The heir sounds reassuring. Asked whether they will keep their parents’ wealth adviser, they say yes. In EY’s 2025 research, 82 percent gave that comforting answer. Then Aaru asked a population of simulated heirs. Its answer was 43 percent. Industry evidence cited by EY put actual retention at 20 to 30 percent. The artificial respondents were less loyal, and closer to the observed behavior. A family fortune, it seems, does not come with an inherited affection for the person managing it.
- Aaru creates simulated audiences for product, pricing and marketing decisions.
- EY’s wealth-study comparison took one day instead of six months of fieldwork.
- The useful test is whether a model fits your decision, not whether its answers sound human.
The promise and the purchase
That discrepancy gives Aaru its opening. Companies routinely spend money on what customers say they want. Aaru sells a way to examine what they might choose when a price, a competing offer or an inconvenient circumstance enters the picture. The difference can turn a pleasant research presentation into an unpleasant launch.
EY challenged the company to recreate research covering 3,600 affluent investors across more than 30 markets. The blinded comparison produced a median Spearman correlation of 0.90 across 53 single-choice questions. That measures agreement in ranked response patterns. It does not mean nine out of ten future decisions will be predicted correctly. The distinction is less glamorous than the number, and considerably more useful.
The practical change followed later: EY announced an alliance with Aaru in September 2026. A test became a commercial relationship. That progression is more informative than any claim that machines have finally understood human nature.
A population, not a costume
Founded in 2024 by Cameron Fink, Ned Koh and John Kessler, Aaru builds computational populations grounded in real data. Kessler, its chief technology officer, previously conducted simulation research at MIT’s City Science Lab. The company’s expertise combines population modeling, behavioral research and software. Its stated ambition is to let organizations test consequential choices before committing money, time or reputation.
Consider the grocery example in Aaru’s public walkthrough. A fictional retailer wants to know which store-brand claim will persuade shoppers to try something new. It defines audiences, supplies questions and compares price points. The walkthrough separates adventurous buyers, value seekers and habitual shoppers. These are demonstration audiences, not disclosed customer results.
For a product team, the same approach could compare a feature bundle with a cheaper alternative before development begins. For a communications team, it could examine how employees, investors and customers interpret the same announcement. A sentence that comforts shareholders may make employees update their résumés. Testing those readings separately is the point.
Aaru’s technical account emphasizes relationships between traits. Age, income and employment cannot be assigned independently merely because each column looks plausible. A population can match its headline demographics while assembling people whose combinations make little sense. The company says it learns supported relationships, samples coherent profiles and checks both individual effects and population totals. A convincing character is only the beginning.
The consultancies bring the crowd
Aaru’s route into the market runs through organizations already paid to understand customers. In March 2025, Accenture Ventures invested, while Accenture Song said it intended to integrate Lumen, Aaru’s private-sector model, into its AI products and services. The announced applications included product development, marketing and customer strategy.
Interpublic followed with an August 2025 partnership connecting simulation to its audience and marketing capabilities. For an agency, the appeal is straightforward: compare campaign ideas before production and media commitments make revisions expensive. The research becomes part of an existing client relationship, rather than another vendor the client must discover.
The audience can also be investors. In July 2026, Aaru and Breakwater Capital Markets launched a quarterly tracker covering 71 judgments across 40,000 simulated investors. Different mandates can produce different reactions to the same market conditions. Treating investors as one undifferentiated crowd would rather defeat the exercise.
The business sells enterprise software and simulation services through negotiated orders specifying fees and deliverables. Buyers can use the work for their own clients. Its competition includes conventional research agencies and panels, alongside a growing market for synthetic research. Human interviewing retains a distinct advantage: it can discover an experience the model’s inputs have missed.
Redpoint confirmed an $80 million Series A in April 2026. That gives Aaru resources to build and expand; it does not establish a customer’s return on investment. For a buyer, the relevant cost calculation includes the contract, the work of preparing data and the consequences of acting on a mistaken answer. Faster research earns its keep when it changes a decision worth changing.
Read the small print on certainty
Aaru’s September 2026 evaluation broadens the evidence. It reports tests across 2,993 questions, 186 studies and nine industries, with unadjusted mean absolute error of 3.53 percentage points. These are company-reported results. They describe performance on a defined evaluation, rather than a general license to predict any market.
“We can’t claim that AI simulation guarantees what people will do”EY, on its Aaru comparison
A useful pilot starts with one decision, explicit alternatives and a measurable outcome. Reserve historical results the model has not seen. Agree on acceptable error before opening the report. Compare the output with existing research and, when practical, a live experiment. The copyable lesson is the discipline of that test.
Thin data, a poorly represented audience or a changing relationship between past behavior and future choices can weaken the result. Aaru itself describes predictions as conditional on data and assumptions. Its published culture asks employees to revise beliefs when evidence changes. Buyers would do well to demand the same habit from their simulation. An artificial crowd is useful company, provided reality keeps a seat at the table.