Aaru x EY: The 90% Accuracy Partnership Explained (2026)
Aaru's partnership with EY validated synthetic research at 90% correlation with real-world results. What it means for the category, the methodology, and the buyer.
Aaru x EY: The 90% Accuracy Partnership Explained
Aaru is the most-discussed company in AI synthetic research. Approaching unicorn valuation, $50M-plus Series A, multi-agent behavior simulation engine. The most cited validation point: a partnership with EY that produced approximately 90 percent correlation between Aaru's synthetic simulations and real-world research results.
That number matters. The synthetic-research category spent two years arguing about whether the accuracy was real. Aaru and EY shipped a number with a Big Four logo attached. The category took the validation seriously, and the buyer conversation shifted from "does this work" to "where does this fit."
This piece is the explainer: what Aaru actually does, what the EY validation actually shows, what 90 percent correlation means in practice, and how the Aaru pricing model fits the resulting category structure.
What Aaru Actually Does
Aaru is a multi-agent behavior simulation engine. The platform models populations of synthetic agents, each with demographic profiles, psychographic structures, and behavioral histories grounded in panel-data calibration. Researchers query the population by specifying a stimulus (a campaign, a product, a policy intervention, a market shift) and Aaru simulates the population-level response.
The output is closer to a quantitative campaign forecast than a survey result. Distribution of stated reactions, aggregated behavior estimates, and (in deeper engagements) intertemporal dynamics: how attitudes propagate, how attention decays, how segments influence each other.
The market positioning is enterprise-only. Pricing is six-to-seven-figure ACV. Customers are Fortune 500 enterprises, major consulting firms, and large research agencies. Implementation involves significant calibration and integration work, typically weeks to months, often with Aaru's professional-services team.
What the EY Partnership Actually Validated
The headline number: approximately 90 percent correlation between Aaru's synthetic simulation outputs and EY's real-world research results on parallel research questions. The validation was run across a number of studies where EY had both real-respondent results (the human baseline) and Aaru synthetic results (the simulation), and the correlation between the two was measured.
A 90 percent correlation is a strong number for this category. The published silicon-sampling literature (Argyle 2023, Horton 2023, Bisbee 2024, Aher 2023) reports accuracy ranges between 70 and 95 percent on stated-preference and concept-reaction questions, with the upper bound climbing each year as the underlying LLMs improve. Aaru sitting at 90 percent, on the kinds of questions an enterprise research agency cares about, is at the top of the published range.
The validation matters less for what 90 percent means in any single study and more for what it means for the category as a whole. The synthetic-research category needed a Big Four sign-off to move from technical curiosity to procured tool. EY's willingness to publish the correlation under its own brand was that sign-off.
What 90% Correlation Actually Means in Practice
A 90 percent correlation means that if you ran an Aaru simulation against a parallel real-respondent study and plotted the simulation output against the real result, the points would cluster on a line with very high consistency. The simulation moves up when the real result moves up. The simulation flags the same winners, the same losers, the same magnitude differences.
What 90 percent correlation does not mean: the simulation matches the real result exactly. There is variance. There are individual questions where the simulation over- or under-estimates by 10 to 20 percentage points. Across a portfolio of questions, the central tendency is right; on any single question, the simulation is a decision input, not a final answer.
This is the standard reading in synthetic-research methodology: the simulation tells you the direction and the relative magnitude reliably; it does not tell you the absolute single-question number to within statistical-significance precision. Buyers who understand this read the 90 percent correlation as "use this to make 100 decisions where you are mostly right" rather than "use this to replace a single regulatory-grade study."
Aaru Pricing and the Buyer Profile
Aaru does not publish pricing. The category-discussion estimate, consistent with what enterprise procurement teams report, is six-to-seven-figure annual contract value. Total cost of ownership includes professional services for setup, calibration, and ongoing engagement; full programs typically land in the high-six to seven-figure range over a year.
This pricing structure makes sense for the simulation methodology. Multi-agent population modeling at the fidelity Aaru delivers requires significant computational and engineering resources per study. The deep-tech infrastructure is appropriate to the questions Fortune 500 enterprises ask: will this campaign move share at the national level, will this product launch trigger competitive response, will this policy intervention cascade across population segments.
It also defines the buyer profile. Aaru is built for organizations with the budget to support enterprise research-program investments, the procurement timeline to handle multi-month enterprise sales cycles, and specialist research or analytics teams to operate the platform. Mid-market companies, growth-stage startups, and team-level research workflows are not the target buyer.
What This Means for the Synthetic-Research Category
Aaru's EY validation reshaped the category in three ways.
First, it ended the "does this work" debate. After EY's published correlation, no serious enterprise procurement team can dismiss synthetic research as a technical novelty. The category is real, the accuracy is commercial-grade, and the question is no longer whether to use it but where.
Second, it clarified the segmentation. Aaru sits at the deep-tech enterprise end of the category. Below it, in pricing and operational complexity, the category splits into mid-market synthetic-panel platforms (Minds, Synthetic Users, Evidenza) and self-serve generators (a wider field of newer tools). The category is not one product, it is a stack, and Aaru defines the top of the stack.
Third, it accelerated procurement at every tier. A Fortune 500 research team evaluating Aaru also evaluates the mid-market tier as the everyday-use complement. A mid-market team evaluating Minds also evaluates Aaru's case studies as the proof point for the category. The validation lifted procurement velocity across the whole stack.
Where Aaru Wins and Where It Does Not
Aaru wins for population-scale behavior questions. Will this campaign actually move share at the national level. Will this product launch shift the competitive equilibrium. How will a policy intervention propagate across segments. These are the questions multi-agent simulation is built for, and Aaru's EY-validated 90 percent correlation is the strongest credential in the category.
Aaru does not win for routine team-level research. A growth team running fifty messaging tests a quarter does not need a six-figure simulation platform; they need a self-serve synthetic-panel tool that any team member can run in minutes. A product team validating five features per month does not need population-scale simulation; they need conversational personas they can interrogate on demand.
This is why most organizations evaluating both Aaru and the mid-market tier end up using both. Aaru for the flagship simulations, mid-market synthetic-panel tools for the daily research surface.
How Minds Fits the Resulting Stack
Minds is in the mid-market synthetic-panel tier. Self-serve access for any team member, persistent persona library that grows over time, multi-mind panels for distribution analysis, conversational interaction for qualitative reasoning, 80 to 95 percent accuracy on historical benchmarks, GDPR-native German infrastructure.
Pricing: 5 EUR per month per user (Lite) through 30 EUR per month (Premium) and 15,000 EUR per year for Enterprise plans with SSO and DPA. The pricing is intentionally a different order of magnitude from Aaru because the question Minds answers (daily team-level customer intelligence) is a different category from the question Aaru answers (flagship-tier population behavior simulation).
A typical mature stack: Minds for the 90 percent of research that is exploration, message testing, concept testing, and team-level decision support. Aaru for the 10 percent of research that is flagship campaign simulation or strategic market-shift modeling. Real-respondent platforms (Cint, Prolific) for the validation studies at the end of the cycle that need real humans.
Frequently Asked Questions
What is Aaru's valuation?
Aaru is reported to be approaching unicorn valuation following a $50M-plus Series A. The exact figure has not been publicly confirmed; category coverage cites figures in the high hundreds of millions to approximately $1B range as of 2026.
Who are Aaru's customers?
Aaru's customer profile is Fortune 500 enterprises, major consulting firms, and large research agencies. The EY partnership is the most-cited public customer. Other customers have not been publicly disclosed in detail; the enterprise pattern is typical for the deep-tech-research tier.
What does Aaru cost in 2026?
Aaru does not publish pricing. The category-discussion estimate, consistent with what enterprise procurement teams report, is six-to-seven-figure annual contract value, with professional services and integration costs on top. Total cost of ownership for a full program typically lands in the high-six to seven-figure range over a year.
Is the 90 percent EY correlation peer-reviewed?
The 90 percent correlation is a partnership-published figure, not a peer-reviewed publication. Independent published silicon-sampling research (Argyle 2023, Horton 2023, Bisbee 2024) reports accuracy ranges between 70 and 95 percent on stated-preference questions, which is consistent with Aaru's 90 percent figure landing at the upper end of the published range.
What kinds of questions is Aaru best for?
Population-scale behavior questions where the dynamics of audience response matter as much as the individual reactions. Flagship campaign forecasts, strategic market-shift modeling, policy-impact simulation. Questions where the budget at risk justifies an enterprise-grade simulation investment.
What is the alternative to Aaru for mid-market teams?
Mid-market teams typically run a self-serve synthetic-panel platform like Minds for daily research, paired with Cint or a real-respondent panel for validation studies. The annualized cost is two to three orders of magnitude lower than Aaru, and the question shape (individual stated preferences, team-level decisions) is what mid-market synthetic panels are optimized for.