AI in document review: best practices and lessons learned

Panel: Natasha Pearman and Joel Mitchell (Milberg), Daniel Heinrichs and Daniel Rupprecht (Sky Discovery UK)

At LIDW2025, Milberg and Sky Discovery UK led a refreshingly practical panel on AI in document review. Their consensus? AI isn’t here to replace lawyers - but the lawyers who embrace it will outperform those who don’t.

Predictive coding and continuous active learning (“CAL”) continue to make an impact. AI today is “like a very keen paralegal”: fast, eager, and impressively helpful (but requiring supervision) - especially for reviewing already-disclosed data or generating highly useful document summaries for second-level review.

That said, complex case law analysis or strategic drafting still demand a human touch. The panel cautioned against over-reliance on AI for deep legal reasoning or key decision-making, particularly in areas where legal arguments hinge on subtle distinctions and context.

Retrieval augmented generation (“RAG”) is the next frontier, enabling AI systems to work with firm-specific datasets, helping to identify key actors, timelines, or patterns early in large investigations.

The panel also flagged the possibility of future CPR Practice Directions to clarify acceptable AI prompting and review processes - a move that could help avoid retrospective disputes or satellite litigation about whether technology was used appropriately.

Ultimately, the message was clear: it’s not about replacing human judgement but about freeing up valuable time. Strategic cooperation between lawyers and tech experts will define the next phase of adoption.

I attended three other compelling talks at LIDW2025 - catch up here if you missed them: 

• The culture of London arbitration: going viral or an English virus?
• Use your discretion: good faith constraints on discretionary contractual powers.
• Enforcing arbitration awards - cradle to grave strategies, from interim relief in support of the arbitration through to final asset recovery.

Burges Salmon also co-hosted two events for LIDW2025: Complex Disputes in the Public Eye I: Balancing Risks and Demands across Civil, Criminal and Regulatory Spheres; and Part II: Leveraging Technology and Artificial Intelligence to Get Ahead of the Game. Check out Tom Whittaker’s LinkedIn post and share your thoughts with us if you attended.

Tom has also co-authored a draft Generative AI (GenAI) Best Practice Guide, which supplements ILTA’s Active Learning Best Practice Guide on using GenAI tools or other advanced review technologies in a document review exercise. The Guide provides practical guidance on using GenAI tools or other advanced review technologies in a document review exercise under PD 57AD. This is important because the PD imposes a duty to liaise and co-operate with other parties to promote the reliable, efficient, and cost-effective conduct of disclosure, including through technology. 

The guide is open for feedback until the end of June 2025 and the final version will be published in the Autumn. Tell us what you think: visit Tom’s post!

AI isn’t here to replace lawyers - but the lawyers who embrace it will outperform those who don’t.