but let alone that we do have tons of external validation. We did a really good study with EY on a really tough to reach audience like 3500 individuals uh who are ultra high net worth, right? Can't imagine anyone with a $30 million net worth and more taking a survey. Yeah. Uh and we were able to recreate their behavior even more accurately than the survey, which was pretty cool. So great example there.

Uh how concentrated is the customer base? Because I feel like with prediction markets, it was just like, you know, people had the page bookmarked and they were refreshing it going into the election. Very like general consumer. It feels like there's like a lot of customers maybe. Of course, there's like whales, but uh with you, I imagine that you can walk into like the CEO of Coca-Cola's office and say like I can move the needle for you and that's like a big ticket client. Um what's the shape of the customer base right now and where do you want it to evolve?

Yeah, I mean behavior is everything, right? So when we talk about predicting behavior, we can sell to like every type of business on the globe. We have film studio clients, we have

podcast company clients, we have CPG clients, we have utilities businesses, and we sell to governments, right? So it's super widespread. But I would say the three biggest areas today are consumer, you know, whether that be retail technology or CPG. Yeah. And then we do a lot of work as well for financial services businesses. And then I would add on top of that kind of like the government policy use case. We do a lot of work like emulating the impact of new tax changes.

What about like a selfs serve like you know for small business that might want to put down like $100 a month for a service on a credit card. Is that an option or do you think that will be an option or or do you want to stay in like enterprisy like let's let's actually

they want to help big business get even bigger.

I mean I I'm not going to be upset about that.

Ask curious.

I'm not I'm not I'm not saying we reject small business forever, right? We we we we work with plenty of really cool businesses across the full size just like we work with the massive CPGs we work with spin drift

uh as well but you know in terms of what our core customer base is look I think every business on the globe is going to want to use our someday right like there in five years from now there shouldn't be a decision you're making without using our software

um and I would like that to be as accessible as possible I think it's just a question of making sure that people are primed to use technology as powerful as this

such is this is this is this company somewhat uh and I'm going to give you ample time to push back but somewhat short AGI like I assume a sufficiently advanced uh model from a frontier lab in the future I could talk to it and say hey I need to make a decision on this product launch uh this is my customer base I can feed it some data uh uh predict the outcome for me based on all the data that you have access too. I imagine you guys if you just work harder on collecting the right kind of data could always have a differentiated data source. But how do you uh imagine kind of competing with other you're an intellig I view you as like an intelligence provider right like you guys are predict you're but but more narrow than some of the more general

I've been running every life decision through GPT2 and people say that my behavior is really chaotic but it's been working out so far. Uh well it's actually funny you mentioned that because what we've noticed is the foundation models over time they actually get worse and worse at predicting behavior right like this is something we've seen we used to be a business where we just give like you know survey data to an LLM and then tell an LLM try and predict the behavior based off of this p survey data uh but what you notice is it just doesn't reach the edges right like really consistently it is not going to be able to predict things at the margin and that's a big issue because it's the predictions at the margin that are the most valuable Right? Those are the predictions of what our Fortune 500 CEO is going to do that we can nail that. And so, you know, there is a future where like Claude is going to be able to tell you, yeah, I suspect that American household purchase decision makers might buy this product, but for the biggest decisions on the globe, why would you risk it? And that's why people trust us for the toughest decisions they have.

Makes sense.

Very cool.

Well, thank you so much. Do we have a I think we got there's a gone order.

Is there an official round announcement?

Uh I'll say we're very well capitalized.

Very well capitalized.

Well capitalized.

We we need a soundboard cue for that. Well capitalized.

That'd be fantastic.

Thank you guys so much for having me on Cameron. I'm sure he'll be back on and uh congrats to the whole team on the progress and look forward to uh

uh we don't as well.

We don't really want a sim for TDP. We're here anytime.

We'll figure something out. Yep. We'll