Case-law research: how the workflow works

Does the quotation actually support the argument?

A lawyer needs a passage of a judgment together with the place it comes from. A list of similar documents becomes useful only when you can open the full context and check whether the passage supports the argument or merely contains similar words. A model that paraphrases the reasoning produces text that cannot be quoted in a pleading.

That is why a quotation is a separate data type: the exact passage, the case number, the record's address and the time it was fetched. The system checks that the passage appears word for word in the source text. A claim without that support is stopped. In our demonstrator, the collection held 40 passages of judgments from SAOS, the Polish court judgments database, each with its address and fetch time, and the local run of 10 questions needed no fees for external APIs.

Set the scoring separately for the relevance of the source and the accuracy of the quotation. Topic labels in a small collection are weak and can distort how retrieval is judged, which is why the full context is always checked in the canonical SAOS record.

What case law research AI costs and when it pays back

The count covers the work around a case: searching, reading, writing and passing things on. What stays is the time to check the result, 40% of today's work in this scenario. Before the quote we measure both times on 20 to 100 cases the team has already handled, in a free test that takes 5 working days.

The error that never happens matters more than the hours: a quotation in a pleading that does not exist in the judgment. That is why the main measure of the trial is the number of claims that passed validation without word-for-word support in the source. The target is zero.

What the evidence covers

The corpus held 40 targeted passages, without the full SAOS database or case files. The benchmark had 10 saved questions, and the tool gives no legal advice and was not deployed at a client. A rollout in a law firm starts with one area of law and a set of questions with passages chosen by hand, on which we measure accuracy. The scoring principles are described in OpenAI's guide to evaluation: the same set of cases before a change and after it.

A first trial in the business

The first scope covers one area of law and a set of questions with passages chosen by lawyers. The demonstrator's screens are shown on the case law research page. We build this kind of system as part of our AI apps. For the free analysis, a description of one case from the last few weeks and the tools it went through is enough. You do not need to send confidential documents, and you get a written plan within 2 working days.