Context

We worked with a Swiss law firm operating in a highly regulated professional environment, where confidentiality, data residency, and compliance are fundamental to daily practice.

The firm maintained a large body of internal documentation, including precedents, templates, internal guidance, and legal research. Over time, this knowledge base had grown extensive but difficult to navigate efficiently, particularly under time pressure.

Public cloud AI tools were not an option. Client data, internal legal reasoning, and firm knowledge could not be exposed to external systems, and any AI-based solution needed to operate within strict privacy and compliance boundaries.

The Real Problem

The issue was not legal expertise. It was time.

Lawyers spent significant portions of their day:

• searching for relevant precedents

• reusing and adapting existing documents

• answering recurring internal questions

• switching between systems to assemble context

This work was necessary, but it was not always billable. It reduced the time available for higher-value legal work and placed a growing administrative burden on the practice.

Any AI system introduced into this environment needed to save time without compromising professional responsibility, confidentiality, or legal accuracy.

Constraints That Shaped the Design

Several constraints defined what was possible.

All data needed to remain private and controlled, with clear data residency guarantees aligned with Swiss legal and professional standards. Internal knowledge could not be used to train shared or external models.

Access to information needed to respect existing internal boundaries, ensuring that lawyers only saw material appropriate to their role and practice area. Outputs needed to be predictable, reviewable, and grounded in trusted sources.

Just as importantly, the system needed to fit naturally into legal workflows rather than disrupt them.

What We Built

We designed and deployed a private AI knowledge system centred on retrieval rather than generation.

At its core, the system used a retrieval-augmented generation (RAG) approach to make the firm’s internal documentation searchable and usable through natural language queries. Lawyers could ask questions in plain language and receive responses grounded in the firm’s own documents.

Building on this foundation, we trained the system on relevant Swiss legal frameworks and structures, enabling it to support legal reasoning without acting as a substitute for professional judgement.

Rather than a single generic assistant, we created multiple specialised assistants, each focused on a specific aspect of the firm’s work. These included assistants for internal research, document drafting support, procedural guidance, and recurring practice-specific questions.

All assistants operated within private, compliant infrastructure, with no client data leaving controlled environments.

Design Considerations

The system was deliberately designed to assist, not decide.

Responses were grounded in internal documentation and recognised legal structures, but lawyers retained full responsibility for interpretation and final output. Predictability, traceability, and clarity were prioritised over linguistic sophistication.

Particular attention was paid to ensuring that the system reduced friction rather than introduced new review overhead. The goal was to make existing knowledge easier to access, not to automate legal judgement.

Outcome

The system significantly reduced time spent on internal research and document preparation.

Lawyers were able to locate relevant information more quickly, reuse existing work more effectively, and focus a greater proportion of their time on billable, client-facing activities. As a result, the firm saw a meaningful increase in productive hours without increasing workload or compromising quality.

Equally important, confidence in the system grew steadily. Lawyers trusted it because it respected confidentiality, stayed within defined boundaries, and behaved consistently.

Why This Matters

This deployment reinforced a key insight.

In professional services, the value of private AI lies not in replacing expertise, but in removing friction around knowledge work. When internal knowledge becomes easier to access and apply, professionals can focus on the work that actually matters.

For regulated practices such as law firms, private AI succeeds when it increases leverage without increasing risk.