A Formula 1 race looks like a driver problem. It is mostly a data problem. Behind every pit call sits a team of analysts turning tyre wear, lap times, and weather into a decision that has to be made in seconds and cannot be taken back.
We ran that same kind of work at scale. Over one project we analysed 131 Formula 1 races from the 2020 through 2025 seasons, pulling lap, stint, and telemetry data through the FastF1 library and cross-checking the 2024 season against an independent race-summary dataset. The goal was not to predict a champion. It was to see, in a domain where the numbers are unforgiving, what real analytical work looks like end to end. Here is what the data showed, and what it tells any business trying to hire people who can do this work.
The four questions every dataset has to answer
Good analytics is not one skill. It is four, stacked in order, and most hiring goes wrong because it tests only the first.
The first layer is descriptive: what actually happened? For our study that meant median pace per tyre compound, how long stints lasted, and how much time a car lost in the pit lane. Across all six seasons the pit-lane loss stayed remarkably stable, with season medians sitting between 22.7 and 23.7 seconds. Circuit layout drove almost all of the variation.
The second layer is diagnostic: why did it happen? Here we regressed lap time against tyre life to isolate how fast each compound degrades. The soft tyre wore roughly three times faster than the hard. That single gap explains most of what looks, on television, like driver error or luck. It is the difference between a stint that ends on plan and one that falls apart.
Key insight
An analyst who can only deliver the descriptive layer gives you a dashboard. An analyst who can carry a question through diagnosis, prediction, and prescription gives you a decision. That distance is the whole game.
The third layer is predictive: what happens next? We used gradient boosting to forecast finishing position from grid slot, tyre, and weather features, then checked it against the actual finishing order. With a limited number of races the model is illustrative rather than production-grade, and a good analyst says so out loud instead of overselling it.
The fourth layer is prescriptive: what should we do? We ran Monte Carlo simulations across compound sequences, feeding in the degradation slopes and measured pit loss from the earlier layers, plus a Safety Car probability. The output was not one answer but a distribution of expected race times per strategy, which is exactly how a real strategy call gets made under uncertainty.
The habits that showed up in the good work
Three things stood out across the project, and they map directly onto what businesses should screen for.
Interpretable models came first
We used linear regression for tyre degradation before reaching for anything heavier, because a race engineer has to trust and question the number, not just accept it. The best analysts add complexity only where it earns its place. In a boardroom, a model nobody can explain is a model nobody will act on.
Uncertainty was reported honestly
The prescriptive layer produced a spread of outcomes, not one confident figure. When a thin sample made the predictive model weak, that limitation went in the report rather than getting buried. An analyst who hands you a number without an error bar is hiding something, usually by accident.
The data engineering was invisible but decisive
Merging lap data with the summary spine on year, round, and driver, filtering to clean laps, correcting for fuel load: none of it appears in the final chart, and all of it decides whether the chart is true. Most analytical failures happen here, long before the modelling starts.
Why this matters when you are hiring
Most technical interviews test the wrong thing. They ask a candidate to reverse a linked list or optimise a query, then put that person in charge of decisions worth far more than a clean algorithm. The F1 project is a useful mirror because it is unforgiving in a way business data often is not. There is a real answer, and it shows up on the timing screen.
What you want is someone who asks what happened, then why, then what is coming, then what you should do, and who can walk you through each step with the uncertainty attached. That person is rare, and they rarely announce themselves well on a standard CV. Screening for the full four-layer journey, on real messy data, surfaces them far better than a coding puzzle does.
This is the standard we hold at TalentRack. Our Data Science and Data Analysis work is built by senior specialists and anchored in a real business question, never technical complexity for its own sake. If your team is sitting on data it cannot turn into decisions, that is usually a talent gap, and it is a solvable one.
Frequently Asked Questions
The four layers are descriptive (what happened), diagnostic (why it happened), predictive (what will happen next), and prescriptive (what to do about it). Each layer builds on the one before it. A team that stops at descriptive dashboards is only doing a quarter of the work.
Look for analysts who can move across all four analytics layers, who reach for interpretable models before complex ones, and who report uncertainty honestly instead of presenting a single number as fact. Domain judgement, clean data engineering, and clear communication matter as much as raw modelling skill.
In most business settings the data is small and the cost of a wrong decision is high. A regression a stakeholder can question and trust often beats a black box that is marginally more accurate but impossible to explain. Add complexity only where it genuinely improves the decision.
TalentRack delivers Data Science and Data Analysis solutions built by senior specialists, and assesses talent on real end-to-end analytical work rather than isolated coding puzzles. Every model is anchored in a real business question.
