What US Judge-Analytics Tools Got Right, and Why the UK Is Different

American legal analytics proved that the public record holds real, usable signal about how judges behave. France decided the same insight was dangerous enough to make a crime. The UK sits between them, and that in-between position defines the only kind of tool that belongs here.

The UK is late to judicial analytics, and for once lateness is an advantage. Two very different jurisdictions have already run the experiment and reached opposite conclusions about it, which means anyone building here now gets to learn from both. Between them, the American enthusiasm and the French prohibition mark out with unusual clarity where the useful and defensible version of this technology actually lives.

What the American tools got right

Over the past fifteen years, a wave of US products did something genuinely valuable with the public record. Lex Machina, which began life as a patent-litigation project at Stanford in 2008 and was bought by LexisNexis in 2015, took millions of federal court dockets and decisions, drawn from the public PACER system, and turned them into structured, queryable insight into how courts and judges actually behave. Ravel Law, acquired by LexisNexis in 2017, did something comparable for judicial reasoning. Bloomberg Law and Westlaw Edge followed with litigation analytics of their own.

What these tools surfaced was real and useful: how often a particular judge grants a particular kind of motion, how long they take to do it, which arguments and authorities tend to succeed in their courtroom and which fall flat. Ravel's analytics went as far as identifying the language a given judge found persuasive on an issue. The pioneers of the field liked to say they were freeing lawyers from anecdote dressed up as data, replacing the corridor war story ("I hear this judge cannot stand that argument") with something drawn from the record itself.

That was the achievement, and it is worth stating plainly, because it is the sound core of the whole enterprise: the public record contains genuine signal about how judges reason, and making that signal legible helps lawyers prepare better. The insight is correct, and it does not stop at the American border.

There is, though, a more aggressive end of this spectrum, and it is worth naming because it is where the tension begins. Some tools moved from describing how a judge has behaved toward forecasting how they will. One current example profiles hundreds of federal judges and offers to predict how each will rule on a motion to dismiss, in some accounts from little more than the case's docket number. That is a different proposition from surfacing a judge's past reasoning. It treats the judge as a function to be solved, and it is precisely this step, from understanding a court to predicting a named individual, on which the next part of the story turns.

Where the same insight became a crime

France looked at exactly the same capability and drew exactly the opposite conclusion.

In 2019, Article 33 of its Justice Reform Act made it a criminal offence, carrying up to five years in prison, to reuse the identity data of judges in order to evaluate, analyse, compare or predict their professional practices. It is widely thought to be the first ban of its kind anywhere in the world. The detail that matters most is what France did not do. It did not close its courtrooms or stop publishing decisions; French judgments remain public as open data. What it prohibited was a specific use of that open data: the profiling and prediction of named individual judges. The concern, upheld by the Constitutional Council, was that judge-by-judge prediction would fuel the strategic selection of courts and judges and distort the functioning of justice itself.

You do not have to agree with the French response to take its underlying point seriously. There is a line between understanding how a court reasons and reducing a named judge to a predicted output, and France decided that a tool built to profile and forecast the individual had crossed it.

Two lessons, not one

Put the two experiences side by side and they teach something that only looks contradictory.

The American story shows that analytics of the public record is genuinely valuable and worth building. The French story shows that one particular framing of it, the profiling and prediction of named individual judges, is fraught enough that a mature legal system was willing to criminalise it. The error would be to absorb only one of these lessons. The value is real and so is the risk, and the striking thing is that they attach to different framings of the very same underlying data. The signal is the same. What changes everything is whether you present it as preparation for a professional or as a prediction about a person.

The defensible product is the one that keeps the first and refuses the second.

Why the UK is its own case

Britain is neither America nor France, and its particular mix of characteristics points clearly at which framing belongs here.

On one side, the UK has a deep open-justice tradition and, increasingly, the infrastructure to match it. Judgments are public, and services like Find Case Law publish them under an open licence for reuse. There is no French-style prohibition on analysing them. The raw material, and the permission to work with it, are both present.

On the other side, British legal culture is notably wary of anything that smells of predicting or second-guessing the bench. That wariness is reinforced from two directions: by professional regulators who are insistent that the lawyer, and never a tool, remains accountable for the work, and by a public that polls sceptically whenever AI is mentioned in the same breath as judicial decisions. The UK, in short, has the openness that makes analytics possible and the culture that makes prediction unwelcome.

There is also a quieter structural difference. The American tools were built on rich, standardised docket data, win and loss recorded case by case, which lends itself naturally to scoreboards and rates. The UK has no equivalent public dataset of outcomes broken down by individual judge. What it has instead is one of the world's great collections of reasoned judgments. That difference in the raw material nudges the British product in the same direction its culture already points: toward how a judge reasons rather than how often they rule a particular way, which is to say toward preparation rather than a scoreboard.

The right tool for this jurisdiction

That combination is not a narrow gap to thread. It is a clear instruction.

The tool that fits the UK is one that takes everything the American analytics got right, the real signal in the public record about how a court reasons, and delivers it in the register this jurisdiction actually wants: preparation, not prediction. Not "here is how this judge will rule," which is the framing France criminalised and British instinct resists, but "here is how this court has reasoned on this issue before, here is the precedent and the climate around it, now bring your own judgment to bear." That is a fundamentally different product from a prediction engine, and deliberately so. It leaves the professional accountable, it stays on the safe side of the line France drew, and it serves a profession that wants to walk in better prepared rather than be handed a forecast.

This is the design Probatur is built around, and not as a concession wrung out by caution. The caution is correct. The UK neither needs nor would accept a machine that predicts its judges. It can make excellent use of one that helps its advocates read them. The signal the Americans found is here too, waiting in one of the most open judicial records anywhere. The task is not to predict what our judges will do with it, but to help the people who appear before them prepare. That is at once the safer path and, for this profession, the better product.

The lateness, then, turns out to be luck. The UK gets to learn from a jurisdiction that proved the value and a jurisdiction that named the danger, and to build the version that honours both.