Accurate AI review starts with collaboration, not a clever prompt

Claira, eDiscovery Collab's AI-powered document review, can complete first-pass review by assessing thousands of documents for relevance and coding them against the issues in a matter.

Before using AI for document review, people need to do some important teamwork. That teamwork is what determines how accurate the AI will be.  

Claira's first-pass review assesses thousands of documents for relevance, but it needs more than a list of issues or a copy of the pleadings. It needs the context behind the matter to know what it is looking for.

What happened? Who are the key people? What language did they use? Which events, allegations and defences matter? What makes a document relevant and, just as importantly, what does not?

This knowledge usually sits across several teams:

• The lawyers understand the legal issues and case theory.

• The client understands the people, events and business language.

• The eDiscovery team understands the documents, the data, and how to turn that knowledge into effective review instructions.

The good news is that lawyers do not need to learn how to write complicated AI prompts.

Lawyers can describe what they want Claira to identify in plain language, just as they would explain the task to a junior lawyer in the team. Claira then develops those instructions into a prompt designed to work effectively with the underlying AI model.

However, Claira still needs the right information to work with.

A request such as “Is this document relevant?” leaves too much open to interpretation. Clear instructions should explain the issues, relevant events, key people, important terminology and how the document should be coded.

The prompts are then tested against real documents. The results are reviewed, unclear instructions are refined, and the prompts are tested again.

If Claira returns an unexpected result, it does not automatically mean the technology is wrong. The prompt may need more context. The issue definition may be too broad, or a term may have a particular meaning in the matter that has not yet been explained.

Every round of feedback improves the instructions and, in turn, the accuracy of the review.

Claira can do the heavy lifting at scale, but the lawyers, client and eDiscovery specialists provide the knowledge and judgement that guides the AI process.

The best results do not come from handing documents to AI and stepping away.

They come from people collaborating and using their case knowledge to teach Claira what really matters.

Click here to book a Claira demonstration.

This blog is for general information only.