Aaru, a two-year-old AI startup, simulated all of those voters\u2014agents built from the kind of data the company collects, like credit card purchasing history, food-delivery orders, and demographic records\u2014and tried to predict how the votes would be cast. The simulation came within 2,000 votes of the final count.<\/p>\n
Aaru\u2019s 21-year-old founder and president Ned Koh doesn\u2019t care if you believe that. He said at Fortune\u2019s Brainstorm Tech conference<\/a> recently that while he lacks years of experience in enterprise sales and market research, he\u2019s let the company speak for itself.<\/p>\n \u201cOur entire sales methodology is to go to a business and say, \u2018Do not trust us. Do not trust our model,\u2019\u201d Koh said. \u201cIn fact, I hope you\u2019re skeptical, because it means you\u2019ll be a better customer in the long run.\u201d<\/p>\n Koh cofounded Aaru in March 2024, at 19, with Cameron Fink, then 18, and John Kessler, then 15 (and too young to sit on the board). The company organizes thousands of AI agents into statistically representative populations\u2014each agent assigned an age, an income, a zip code, a gender, among other things\u2014and surveys them in place of human respondents.<\/p>\n The company begs a somewhat uncomfortable question: Are humans so pliable to conditioning, that you can predict their behavior using second-order data? The company builds AI to train and benchmark on what it calls \u201coutcomes\u201d: election results, purchase data, the things people do rather than the things they say.<\/p>\n The outcomes are pretty spot on, though not always in the way clients anticipate. Aaru counts on clients\u2019 incredulousness for the success of this framework. Prospects hand over an old survey\u2014the people they polled, the questions they asked\u2014and Aaru reruns it blind against the known result.<\/p>\n The proof he cited: Ernst & Young, a partner of Aaru, which spent six months surveying 3,600 people across 30 countries for its global wealth study<\/a>. One question asked whether respondents would keep their parents\u2019 wealth manager after they passed away, and 82% said yes. However, real world retention is actually closer to 20% to 30%, Koh said. Aaru reran the entire survey blind\u2014no humans\u2014and its simulated respondents said roughly 40%. So the simulated people were closer to the truth than the real ones.<\/p>\n Eliminating people bias<\/p>\n Aaru\u2019s success hinges upon people\u2019s own bias toward themselves, Koh argued. Intentions are often different from behaviors, so when surveys ask people what actions they plan to take, it often deviates from what they actually do. Scientists call this the \u201cintention-behavior gap<\/a>,\u201d and have used the concept to explain why individuals don\u2019t go to the gym despite their New Year\u2019s resolutions, or why people make impulsive purchases even after planning their monthly budget.<\/p>\n Koh puts it more simply. \u201cPeople lie,\u201d he said. \u201cIf you ask someone how much they drink alcohol, they\u2019re never going to be honest.\u201d<\/p>\n Koh has seen this play out in the top demographic Aaru is asked to survey (though he did not disclose which clients requested this information, or what results would be used for): GLP-1 users, who he claims tend to lie by omission.\u00a0<\/p>\n \u201cThey\u2019re not even telling their family they\u2019re on these medicines,\u201d he said. \u201cThey\u2019re not going to tell your study.\u201d<\/p>\n Investors and clients are buying into Aaru\u2019s methodology, including EY<\/a>, Accenture<\/a>, Interpublic Group<\/a>, McDonald\u2019s, Boston Beer<\/a>, A24, and Bayer<\/a>. The startup recently worked with Spindrift on product innovation, which led to the sparkling water company launching a still tea drink.<\/p>\n \u201cThat\u2019s a totally separate category,\u201d Koh said. \u201cThey\u2019ve never done that before, but we can be confident in our predictions of that space, because we can be confident that our models are good at thinking through the profiles.\u201d<\/p>\n In December, Aaru raised a Series A led by Redpoint Ventures at a $1 billion headline valuation<\/a>\u2014on annual recurring revenue still under $10 million, according to TechCrunch. The round used a multi-tier structure: some investors paid the $1 billion price, and others paid less, putting the blended valuation below the headline number. Total raised to date is roughly $88 million<\/a>, per PitchBook.\u00a0<\/p>\n