An AI agent supporting the legal department: faster contract review, with decisions always made by humans
Can AI read a contract like an experienced lawyer, without ever taking the place of that lawyer's decision? For the legal department of a multinational manufacturing company, the answer is yes.
Overview
The Legal Affairs department of a multinational manufacturing company handles a significant volume of contracts and negotiation documents, involving high-value but repetitive review work in the initial stages.
Digital Attitude has designed and released Legal Agent: an AI assistant, available as an app within Microsoft Teams, that analyzes contractual documents, identifies critical clauses, and proposes a reasoned markup based on the company’s standards and policies.
The agent is built on Copilot and Azure OpenAI and uses semantic search of a legal knowledge base: it highlights risks and suggests annotations, but never modifies the body of the contract on its own; each suggestion remains a recommendation that the legal professional evaluates and approves.
Developed in just a few weeks as an MVP, Legal Agent demonstrates DA’s “Human-in-the-Loop” approach: AI accelerates and standardizes the preparatory work, while decision-making remains entirely in human hands.
Solution
We have developed a Legal Agent integrated into Microsoft Teams to assist the legal department in analyzing and annotating contracts. The engine, based on Copilot and Azure OpenAI with semantic search of the corporate knowledge base (Azure AI Search), operates according to two fundamental principles: non-destructive markup, which always preserves the original text by adding annotations on a separate layer, and the Human-in-the-Loop approach, whereby the final review remains the responsibility of legal professionals in all cases.
The agent does not replace the legal department; rather, it supports it by accelerating the preliminary analysis. Outcome KPIs (average review time, clauses identified) will be measured once the system is fully implemented.
Next Case Study
