Ashurst
Ashurst tests generative AI with its lawyers, then rolls Harvey out to all offices
UK law firm Ashurst ran a months-long generative AI experiment with its lawyers and reported that first drafts of briefs took about half the time. A separate Harvey trial involved over 525 users in 23 offices, and the firm then made Harvey available to all its lawyers and business services staff.
Key figures
Copied exactly from the sources. Copy any figure with its citation.
The problem
Ashurst, a UK-based law firm, tested whether generative AI could speed up lawyers' work, such as drafting briefs, before committing to a firm-wide tool.
What they built
The firm ran an experiment with its lawyers, then ran a proof of concept with Harvey, a legal AI platform, and made Harvey available to all of its more than 4,000 lawyers and business services staff. Legal IT Insider reports that the firm points out it is the only global law firm so far to roll Harvey out across all offices from day one. Staff use it for tasks such as document review, due diligence, drafting and market research.
How they did it
Ashurst first ran a five-month experiment involving more than 400 staff. A Harvey proof of concept followed, with over 525 users from 23 offices who sent over 4,000 queries. The firm published a report called VoxPopulAI, whose findings were based in part on the positive outcome of the Harvey proof of concept, and the firmwide rollout announcement followed it. The firm's chief digital officer led the trials, and the knowledge team said training, learning sessions and regular reviews would continue.
Results
In the firm's experiment, lawyers using generative AI wrote the first draft of briefs in about half the time it took them unaided, according to Bloomberg Law. The opening of that article also says the firm found AI output at least as legally correct as a lawyer's. The Harvey trial reached over 525 users in 23 offices, and the firm then extended access to all of its lawyers and business services staff.
All reported figures
| Measure | Figure | Citation |
|---|---|---|
| Time to write a first draft of a brief with generative AI versus without, in the firm's experiment | about half the time1 | |
| Staff involved in the firm's generative AI experiment | more than 4001 | |
| Users in the Harvey trial, across offices worldwide | over 525 users2 | |
| Offices the Harvey trial users came from | 232 | |
| Queries sent during the Harvey pilot phase | over 4,0002 |
Lessons for your project
Test with a large, mixed group before buying firm-wide. Ashurst's trial drew over 525 users from 23 offices, so the rollout decision rested on broad evidence.
Measure time on a specific task, such as the first draft of a brief, rather than asking people whether AI helped.
Pair the rollout with training and regular reviews. Ashurst said it would run learning sessions and ongoing reviews after launch.
Write up and share what the trial found. Ashurst published its findings in a report, and the rollout announcement followed it.
Caveats
- The drafting-time result is the firm's own, as reported by Bloomberg Law. Only the opening of that article could be read (the rest is behind a paywall), so the method and sample are not known.
- The Bloomberg article describes a five-month firm experiment and does not name Harvey. The time saving cannot be attributed to Harvey alone.
- The Harvey trial figures come from Legal IT Insider reporting on Ashurst's announcement. The trial's length is not stated.
- No source names the underlying AI model. Harvey is a legal AI platform; its models are not specified in the sources used.
- The sources give an announcement date, not a go-live date. The deployed date uses the month of the Bloomberg article, which reports the experiment results. The firmwide rollout was announced on 25 June 2024.
Sources
- Gen AI cut lawyers' drafting time in half, UK's Ashurst says
Bloomberg Law · 2024-06-11 · News · news.bloombergtax.com
- Ashurst launches global Harvey partnership following firmwide trial
Legal IT Insider · 2024-06-25 · News · legaltechnology.com
Summarized in our own words from the sources above. Spotted an error? See corrections.