Clifford Chance

As one of the world's largest law firms with 7,000+ staff across multiple jurisdictions, Clifford Chance faced a defining challenge when generative AI emerged: how to enable firm-wide adoption without exposing client data, breaching professional obligations, or falling foul of the EU AI Act. Lawyers had ethical duties that made ungoverned AI experimentation genuinely risky — not merely inconvenient. Early client sentiment was a flat refusal, with concerns that matter data could become model training input. Meanwhile, retrieving internal policies and precedents from SharePoint and email archives could take up to two days, a friction cost that compounded across thousands of fee-earners every week.

Clifford Chance built its AI programme in two phases. In 2023 it launched Clifford Chance Assist, an internal tool built on Azure OpenAI and trialled with over 1,800 users before firm-wide rollout. In February 2024 it became one of the first organisations globally to deploy Microsoft Copilot for Microsoft 365 and Viva Suite across its entire workforce — a deliberate choice to consolidate on a single, governed hyperscaler platform and eliminate shadow AI usage. Underpinning both deployments was a bespoke governance framework: a traffic-light classification system for tools and data types, mandatory training for all staff, and a risk assessment process enabling the innovation team to proceed on new GenAI projects without escalating to the risk function in every instance. The EU AI Act's direction was incorporated into policy before it was finalised.

The Copilot rollout delivered measurable returns against a clear ROI threshold. The firm established that 15 minutes of time saved per lawyer per week was the breakeven point to justify the investment — and deployment exceeded it.

• Policy retrieval: reduced from 2 days to 4 seconds using document search across SharePoint and email
• Weekly time saved: exceeded 15 minutes per lawyer firm-wide
• Adoption base: 1,800+ users validated the platform during the Clifford Chance Assist trial before full deployment

Top use cases include email chain summarisation, SharePoint document search, information drafting, and creative argument review. Clients who initially refused AI use have since reversed that position following direct engagement.

• Invest in AI principles and governance frameworks before deploying tools — the upfront policy work is what enables faster, lower-risk iteration later.
• A traffic-light classification system (green/amber/red by tool and data type) lets staff act independently without creating compliance gaps.
• Anchoring to a hyperscaler platform (Microsoft/Azure) reduces vendor risk, limits shadow AI, and provides a single governed environment at scale.
• ROI calculations must account for aggregate time savings — even modest per-user gains justify enterprise investment when multiplied across thousands of fee-earners.
• Mandatory training and personal accountability at the supervisor level are essential; partners must understand how the technology works before they can responsibly oversee its use on client matters.