Formula 1 doesn’t guess. It simulates thousands of races before making a single call. Customer marketing is overdue the same discipline.
I found myself watching the Hungarian Grand Prix yesterday. For anyone who follows the sport, this season has changed a lot of the rules and some fundamental principles around car design. Some changes have been good for the sport, others have not quite worked as intended. But halfway through the season it is fascinating to watch eleven teams still making big bets, and big changes, to the strategies they are pursuing, and to see how that plays out in the calls they make during a race. And one thing holds across all of them: no serious team makes a strategy call on instinct.
Before the lights go out, they have already run the race thousands of times. When to pit. Which tyre compound for each stint. How to respond if the safety car comes out on lap 30. What to do if it rains in the final ten laps. Every plausible version of the two-hour race has been simulated, scored and ranked before the ‘lights-go-out’.
Then, during the race, they do it again. Continuously. Using a stream of real data coming off the car and from the environment.
This is now so completely standard that it would be unthinkable to operate any other way. And it is a near-perfect illustration of something customer marketing has not yet accepted: that high-stakes decisions made under uncertainty should be simulated before they are made, not learned from after.
What F1 actually does
I’ve followed and watched F1 for most of my life. I’m drawn to the relentless pursuit to win. To be better than the competition; or even your team-mate. I’m fascinated by the technology. I love the strategy and decisions. The fact that this is a team sport as much as an individual. Perhaps most I reveal in the stories that it creates. The narratives that unfold during a race, but that get told again and again as outcomes clearly show the decisions that were made.
However, the interesting part, for our purposes, is race strategy. The decisions taken on the pit wall, under time pressure, with incomplete information and a real cost attached to getting them wrong. This is where F1 has built a genuine discipline.
The core method is Monte Carlo simulation. A race involves a set of choices – when to stop, which compound, how many stops. Each choice produces an outcome that depends on variables the team cannot control. Safety car probability, rival strategies, tyre performance, and lap time variation. Rather than trying to solve this on a whiteboard, teams simulate the race thousands of times, drawing random values from probability distributions for each uncertain variable in every run. After 10,000 or more simulations, the team has a probability distribution of outcomes for each potential strategy.
The scale of this is worth sitting with. Working with race strategy software, former Aston Martin head of race strategy Bernie Collins has described teams capturing more than 1,000 data points per second to power over two million predictive simulations across a race weekend. Two million simulated outcomes, to inform a handful of real decisions.
And crucially, the simulation does not stop when the race starts. Once live timing data became available to teams, they were able to run real-time Monte Carlo simulations to answer questions like: given everything that has happened up to this lap, what is the best thing to do now if our competitor makes a three-stop strategy? The model updates as reality unfolds. As one strategist put it, they refresh it any time new data arrives.
There is a subtlety here that maps precisely onto customer decisions. The baseline simulation answers a deceptively simple question: without interacting with anyone else, what is the fastest way to do this race? But that baseline is never the final answer, because everyone else is running the same maths and the moment you act, they react. So teams layer game theory on top of the simulation, modelling how rivals will respond and adjusting the call accordingly.
The obvious strategy and the winning strategy are rarely the same thing.
Where the simulation helps make better informed decisions
As I the race unfold at the Hungarian Grand Prix yesterday, where three separate strategy calls decided the shape of the race.
The first was the choices the teams faced at the start. Around the Hungaroring the pitlane loses you significant time, overtaking is hard and the tyre behaviour on Sunday left the leaders genuinely undecided between a one and a two stop race. The time you lose stopping is roughly the time you gain on fresher rubber, so on paper the strategies were close to level. That is not a comfortable position. It means the fastest route to the flag is not obvious, and the decision only become easy once someone, or something, else moves. Which is exactly what happened.
The move came from Ferrari. Lewis Hamilton pitted from fourth to try an undercut, and McLaren had a decision to make with both cars running first and second. It pitted Oscar Piastri to cover Hamilton, because Piastri was ahead and got strategic priority. On first-order logic this was correct. It protected the position from the car behind. But it carried a second-order cost that the obvious call concealed. Piastri came out into traffic, lost his four second cushion, and was then delayed lapping a backmarker. That let Lando Norris, who stayed out longer in clear air, produce a run of fast laps and take the lead his teammate had been holding. The call that defended one car handed the race to the other. Piastri later retired with a gearbox failure, but the race had already turned before then.
The third call came late in the race. When Piastri stopped on track the race went to a Virtual Safety Car, which makes a pitstop far cheaper because everyone is slowed. Max Verstappen, running second, chose not to take it when his closes competitors did. He stayed out on soft tyres already around eighteen laps old and backed them to hold to the flag. On paper it looked marginal. What made it the right call was the degradation data: his used softs were not falling away much faster than a fresh medium would have, so the track position he would have surrendered was worth more than the grip he would have gained. Had it been a full Safety Car rather than a virtual one, the field would have bunched back up and the maths would have flipped entirely. Same driver, same lap, opposite decision, depending on which version of events the model was pointing to. It held, and he kept second.
Three decisions, three different shapes of uncertainty.
One where the options were so close together that the answer would not reveal itself until a rival moved.
One where the obvious defensive call carried a hidden second-order cost.
One where the same action was either right or wrong depending on a single variable outside the team’s control.
None of these were guesses. Each was a read of thousands of simulated outcomes, updated live as the race unfolded. In F1 the cost of getting one wrong is a place, or a race.
In a customer value decision, the equivalent misread costs a year of growth and a chunk of the base.
Engrama’s perspective
Look at how customer decisions get made in your business
A loyalty programme change worth tens of millions in customer value. A benefit is being removed. A tier structure is being reworked. An earn rate is being cut.
How many versions of the future were simulated before the call?
In most businesses, the honest answer is none. There was a business case. There was a slide showing the projected saving or the projected margin improvement. There may have been some sensitivity analysis on the headline number. But there was almost certainly no simulation of how different customer cohorts would respond, no modelling of the second and third-order effects on retention and spend, and no probability distribution of outcomes.
The decision went to the track on a hunch and was learned from after the cost was sunk.
This is not because customer marketers are less rigorous than race strategists. It is because the discipline has not yet been adopted as standard. The data exists. Businesses are rich with customer data. The compute exists. The simulation methods exist and are well understood. What is missing is the cultural expectation that a decision of this size should pass through simulation before it is committed. The same expectation that is now absolute on an F1 pit wall.
Customer marketing in 2026 is roughly where F1 strategy was in the 1990s: the tools are arriving, but the instinct to use them before acting has not yet hardened into practice.