How cricket teams use data: analytics behind the game

The performance analyst is now a fixture in every serious cricket set-up, and almost nobody outside the dressing room knows what the job involves. Coverage tends to swing between two wrong pictures: the analyst as a laptop oracle who tells the captain where to put the fielders, and the analyst as a harmless clerk who cuts video clips. The reality is more specific and more limited than either, and understanding it explains a good deal about how modern cricket is played.

Analysis in cricket is not primarily about prediction. It is about reducing the number of things a coach or a captain has to guess.

Where the numbers come from

Every ball in professional cricket is logged. At the simplest level that means a scorer's record: bowler, batter, delivery type, line and length as judged by a coder, shot played, where the ball went, runs scored, outcome. That coding is done by a human watching video, usually to a defined taxonomy, and it is slow, dull and the foundation of everything else.

On top of that sit the tracking systems. Ball-tracking gives release point, speed, bounce, deviation off the pitch and through the air, and the projected path after impact. Edge-detection systems give a sound and vibration record. Multi-camera systems increasingly give fielder positions and batter movement. Training data comes from wearables and load monitors, which say more about who is fit to bowl a fourth spell than about tactics.

None of this is exotic. What distinguishes a good analysis department is not access to data, which is broadly similar across the top teams, but the quality of the coding taxonomy and the discipline of applying it consistently. If two coders disagree about what counts as a good length, every conclusion downstream is noise.

Matchup analysis

Matchup analysis is the most discussed and most misused part of the discipline.

The naive version looks up how a specific batter has fared against a specific bowler and treats the result as a plan. This almost never works, because the sample is tiny. A pair of international players may have faced each other a small number of times across a career, and a handful of deliveries tells you nothing that is not swamped by chance. A batter who was dismissed twice by a particular bowler in one series is not thereby vulnerable to him.

The version that works aggregates by type rather than by individual. Instead of asking how a batter goes against one left-arm orthodox spinner, you ask how he goes against left-arm orthodox spin generally: how he scores, where he scores, how often he plays a false shot, whether he attacks the ball turning away from him or waits. That gives a usable sample and a transferable conclusion, because bowlers of a type share the essential problem they set.

Good matchup work is structured around the mechanics of the contest rather than the identities involved. The questions are things like:

• Does this batter score more freely against pace on the ball or pace off it?
• Which lengths does he attack and which does he block, and does that change against a moving ball?
• Does he prefer the ball coming into him or leaving him, and is that preference strong enough to be worth changing the bowling to exploit?
• Where does he score, and can that region be protected without giving away an easier one?
• What has he done when a particular field has been set, as opposed to what he has done in general?

The answers are probabilistic and they are inputs to a discussion, not instructions. A captain who takes a matchup as a command ends up bowling an off-spinner to a left-hander in the last over because the numbers said so, having ignored the fact that the off-spinner is nineteen and terrified.

Phase-by-phase planning

Limited-overs cricket is not one contest. It is a sequence of phases with different rules, different fields and different economics, and analysis is organised accordingly.

The fielding restrictions define the phases. In the powerplay, with few fielders permitted outside the ring, boundaries are relatively cheap and wickets are relatively expensive. In the middle overs the ring opens up, singles become available and boundaries become harder. At the death the value of a wicket falls sharply, because there are fewer overs left for the incoming batter to hurt you, and the value of a dot ball rises.

Analysis produces phase-specific benchmarks for a ground and a match type: what a competitive score looks like at the end of the powerplay, what a chase needs at the halfway point, how much a wicket is worth in each phase. These are ground-specific because grounds differ enormously, and the same total means different things on a small square with true bounce and on a slow surface with long straight boundaries.

The practical output is not a target score. It is a set of thresholds that tell a captain when the game state has changed. A side that is behind the phase benchmark but has wickets in hand is in a different position from a side that is level on runs and has lost its top order, and quantifying that is genuinely useful, because human judgement under pressure tends to overweight the last two overs.

Phase thinking also drives bowling allocation. If a particular bowler is at his most effective in the powerplay and least effective at the death, and the captain has been holding him back out of habit, the analysis makes that visible in a way that a season of watching does not.

Wagon wheels, pitch maps and what they actually show

A wagon wheel plots where a batter scored, usually as lines radiating from the middle. A pitch map plots where the ball landed. On their own, each is close to useless, and the pair of them together is the beginning of a plan.

A wagon wheel alone tells you where a batter has scored, which is a function of where he has been bowled to. A batter who scores heavily square on the off side may be a great cutter, or may simply have been fed width all season. The pitch map answers that question. Filter the wagon wheel by delivery length and suddenly it says something: this batter scores square of the wicket against anything short of a length, and almost nothing when the ball is full and straight.

The same combination works from the bowler's side. A bowler's pitch map shows his density of lengths, and overlaying the runs conceded shows which parts of his own map are hurting him. Most bowlers have a length band where they are close to unplayable and a band where they are ordinary, and the difference between the two is often a matter of a few inches. That is a coaching conversation, and it is one that video alone rarely produces because the human eye does not aggregate.

Two derived measures do a lot of work here. Control percentage counts the proportion of deliveries a batter met as he intended, judged by a coder, and it is a better short-term guide to how a player is batting than runs are, because it is less dependent on luck. False shot rate is the same idea from the bowler's perspective, counting how often a bowler induced an error. A bowler with a high false shot rate and poor figures has usually been unlucky. A bowler with good figures and a low false shot rate has usually been lucky, and will be found out.

Expected runs

Expected runs is the idea that has changed analysis most in the past few years, and it is simple in principle. For every delivery, given its length, line, speed, movement, the phase of the innings, the ground and the match situation, you can ask what happened historically on average across a very large number of similar deliveries. That average is the expected outcome for that ball. Aggregate over an innings and you get an expected score. Aggregate over a bowling spell and you get what the spell should have cost.

The value of this is that it separates process from result. Cricket is a game with enormous variance in outcomes. A bowler can bowl six deliveries in exactly the right area and be hit for four boundaries by a batter who took risks that came off. Under conventional figures the bowler had a bad over. Under expected runs he bowled a good over and got a bad result, and the sensible response is to give him the ball again.

The same logic applies to batting. A batter who plays a series of high-risk shots that come off has scored more than the expected value of what he did, and the analysis says so, quietly, before the coach picks him again on the strength of one innings.

Expected models are also the basis of most in-game win probability displays. Those are legitimate as a description of the average outcome from similar positions, and misleading if read as a forecast of a specific match, because the model does not know that one side's best bowler has a hamstring problem.

Field placement from data

Field setting is the clearest example of analysis becoming a physical thing on the ground.

The raw material is a scoring map: where does this batter score against this type of bowling at this length, weighted by how often. From that you can work out which regions are worth protecting and which can be conceded. The critical constraint is that a fielder moved to one place is a fielder taken from another, so every field is a trade, and the analysis is about which trade is least bad.

Two principles come out of the work repeatedly. The first is that the field must match the bowling plan. A field set to protect the leg side is worthless if the bowler misses his lengths and offers width, and most fields that look stupid on television are fields that were sensible until the bowler stopped executing. The second is that boundary riders should be positioned by scoring density rather than by convention. The traditional straight boundary rider is often standing where very few balls go, and moving him ten yards squarer changes the arithmetic more than any change of bowler.

Catching positions are a probability question. A slip is worth having when the chance of an edge carrying multiplied by the value of a wicket exceeds the runs that fielder would have saved in the ring. That calculation changes with the age of the ball, the pace of the surface and the batter, and it is one of the places where a good analyst can genuinely improve a captain's default.

What analysts actually do

The day-to-day work is less glamorous than the concepts.

Before a series, an analyst prepares dossiers on the opposition, built from coded footage and aggregated data, and reduces each to something a player will actually read. The reduction is the skill. A batter does not want a fourteen-page report on an opposition attack. He wants three sentences about the one bowler who will trouble him and a clip reel of that bowler's stock ball.

During a match, communication is restricted. There is no live coaching from the dressing room in most formats, so the analyst's work has to be done before play and at intervals. Between innings, at drinks and at the break, the analyst provides what has actually happened: which lengths have gone for runs, whether the ball is holding in the surface, which side of the ground is playing shorter.

After the match there is coding, review and the unglamorous business of maintaining the database. Over a season the analyst also becomes the institutional memory of a side, the person who knows that a plan was tried against a particular batter two years ago and why it failed.

The best analysts are translators as much as technicians. A finding that cannot be expressed in a sentence a player will accept is a finding that will not be used, and dressing rooms are not generous to people who arrive with a spreadsheet and an air of certainty.

The limits of the numbers

The case for honesty about limits is not modesty, it is accuracy. Analysis fails in predictable ways.

Samples are small. Cricket generates fewer meaningful events than most sports. A career of international innings is a small dataset by any statistical standard, and splitting it by opposition, by ground and by phase produces cells with almost nothing in them. Most confident-sounding matchup claims are built on samples that would not survive a first-year statistics course.

Players change. A batter's record against short bowling from three years ago describes a player who no longer exists. Technique, fitness and confidence all move, and models built on career aggregates assume a stability that is not there.

Conditions dominate. The same delivery is a different delivery on a seaming surface, at altitude, with a wet ball, under lights. Data pooled across conditions hides more than it shows, and data split by conditions runs into the sample problem again.

Selection effects are everywhere. A bowler's record against a batter exists because a captain chose to bowl him at that batter, usually in circumstances that favoured the matchup. Observational data records what teams already believed as much as what is true.

Measures become targets. Once a squad is judged on a metric, players optimise for the metric. A batting side told to value dot-ball percentage will take singles it should not take. This is not an argument against measurement, it is an argument for choosing measures that reward what you actually want.

Description is not prescription. Knowing where a batter scores does not tell you how to stop him, because the alternative you would push him towards may be worse. Every field change is an experiment on a system that responds.

The teams that get value from analysis are the ones that treat it as a way of forming better priors rather than as a source of answers. The question is not what the data says. It is what the data would have to say for us to change our minds, and whether we have enough of it to be worth changing them.

If you want to develop a feel for this yourself, the cheapest exercise is to watch an innings with one question in mind rather than none. Pick a batter, write down where you think he scores, then check the scoring map afterwards on a player page. The gap between what you saw and what actually happened is precisely the gap that analysts exist to close.