Synthetic Audiences, $200 Million, and Creative Risk

Simile, which builds AI simulations of real consumers for companies to survey, raised $200 million at a $2 billion valuation in July, five months after its Series A funding round.

Adobe offers marketers a related tool through Brand Intelligence, where they can test creative work and messaging against AI-generated customer profiles.

The essay frames the pressure on marketing departments: produce more, spend less and improve the work. Its efficiency claim is specific: AI can test ideas faster and at a fraction of conventional research cost.

By Marcus Bennett

What do synthetic audiences actually measure?

Synthetic audiences (simulated groups of consumers) can use customer records, interviews, demographic profiles and other data. A team can present its work, examine predicted responses, revise and try again without recruiting another human panel.

For a company that understands its customers and validates predictions against actual behavior, the approach can be useful. Marketing often improves established work; avoiding an expensive mistake or strengthening a weak campaign is a respectable result.

Beyond efficiency, synthetic audiences provide a defensible reason to proceed. The phrase “We tested it, and this is what customers preferred” carries more authority than “we think this is a good idea,” even when those customers are simulated.

As approval tools become part of creative work, they will influence what people consider worth proposing. Teams may grow better at anticipating an audience while losing the ability to surprise it, according to the essay’s central warning.

How does synthetic research change the learning loop?

Steven Waterhouse argues in Learning Machines All the Way Down that AI lowers the cost of trying things, examining results and adjusting. Companies built around that sequence can feed each lesson into the next decision, improving and accelerating as they go.

Marketing has always followed part of that sequence: form a judgment, put the work in front of people and discover how sound it was. Research prepares, experience informs interpretation, and the audience retains the right to surprise us. United States Air Force Colonel John Boyd formalized the process as the OODA loop (Observe, Orient, Decide, Act), then repeat.

Orient draws on experience, culture and new information. A decision functions as a hypothesis; an action becomes a test, with results feeding back into understanding.

Synthetic audiences can help teams work through that understanding before they commit resources. They may expose an unexamined assumption or suggest an interpretation the team missed. Another simulated response, however, adds no new observation of actual human behavior.

The distinction matters when revision exists mainly to improve prediction. A team can run through hundreds of iterations and become increasingly confident in a campaign without learning whether people share that confidence. The process accelerates. The uncertainty remains.

Boyd also wrote that “adaptability implies variety and rapidity” (a broad range of possible responses and fast adaptation), linking both to unpredictability and coping with unforeseen circumstances. Applied to creative work, that requires preserving possibilities broad enough to attempt the unfamiliar.

Deadline and budget pressure narrow that range. The idea everyone already understands has an immediate advantage over an unfamiliar direction with no evidence yet.

Why can better information produce repetitive work?

David Perell described the cultural risk on TBPN, crediting Derek Thompson with the observation that Hollywood has succumbed to the “Moneyball-ification of our world.” In his account, Hollywood became boring partly because it became intelligent about risk.

Thompson’s diagnosis was blunt: “The problem is you end up getting Spider-Man 74 or Lord of the Rings 7 rather than a new show.”

The complaint has commercial logic because an established franchise supplies an audience, a performance record and some knowledge of expectations. Financing another installment can be sensible, and audiences can take real pleasure in returning to a familiar world.

Yet the first installment had to accept uncertainty before an audience existed. A studio can make a succession of reasonable decisions and still create a culture in which unfamiliar stories struggle to get made. We routinely approve ideas we are progressively less excited about.

Better information does not have to produce that outcome. Understanding risk can encourage more informed bets, especially where teams exaggerated the downside. But information alone does not determine an organization’s appetite for uncertainty.

Synthetic research moves that calculation into daily creative work. Teams can consult an audience earlier, more frequently and on increasingly small decisions. This can improve campaigns, yet it can also keep anticipated reactions constantly present.

People may begin accommodating that reaction before developing an original idea of their own. The model’s response becomes part of the brief, even without an explicit instruction to obey it.

How does the audience train its users?

The source describes a second learning loop. A favorable score helps secure approval, so team members naturally notice which proposals move forward and which require a lengthy defense. They learn what their environment rewards before the model issues an instruction.

An idea gets adjusted before submission, an awkward element comes out and an unfamiliar direction goes unexplored. Each accommodation may be minor. Across many decisions, they can narrow the range of work the team produces.

Success within the process can become the team’s working definition of good creative judgment. Members may retain the formal authority to disagree while exercising it less often, making compliance appear voluntary because the incentives favor it.

Approved campaigns meet expectations, and model recommendations look sensible. That reliability says little about work never proposed or audiences it might have reached. A system can excel at selecting familiar possibilities while its users stop noticing what they have stopped considering.

Waterhouse writes in the “Mirror Image” section: “We’re building intelligence in our likeness.” He argues that people then reorganize their companies and institutions around that image.

In a creative department, that reorganization can occur inside individual habits. Teams adapt until the tool’s preferences start to feel like their own judgment, then produce what the system understands because approval has become familiar.

What does Asimov’s Mule reveal?

Isaac Asimov imagined a disruption to a predictable future in Foundation and Empire. Hari Seldon’s psychohistory (a method for forecasting vast populations) lets him devise a plan extending centuries beyond his life and gives its dependents extraordinary confidence.

Then the Mule appears: a mutant able to alter people’s emotions and loyalties. Seldon’s calculations did not accommodate that influence. The population begins behaving differently, and the Plan is thrown off course.

One individual outside the model’s assumptions changes the future it was supposed to predict. Asimov’s device turns prediction into a contest between a system’s assumptions and forces that can escape them.

For a brand, the corresponding ambition is to create something compelling enough to change what people expect. An audience can trust the quality of the work without knowing what the brand will do next; that tension may sustain the relationship.

How can teams preserve room for surprise?

The source places surprise inside the creative task. Wonder depends on encountering something that could not have been specified in advance and still recognizing its value. Campaigns can change a product’s associations, make an unfamiliar experience desirable or give people a new way to understand a brand.

Audience preferences develop partly through encounters with the world, including the things marketers choose to put into it. An account of what people currently prefer can therefore do only part of the job.

Teams must also judge what might interest people once it exists and what they want the brand to contribute. Those ambitions often begin before there is enough evidence to make them comfortable, which is where synthetic forecasts cannot settle the choice.

The authors of Peak Analytical Humanity argue that a growing abundance of analytical capability places more weight on the choices made with it. The distinction between what a tool can do and what an organization should do becomes more consequential when AI can generate and evaluate an apparently endless supply of possibilities.

A reliable prediction can inform that choice without resolving it. A team may correctly expect a familiar approach to perform well and still pursue a more interesting ambition, based on its understanding of the brand, the relationship it wants with the audience and the period over which the work should matter.

The forecast contributes to judgment, but the organization retains responsibility for the ambition. Independence does not