Disrupting Formula 1

Background: The Pit Problem

Formula 1 is the fastest and most technologically advanced racing series in the world. Twenty-two cars compete across twenty-four Grand Prix each season, and the single most decisive factor in any race is often not the car or the driver. It is the pit strategy call. When to stop for fresh tires, which compound to fit, and how many stops to make over a race distance can be the difference between a podium finish and a points-less afternoon. Get it right and a midfield car beats expectations. Get it wrong and a race leader drops to tenth.

Today, pit strategy is built by small teams of human strategists making real-time decisions under extreme pressure with incomplete information. They must anticipate what twenty-one other teams are planning while conditions change lap by lap. The approach is reactive, treats each car as an independent problem, and degrades under the cognitive load of a live race. For a new constructor entering the grid, like Cadillac in 2026, there is no historical playbook, no institutional memory, and no margin for the learning curve that established teams spent decades building.

Our Solution: Proprietary Deep Learning for Full-Field Strategy

We developed a proprietary AI system that predicts and prescribes optimal pit strategy for every car on the grid simultaneously, in real time, lap by lap. Our technology achieves 100% strategy classification accuracy on held-out circuits never seen during training, a 0.58 lap mean pit timing error, and 82.6% of finish position predictions within three positions of actual. These results span nearly 10,000 test samples drawn from nine seasons of Formula 1 data.

This is not a dashboard or a visualization layer. It is a prescriptive engine that tells the pit wall what is going to happen, what they should do about it, and why. Each prediction includes confidence scores, named competitor targets, and specific lap-by-lap recommendations. The system runs on portable trackside hardware with sub-second inference, requires no internet during the race, and retrains automatically after every race weekend.

How This Compares: Published Research Benchmarks

The best-performing published model for F1 pit stop prediction is a Bi-LSTM architecture evaluated across five deep learning approaches (Frontiers in AI, 2025). It achieved an F1-score of 0.81 on binary pit stop detection, which predicts whether a driver will pit on a given lap. On the 2025 Chinese Grand Prix, that model reached 70.59% accuracy. The Deep-Racing neural network (IJML, 2023) achieved an F1-score of 0.67 for pit prediction and explained less than 40% of finish position variance. An SVM-based approach (García Tejada, 2023) reached F1-scores of 0.44 and 0.62 on pit stop quality and occurrence respectively.

Our system solves a harder problem at higher accuracy. Published models answer a binary question: will this driver pit this lap? Our system predicts the complete race strategy, the specific pit lap, and the finish position for every driver on the grid. It achieves 100% strategy accuracy and sub-lap timing precision on circuits it has never seen. The published state of the art tops out at 81% on a simpler formulation of the same problem.

Why This Matters: Real-World Impact

For the Strategy Department:

  • Full-field intelligence. Predictions for all 22 cars simultaneously, not just your own driver, updated lap by lap in real time.

  • Prescriptive recommendations with named undercut and overcut targets, predicted rejoin positions, and net time deltas for every pit scenario.

  • Pre-race preparation. Complete strategy reports generated automatically after Saturday qualifying, giving the pit wall a detailed plan before lights out.

  • Transparent reasoning. Every recommendation shows its confidence score and risk assessment. The strategist can override any call with a single decision.

For Team Leadership:

  • New-entrant disadvantage reduced. Nine seasons of learned patterns deployed from race one. This does not replace decades of institutional knowledge, but it closes the strategy gap significantly and gives a first-year team a foundation that would otherwise take years to build.

  • Edge computing, solar powered. Runs trackside on portable hardware that fits in a flight case. No cloud dependency, no internet required during the race, with an option to be powered entirely by solar energy.

  • Validated and ready. Tested on Grand Prix data spanning 2018 through 2025 with predictions verified against actual race outcomes.

For Sponsors and Stakeholders:

  • Competitive differentiation. AI-driven strategy is a visible, communicable technology story that positions the team as a modern, data-led operation.

  • Scalable platform. Adaptable to any constructor, any series, any regulation change through automated retraining on new season data.