The Challenge
A mid-sized Geneva law firm received about one hundred postal documents and over 200 emails daily. The sorting, classification, and assignment process to the correct client files required two legal assistants for several hours each day. Classification errors led to time losses searching for crucial documents, and important deadlines risked being missed. Urgent emails could remain unprocessed for hours if they arrived during busy periods.
Management wanted to modernize this process without compromising client file confidentiality or depending on public cloud solutions where data transits. The firm handled sensitive matters for family offices and international companies, making data sovereignty absolutely critical.
The Deployed Solution
We designed an intelligent automation solution combining several Microsoft technologies hosted in their own Azure environment based in Switzerland. The system relies on Azure Document Intelligence to extract text and structure from scanned or PDF documents, Azure OpenAI Service with a GPT-4 model deployed in private mode to analyze content and determine the nature of documents and emails, and Azure AI Search to index everything and enable instant semantic search.
The flow begins with scanning physical documents, which are automatically deposited into a secure SharePoint folder. In parallel, a Power Automate flow monitors the firm's shared email inbox. For each new document or email, the system triggers and extracts content. Document Intelligence processes scanned files and PDFs to extract full text and identify structural elements like headers, tables, and signatures. For emails, the body content and attachments are analyzed.
The extracted text is then sent to the GPT-4 model with specific instructions to classify according to a taxonomy defined with the firm: client correspondence, court summons, supplier invoice, contract, opposing correspondence, legal document, quote request, urgent matter, and others. The model also identifies key entities such as client names, file numbers, important dates, amounts, and urgency level.
For emails, the system also determines client sentiment and generates a one-sentence summary. All these metadata are indexed in Azure AI Search with the full content. Based on classification, the document or email is automatically routed to the appropriate SharePoint folder and concerned lawyers receive a Teams notification with a summary and urgency indication. Very urgent emails generate an immediate alert.
Measured Benefits
After three months of deployment, results exceeded expectations. Processing time for incoming mail and emails dropped from six hours per day to less than one hour, an 83% reduction. Assistants now focus on ambiguous cases flagged by the system rather than mechanical sorting. Classification accuracy reaches 92% for physical documents and 89% for emails, far superior to the previous estimated human error rate of 10%.
Lawyers can instantly search across all documents and emails by semantic content, for example finding all communications mentioning a non-compete clause even if exact terms vary. Average first response time to client emails decreased by 65%, from an average of 4 hours to 1.4 hours.
The firm also noted improved compliance, as all documents and emails are timestamped and tracked, facilitating audits. Finally, lawyer satisfaction greatly increased, as they appreciate receiving contextualized notifications with urgency indicators rather than having to manually search through folders and the email inbox.
Technical and Security Aspects
The technical architecture ensures total confidentiality. All Azure services used are deployed in the Switzerland North region, data never leaves Swiss territory. The GPT-4 model is deployed in private mode with Microsoft contractual guarantees that data is not used to train other models. Access is controlled by Azure AD with mandatory multi-factor authentication, and all actions are logged for audit.
The total monthly cost of the solution, including Azure licenses and houle support, represents about 25% of the saved salary cost, offering very rapid return on investment.
Conclusion
This automation demonstrates that it's possible to apply artificial intelligence to sensitive business processes including physical documents and electronic communications while maintaining total data control. The law firm has not only gained efficiency but also improved client service quality through shorter response times.