Anand and Anand

Anand and Anand, a full-service intellectual property law firm established in 1923, faced a challenge common to legacy professional services firms: a century of institutional knowledge locked in inaccessible formats. The firm held over one million case files totalling 80 terabytes of digitised data — yet none of it could be systematically queried or leveraged for client insights. In a fiercely competitive IP litigation market where clients expect near-instant assessments of case viability, junior lawyers were spending 3–4 days manually sifting through hundreds of pages per file just to surface relevant precedents. This bottleneck created a direct service gap: the firm could not rapidly substantiate win/loss probability assessments with material facts, putting it at a competitive disadvantage.

CIO Subroto Panda led the initiative, beginning with a foundational data-quality effort in February 2023: applying consistent metadata tagging, naming conventions, and tagging systems across the entire historical repository to make one million-plus files searchable. He then integrated Microsoft Copilot with custom machine learning algorithms trained on Delhi High Court IPR judgments and — critically — on the structured reasoning of the firm's own senior lawyers. Panda conducted extensive interviews with top legal experts to encode case-outcome logic directly into the algorithms' weighted keyword scoring model. The system went live in September 2023, generating concise case abstracts and win/loss probability assessments backed by supporting documents. To sustain accuracy, a mandatory quarterly feedback loop involving all users was enforced by top management, ensuring the model refines continuously as lawyers encounter novel case patterns.

The productivity impact was immediate and measurable. Case file review time dropped from 3–4 days to minutes, compressing a core research task by orders of magnitude and significantly accelerating new lawyer onboarding. The firm can now respond to client queries with AI-generated win/loss probability assessments substantiated by relevant precedents — a capability that was previously impractical at speed. Key outcomes include:

• 80 TB / 1M+ files made fully searchable through metadata standardisation
• Case research time reduced from days to minutes per file
• Faster client response on litigation viability, improving competitive positioning
• Reduced time spent on routine document preparation ('suit-making'), freeing lawyers for higher-value analytical work
• Quarterly model refinement cycle keeps predictions current as case law evolves

• Domain expertise must be encoded deliberately: the highest-value step was interviewing senior lawyers to capture their case-outcome logic — generic AI models without this input produced insufficient results for specialist IP litigation.
• Mandatory adoption beats voluntary uptake: top-management-enforced participation in the feedback loop overcame typical resistance to new technology in professional services environments.
• In-house deployment protects client confidentiality: keeping the model internal enabled faster iteration and avoided the data-sharing risks inherent in third-party legal AI platforms.
• Treat AI refinement as an ongoing process: as expert lawyers solve new classes of cases, those insights must be systematically fed back into the model — the feedback loop is the product, not a post-launch afterthought.