Brox uses AI to revolutionize the market research industry: building 60k ultra-realistic digital avatars, reducing market research from several weeks to just a few hours

American AI Startup Brox Builds a Database of 60,000 Real “Digital Twins,” Allowing Corporate Clients to Query Within Hours, Replacing Weeks-Long Market Research Processes
(Background: Anthropic releases 10 financial AI agents: reading legal statements and earnings reports, creating Pitchbook reports, automating month-end closing, Claude does it all automatically)
(Additional context: Meta is training “Zuckerberg AI Twins” to communicate directly with 80k employees, next step: launching KOL avatars?)

Table of Contents

Toggle

• 60,000 avatars, each with 300 pages of data
• Speed and cost models have been overhauled
• Who will be replaced by this logic

A market research project that traditionally takes weeks: recruiting respondents, designing questionnaires, running surveys, cleaning data. Brox compresses this process into a few hours.

This AI market research startup founded in 2023 has established 60,000 “digital twins” in the US, UK, Japan, and Turkey, allowing companies to query this virtual population instead of repeatedly recruiting and surveying.

60,000 avatars, each with 300 pages of data

According to VentureBeat, Brox’s core claim is: its avatars are not synthetic data, but 1:1 digital clones of real humans.

The method involves first recruiting real respondents, then conducting deep interviews lasting several hours, alternating between human interviewers and AI interviewers. The interviews don’t just ask about consumer preferences but delve into “decision-driving factors” (the underlying motivations behind why a person makes a certain decision): upbringing, social networks, even marital stability. Each avatar can ultimately accumulate about 300 pages of textual data.

CEO Hamish Brocklebank calls it “the deepest data set per person to date.”

The opposite approach is to generate thousands of virtual respondents using large language models and then ask them questions. Several research institutions and startups are experimenting with these purely synthetic avatars—low cost, fast speed—but Brocklebank’s rebuttal is direct:

“You can create ten thousand purely synthetic digital avatars, but their answers will ultimately normalize into a very narrow distribution (answers clustered around the middle value, lacking diversity), which is completely unlike real human responses.”

In other words, the problem with purely synthetic avatars isn’t that they’re not smart enough, but that they’re too average. The interesting aspects of humans are precisely those statistical outliers and edge cases, which can only be captured through real interviews.

Speed and cost models have been overhauled

Traditional market research companies operate on a business logic: clients have questions → commission a survey company → recruit respondents → conduct the survey → deliver the report. Each project is billed separately, with high time and marginal costs; every time a client wants to ask a new question, the process must be repeated.

Brox’s logic is entirely different. Once the 60,000 avatars are built, enterprise clients buy query permissions, not a report. This is a high-level SaaS (Software as a Service) model—annual subscription, unlimited queries—with starting prices at $100,000 per year, no per-project billing, no single-use options.

Reducing weeks to hours signifies more than just speed; it changes the rhythm of decision-making. In the past, market research cycles were long enough to support strategic decisions: whether to enter a market, whether to adjust pricing. When query cycles are shortened to hours, market research can support product iteration decisions: whether to launch a feature, which ad copy version is more effective.

Brox’s current client list includes Google, Amazon, Sony, Mitsubishi UFJ Financial Group (MUFG), Dentsu, YouTube, and Japan’s TBS. These are large corporations capable of conducting their own user research, but they still choose to pay annually. The reason may not be that they don’t ask questionnaires, but that they can’t get results within hours.

Who will be replaced by this logic

Brox’s target market is an industry heavily reliant on manual labor. Traditional market research firms like Ipsos, Nielsen, YouGov (Brocklebank’s former employer) depend on maintaining large respondent panels and survey execution capabilities. The core of this moat is “scale,” but the cost of scale is slow speed.

When Brox turns 60,000 avatars into a permanent asset, the traditional firms’ execution capabilities shift from advantage to burden. They have thousands of respondents they can call on, but each call takes time and costs money, whereas Brox’s clients can query avatars at nearly zero cost.

More importantly, the direction of marginal benefits is changing. Every new avatar Brox creates increases the platform’s value for all clients—a classic data flywheel effect. As the number of avatars grows from 60,000, the coverage of population segments becomes more precise, enabling companies to ask more specific questions.