A young American startup tests 200 concepts before submitting six of them to a human panel. For luxury marketing teams, the question is no longer how much data is processed, but who decides, early on, where that scarce attention is directed.
Two hundred. That’s the number of concepts—beverages, packaging, campaigns—that a model can now evaluate before a single human consumer even tastes or sees them. On stage at VivaTech in Paris, Cameron Fink, 20, co-founder and CEO of the American startup Aru, describes this process with the calmness of someone who has already explained it a hundred times: out of two hundred simulated concepts, only six make it to the human jury, and those six, he says, have already been sorted out by the market before they even exist.
Aru’s story can be summed up in a single number that Fink himself mentions: eight hundred thirteen days. That’s how old the company was at the time of the interview. Before that, there was another company—founded when he was fourteen with a future co-founder—that focused on skin cancer screening; it operated for two and a half years. Then came a private message on LinkedIn connecting him with John, whom he didn’t know at the time: a call that was supposed to last thirty minutes ended up lasting one hundred and fifty. John is seventeen years old—not yet old enough to vote in the United States—and oversees the technical operations of a company with thirty-three employees, a number that will exceed forty by the end of the month.
The Argument Against the Poll
Aru’s approach stands in direct contrast to a century of declarative marketing research. Fink cites a figure he attributes to the consumer goods industry: only three out of every ten products launched on store shelves are still there two years later. For him, this failure rate highlights the limitations of surveys, panels, and focus groups—tools that measure what people say they want, not what they actually do.
Aru turns conventional wisdom on its head: rather than training a model on self-reported responses, the company trains it on traces of actual behavior—credit card purchase histories, foot traffic data, music listening habits, and podcasts. The stated goal is to reconstruct, for any given audience worldwide, its likely composition and future decisions—whether regarding a product, a campaign, or an election. Fink claims that his model is 97 to 98 percent accurate in replicating the results of the General Social Survey, the leading U.S. sociological survey—a figure that, at this stage, remains a claim by the founder rather than an independent audit.
The raw data comes from three distinct sources: public data—government statistics, academic research—; data purchased from specialized providers—credit card data, audience metrics, measurements—; and partnerships with associations or small organizations, to whom Aru provides simulations in exchange for the right to train its models on their data. Clients fall into three categories: consumer and retail brands, financial institutions—wealth and asset management—and media and marketing agencies. The political sector, where Aru’s technology reportedly demonstrates its highest electoral accuracy, according to Fink, is no longer among them.
Details
Aru’s business model is based on a precise reallocation of human effort. A consumer goods brand that wants to test six to eight product concepts must, under the traditional approach, submit each one to a human panel—a process that is costly, slow, and statistically unreliable. Aru proposes reversing the order of operations: simulate 200 variants, select only six—those already preselected by the model as the most likely to succeed—and submit only those to human judgment. Human testing isn’t going away; it’s simply moving to a later stage, applied to a sample that the machine has already narrowed down.
What the machine can’t do
It is within this mechanism that the most revealing statement of the interview lies, and it comes from Fink himself, without being prompted: an agent cannot feel the weight of an object in its hands. An agent cannot describe the taste of a product. This admission is significant for a company whose value proposition is based precisely on its ability to replace human judgment in the decision-making process.
For the marketing department of a luxury brand, the challenge is therefore not the abundance of data—that’s already a given, both at Aru and among its competitors. The challenge is determining who decides—even before a human enters the room—what deserves their attention. Aru does not claim to replace the taster, the fit model, or the artistic director’s eye. It claims to decide, on their behalf, what will reach them. This is a shift in power that is more subtle than automation—and likely more far-reaching: scarcity no longer lies in the data itself, but in the act of paying attention—whether or not we choose to delegate that to third-party software.
Fink’s personal journey sheds light—without stating it outright—on the nature of this venture. At age thirteen, using the money from his bar mitzvah, he imported twenty thousand boxes of politically themed colored pencils—Bernie Blue, Trump Tangerine, Liberal Lime, Conservative Crimson—and sold five hundred of them, for about five hundred hours of work and a net profit of two hundred fifty dollars. The anecdote, which he recounts without self-indulgence, paints less a picture of a child prodigy than of a discipline already firmly in place: measuring a result, accepting the failure to meet a sales target, and deriving an exact figure from it. It’s the same logic, applied since then on the scale of an entire population rather than a market stall.
A rarity that changes its nature
One question remains unanswered by the interview. If a machine now decides which six concepts deserve the attention of a committee, a jury, or a creative director, who decides what the machine itself is allowed to reject without any human ever knowing? The scarcity of attention—which has become the new currency with the arrival of these models—requires an arbiter. For now, that arbiter goes by a startup name and has thirty-three employees. In two months, its founder will be legally old enough to buy a drink at an American bar. Power, however, does not seem to be waiting for it to come of age.
Cette publication est également disponible en :
