Herbert Smith Freehills Kramer Brings Defensible AI Review to High Court Disclosure with Relativity aiR for Review

How did they do it?

• Deployed Relativity aiR for Review with full transparency and the express agreement of the opposing party.
• Reviewed roughly 100,000 documents in about 12 hours with Relativity aiR for Review, surfacing key materials early enough to shape live case strategy.
• Standardized a proprietary, repeatable workflow to adapt across matters, jurisdictions, and global teams.

Large-scale disclosure has always carried a hard tradeoff. The data keeps growing, the deadlines keep compressing, and courts keep raising the bar for how firms must explain and defend their methodology. Meeting all three at once, on a single matter, has historically meant throwing more people at first-level review and absorbing the associated cost and risk.

Herbert Smith Freehills Kramer set out to change that equation. One of the world's leading international law firms, with a strong reputation in complex, large-scale disputes including class actions, regulatory investigations, and cross-border litigation, the firm relies on Relativity as central to how its Digital Legal Delivery practice designs and delivers disclosure at scale. When a group litigation in the English High Court produced an initial dataset of more than five million documents, the firm saw an opportunity to prove that generative AI could carry real weight in high-stakes legal work, without compromising defensibility.

Earning the right to deploy AI on a live matter

Herbert Smith Freehills Kramer's adoption of generative AI was evidence-based from the start. Before bringing Relativity aiR for Review on a live matter, the team ran an extensive pilot with quantitative success criteria defined in advance – specific accuracy thresholds, precision and recall benchmarks, and direct comparison against experienced lawyer review. Only once those criteria were met consistently did the firm move to live deployment.

That discipline mattered, because this was new ground. To the team's knowledge, the matter represented one of the first deployments of generative AI at this scale in English High Court disclosure, so the firm secured the opposing party's express agreement and established a governance framework built to withstand scrutiny. Generative AI operates as a defined component within the firm's proprietary workflow, with its own protocols and quality control thresholds. That same framework, including the sequencing of analytics, the governance model, and the quality control protocols, now gives the firm a model it can adapt rather than reinvent across different matter types and jurisdictions.

From five million documents to early case insight

The Herbert Smith Freehills Kramer team approached the disclosure exercise end to end, looking for where complexity and cost were really accumulating. The answer was clear: effort was disproportionately concentrated in first-level review. So the firm took a different approach to that step, not by automating legal judgment, but by using generative AI to support prioritization and consistency at the front end.

Email threading and targeted keyword searches in RelativityOne narrowed the population from five million to roughly 140,000 documents. Senior case lawyers, Digital Legal Delivery specialists, and e-discovery professionals then developed and validated the prompt across several rounds of refinement, with the outcome depending on genuine engagement from senior lawyers who understood the issues in the case. aiR for Review delivered first-level relevance decisions across roughly 100,000 documents in about 12 hours, providing transparent reasoning for every call and meeting the firm's pilot benchmarks. Lawyers reviewed every relevant or borderline document to confirm relevance, assess privilege, and isolate confidential and irrelevant material, while active learning and quality control checks covered the rest.

"What excites us most about aiR for Review is its potential to fundamentally change the value of document review, from a largely manual, time-intensive exercise into a source of early, strategic insight. There is a shift underway from review as a cost burden to review as a strategic advantage, with aiR playing a central role."
Lyn Harris, Partner, Digital Legal Delivery

Freeing senior judgment for higher-value work

With new efficiency in first-level review, experienced lawyers were able to focus on case strategy, client advice, and witness preparation. The firm can now generate meaningful insight from very large datasets within days rather than weeks, early enough to shape client advice, negotiation strategy, and the direction of a case from the outset.

Lawyers who had approached the process with caution came away genuinely engaged, recognizing that this was about directing their expertise more effectively, not displacing their role.

Now Herbert Smith Freehills Kramer is extending this approach to a broader range of tasks and matter types and scaling it consistently across global teams, embedding generative AI as a standard part of how the firm delivers disclosure at scale. For a firm whose value rests on the quality of its judgment rather than the volume of its effort, that consistency is the point – a way to apply experienced legal thinking, defensibly, at a scale that would otherwise be out of reach.

"aiR for Review adds value when it is used thoughtfully, transparently, and in the right part of the workflow, never without experienced lawyers firmly in control. Used properly, it enables teams to manage complexity, control cost, and make better decisions without compromising defensibility."
Lyn Harris, Partner, Digital Legal Delivery