Fidal

Fidal, France's leading law firm with over 100 years of history serving 69,000 clients, faced a knowledge management crisis inherent to large legal practices. Its 1,200 professionals generate approximately 520,000 documents annually — a corpus of immense institutional value that remained largely inaccessible. Without a way to surface relevant prior work, lawyers repeatedly duplicated research across practice areas including public law, competition law, and real estate law. Beyond lost productivity, the firm's growing data footprint created serious exposure to data leakage risk. The net effect: billable hours consumed by redundant research rather than the high-value analysis and client strategy that differentiate a top-tier firm.

Two years ago Fidal launched 'Fidal IA,' a proprietary generative AI legal assistant built on Azure OpenAI and powered by two large language models — OpenAI and Mistral Large — selected by lawyers based on task requirements. The system uses a Retrieval Augmented Generation (RAG) architecture indexing 6 million documents: 40,000 internal firm sources and 2.7 million items from legal open data. Use cases were developed directly with practising lawyers and include generating summaries, drafting emails, comparing legal conclusions, legal monitoring, and translation. A collaboratively maintained prompt library grows organically through weekly training sessions and firm-wide brainstorming. Hosting on Azure ensures the firm retains full data sovereignty — a non-negotiable requirement given client confidentiality obligations.

Even in beta, Fidal IA has demonstrated measurable impact across adoption and efficiency:

• 30% reduction in average legal research time, per practising tax lawyer Maud Roux
• 119,000 messages processed across 28,000 conversations during the beta phase
• Dozens of new users joining monthly, with self-reported satisfaction among early adopters
• RAG index spans 6 million documents, combining internal precedent and public legal data

Qualitatively, lawyers report redirecting recaptured time toward analysis, strategy, and client relations. The firm also achieved its foundational requirements: data security, client confidentiality, and full technological independence — outcomes Fidal's CIO described as requiring 'complex technical and expertise-oriented choices.'

• Co-design with end users is not optional. Fidal built use cases and the prompt library directly with lawyers; weekly sessions drove adoption and ensured the tool reflected real workflow needs.
• RAG over proprietary data creates durable competitive advantage. A 6-million-document index combining internal precedent and open legal data cannot be replicated by off-the-shelf AI tools.
• Data security must be a design constraint, not a retrofit. In legal contexts, client confidentiality requirements should drive infrastructure choices — Fidal's Azure-hosted deployment gave the firm full data control from day one.
• Change management matters as much as the technology. Treating self-employed professionals as collaborators rather than passive users was central to Fidal's adoption success.