Synthetic Audiences and the Importance of Being Unpredictable

Surprise! The world is accelerating. Marketing departments are being asked to do what everyone else is: produce more, spend less, and somehow make the work better, too.

AI is now almost completely dominating the zeitgeist, and when it offers a way to test ideas more quickly, at a fraction of the cost of conventional research, marketers have good reasons to listen.

Synthetic audiences are simulated groups of consumers, built using data that can include customer records, interviews, and demographic profiles. A team can present its work to these audiences, examine their predicted responses, revise, and try again without recruiting another human panel.

Now, for a company that understands its customers and has checked those predictions against actual behavior, this could be enormously useful. Much of marketing involves improving things we already do, and there’s nothing trivial about doing them well. An expensive mistake avoided or a weak campaign improved is a perfectly respectable outcome. Indeed, most departments would be wise to pursue these opportunities.

But the appeal is more than efficiency for efficiency’s sake. Synthetic audiences also promise something that’s become invaluable: a defensible reason to proceed.

“We tested it, and this is what customers preferred” carries a different kind of authority from “we think this is a good idea,” even when the customers in question are simulated.

Investors are betting heavily on that authority. In July, Simile, which builds AI simulations of real consumers for companies to survey, raised $200 million at a $2 billion valuation, about five months after its Series A. Adobe lets marketers test creative and messaging against AI-generated customer profiles inside its Brand Intelligence product.

That being said, as these tools become part of how creative work gets approved, they’ll influence what people consider worth proposing in the first place. We may find ourselves getting much better at anticipating the audience while gradually losing the ability to give it something unexpected.

Learning Faster

The promise of synthetic research fits into a broader change in how businesses operate in the face of this new technology. As Steven Waterhouse argues in Learning Machines All the Way Down, AI reduces the cost of trying things, examining the results, and adjusting. Companies that organize themselves around this process can incorporate what they learn into the next decision, improving – and accelerating – as they go.

Marketing has always involved some version of this. We make a judgment about what might appeal to people, put it into the world, and discover how good that judgment was. Research helps us prepare, experience informs our interpretation, and the audience usually retains the right to surprise us.

United States Air Force Colonel John Boyd gave this process a more formal expression he called the OODA loop: Observe, Orient, Decide, Act (and repeat). Businesses frequently invoke it as an argument for speed, but Boyd’s account gives considerable attention to how we understand what’s happening. Orientation draws on experience, culture, and new information. In his diagram, a decision is a hypothesis and an action is a test, with the results feeding back into our understanding.

Synthetic audiences can help us work through that understanding before committing resources. They may expose an assumption we hadn’t noticed or suggest an interpretation we hadn’t considered. But another simulated response doesn’t, by itself, add another observation of how people behave. Improving a campaign’s predicted reception still leaves us with something to discover when people encounter it.

The distinction becomes consequential when we start revising primarily to improve the prediction. A team can run through hundreds of iterations and become increasingly confident in a campaign without having learned whether the audience shares that confidence. The process has accelerated, but some of the uncertainty remains exactly where we left it.

Boyd also wrote that “adaptability implies variety and rapidity,” connecting both to our ability to be unpredictable and cope with unforeseen circumstances. Applied to creative work, that combination asks more of us than producing variations quickly. It requires preserving a range of possibilities broad enough to discover something we wouldn’t otherwise have attempted.

Unfortunately, the pressure to move faster can make those unfamiliar possibilities harder to entertain. When a deadline is approaching and a budget needs approval, the idea everyone already understands has certain advantages.

The Cost of Playing It Safe

David Perell recently described this dynamic on TBPN, crediting Derek Thompson with the observation that Hollywood has succumbed to the “Moneyball-ification of our world.” In his telling, the industry has become boring partly because it has become intelligent about risk: “The problem is you end up getting Spider-Man 74 or Lord of the Rings 7 rather than a new show.”

There’s a commercial logic to the complaint. An established franchise comes with an audience, a record of performance, and some understanding of what people expect. Financing another installment can be sensible, and audiences can take real pleasure in returning to a familiar world.

The first installment, however, had to accept its uncertainty before the evidence of an audience existed. A studio can make a succession of reasonable decisions and still produce a culture in which unfamiliar stories struggle to get made.

In other words, a succession of reasonable decisions can produce a culture that feels repetitive, even if each project satisfies the criteria under which it was approved. We routinely “approve” ideas that we’re progressively less excited about.

Now, better information needn’t necessarily lead to these outcomes. Understanding risks could make us more willing to take them, especially if we’d previously exaggerated the downside. But information alone doesn’t determine an organization’s appetite for uncertainty.

Synthetic research brings this calculation into the daily practice of creative work. We can consult the audience earlier, more frequently, and about increasingly small decisions. That can improve the work considerably. It can also make the anticipated reaction so continuously present that people begin accommodating it before they’ve had time to develop an idea of their own.

The Audience That Trains Us

A creative team doesn’t need an instruction to obey the model. If a favorable score helps secure approval, its members will naturally pay attention to what earns one. They’ll notice which kinds of proposals move forward easily and which require a lengthy defense.

Under pressure, that knowledge becomes useful. An idea can be adjusted before submission, an awkward element removed, an unfamiliar direction left unexplored. Each accommodation may be minor. Over time, however, they can change the range of work the team produces.

This is a learning loop too, and the people inside it are learning exactly what their environment rewards. The uncomfortable possibility is that success within the process gradually becomes their working definition of good creative judgment. They may retain the formal authority to disagree while exercising it less and less.

The resulting performance can be reassuring. Approved campaigns meet expectations, the model’s recommendations appear sensible, and the process looks increasingly reliable. Yet this tells us very little about the work that was never proposed or the audiences it might have reached. A system can become excellent at selecting among familiar possibilities while leaving its users unaware of what they’ve stopped considering.

Waterhouse describes a related development in the “Mirror Image” section of his essay. We’re building intelligence in our likeness, he writes, and then reorganizing our companies and institutions around its image. In a creative department, that reorganization can happen within the habits of individual people. We adapt to the tool until its preferences begin to feel like our own judgment.

At that point, some of the predictability we observe may be a consequence of how we’ve learned to work. We’re producing things the system understands because we’ve become accustomed to asking for its approval.

The Mule

Isaac Asimov imagined a particularly dramatic interruption to a predictable future in Foundation and Empire. Hari Seldon’s psychohistory can forecast the behavior of vast populations, allowing him to devise a plan that extends centuries beyond his own life. Its success gives the people relying on it extraordinary confidence in what comes next.

Then the Mule appears. A mutant with the power to alter people’s emotions and loyalties, he exerts an influence that Seldon’s calculations hadn’t accommodated. The population begins behaving differently, and the Plan is thrown off course. An individual who falls outside the model’s assumptions changes the future it was supposed to predict.

Preserving the Ability to Surprise

For a brand, the corresponding ambition is to create something compelling enough to change what people expect. That requires a particular kind of unpredictability. An audience can trust our judgment and the quality of our work while remaining unable to anticipate what we’ll do next. Indeed, that combination may be part of what keeps the relationship alive.

Wonder depends on our capacity to encounter something we couldn’t have specified in advance and nevertheless recognize its value. The creative challenge is to make such encounters possible through work that has the quality, nuance, and coherence to reward people’s attention. As our tools become more capable, we should want the freedom to attempt more of that work.

For marketing, this presents a particular problem. Our work can help people discover something they didn’t already know they wanted. A campaign can change the associations around a product, make an unfamiliar experience desirable, or give people a new way to understand a brand. The audience’s preferences develop partly through its encounters with the world, including the things we choose to put into it.

That means an account of what people currently prefer can only do part of the job. We also have to exercise judgment about what might interest them once it exists, and about what we want the brand to contribute. Those ambitions can begin before there’s enough evidence to make them comfortable.

In Peak Analytical Humanity, we argued that the growing abundance of analytical capability would place more weight on the choices we make with it. The distinction between what we can do and what we should do becomes particularly consequential when a tool can generate and evaluate a seemingly endless supply of possibilities.

A reliable prediction can inform that choice without resolving it. We might correctly anticipate that a familiar approach will perform well and still decide there’s a more interesting ambition worth pursuing. That decision involves our understanding of the brand, the relationship we want with its audience, and the period over which we expect the work to matter. The forecast contributes to our judgment, but we remain responsible for the ambition.

Preserving that responsibility requires more than confidence in our own taste. People are perfectly capable of mistaking an indulgence for a breakthrough, and an unfavorable prediction deserves consideration. Independence becomes useful when we can examine the objection, decide why an idea still warrants a chance, and arrange a real encounter with the audience that teaches us something.

Cheaper preliminary research should make more of those encounters possible. Some of the savings can fund work with unfamiliar audiences or limited tests of ideas whose value remains uncertain. The organization then gains evidence about possibilities it might otherwise have discarded, and the people doing the work retain the habit of acting on a considered judgment.

AI can participate throughout this process, including in the creation of the unexpected idea. Using it well may expand our imagination considerably. The freedom worth preserving is the ability to pursue a possibility whose value hasn’t yet been settled by the systems we use to evaluate it.

P.S. Perhaps AI will eventually anticipate every meaningful change in human preference, including our responses to the most original work imaginable. If that happens, all bets are off. Until then, we still have to put ideas in front of people and see what they make of them.