Simulations Are About to Make Failure Cheap in Advertising

By Rob Hall, CEO & Founder, Parallel

Advertising has answered three different questions with data, more or less in sequence. We began by observing. Censuses, surveys and panels gave us data about people, followed eventually by logs of almost everything they viewed/clicked/whatever. That told us what had already happened.

Secondly, we started modelling. Statistics and machine learning became remarkably good at spotting patterns and projecting them forward. Lookalike audiences, propensity models, reach curves and benchmarks all offered a view of what would probably happen next.

Now, we are moving towards simulation. Rather than producing a forecast based on what happened elsewhere, a simulation attempts to create a working model of the thing itself and run it forward. It sounds like a subtle distinction until you look outside advertising, where simulation is already how the big decisions get rehearsed. NASA rehearsed the Apollo 13 fixes on ground twins in 1970. Your car was crash-tested as a digital twin before it was built. Medicine builds simulated hearts to test treatments on. And a Formula One car laps a virtual track thousands of times before the real one exists.

Simulation gives you a way to experiment

Bear with me through the Formula One version, because it is the clearest. There are two ways to estimate how quickly a new car will lap a circuit. You could begin with last year’s data and apply what you already know about this year’s car to get a rough estimate of how fast it might go. That estimate might be perfectly useful, but it is still borrowed from other cars and therefore has its limits in terms of usefulness. It cannot show the nuances of the new car’s performance.

Alternatively, you can put a digital version of the actual car on a virtual circuit and drive it. Watch where it is quick, see where it struggles, alter the car’s characteristics and go again. Thousands of laps can be completed, in simulated form, before the real vehicle reaches the track.

So just how important is simulation to the F1? We asked Hoagy Nidd, F1 Head of Car Engineering . He said: “In Formula 1, simulation is a critical part of nearly every step of what we do. All the way from the early stages of concept design through to full driver-in-the-loop simulations of our car driving round any circuit. The process of simulation yields a better optimised solution in a shorter period of time. And in a world where we are defined by our results relative to our competitors, our sport would not exist without simulation.”

A prediction gives you an answer. Simulation is different because it gives you somewhere to experiment. You can change things, get them wrong and try again, cheaply and privately, before the consequences become real.

People are more complex than cars

Currently, media is still closer to the old model. Our version of last year’s data is benchmarks and norms: campaigns like yours achieved this; your category usually indexes at that. Sensible figures, perhaps, even when they are your own. But they all describe past campaigns, and none of them can tell you how this new ad will do, or where exactly it will perform the best.

The advertising equivalent of race day is the live campaign. A media plan’s assumptions are checked by spending the budget and reading the results, part way through or at the end. The lessons arrive at full price, in public, using real money and some of the audience’s goodwill. Today, we are largely still operating in this fashion. We are the team that still finds out after the fact.

Simulation itself is not new. Engineers have modelled cars, bridges and aircraft for decades. People are harder to model than any of those, but what has changed very recently is that it has become possible to build simulated audiences that respond in recognisably human ways.

Human validation is still necessary

The foundations also matter enormously. Simulated audiences depend on the data beneath them, the model interpreting it and the tests used to establish whether their answers are any good. In a recent Harvard and MIT-led study, identical simulated people produced answers 71 percentage points apart when researchers changed the AI model speaking for them. The authors stressed that human validation was still necessary.

That is a warning about standards, rather than an argument for ignoring simulation. Those standards are still developing, however, and buyers should be asking what a simulated audience has been tested against, not simply admiring how convincingly it talks.

Formula One uses a feedback loop to mark its own work. The simulator is checked against the real car every race weekend, and a team that stops doing it stops trusting the simulator within a season.
The media equivalent is checking a simulated audience against real people at both ends: what real people said going in, and what the campaign actually did to real people coming out. Simulation does not replace finding out for real. It moves most of it earlier, and makes the rest worth more.

With the way AI is changing the landscape, the most interesting effect will not be efficiency. Much of what AI has done in media so far involves completing familiar jobs more quickly: trafficking, reporting, versioning and so on. Helpful, certainly, but hardly a new operating model.

But suddenly, AI driven simulations can make failure cheap. Advertising has always learnt by releasing things into the world and seeing what happens in reality. As a result, experiments are expensive and tend to be cautious. A handful of things are tested instead of hundreds. Familiar approaches absorb most of the budget, with a small percentage reserved for something new, because discovering that an idea was wrong means paying the full market price for the lesson.

Bring that cost close to zero and the conversation changes. Instead of asking what we can afford to test, we can ask what we would try if being wrong were practically free.

Motorsport did not become better simply because simulation reduced the cost of test days. It became faster because teams could explore ideas they would never have risked on the real race day. Advertising may be approaching its version of the same shift.