BNY Mellon

BNY Mellon's in-house legal team operates under the dual pressure common to financial services: a high volume of routine contract review work and a regulatory environment that demands rigorous oversight at every step. For a global custodian bank managing trillions in assets, the legal department handles a continuous flow of vendor agreements, client contracts, and financial documentation — work that is largely repetitive but carries significant compliance risk. Relying on attorneys to manually review each document consumed substantial billable-equivalent capacity, created throughput bottlenecks, and left little bandwidth for the strategic legal counsel the business actually needed. The status quo was expensive, slow, and difficult to scale.

BNY Mellon's legal department developed an internal generative AI platform named Eliza, built in collaboration with OpenAI and grounded in large language model technology. Rather than deploying an off-the-shelf legal tech tool, the team chose a bespoke internal platform to retain full control over data handling, model behavior, and governance guardrails — a critical requirement in a regulated financial services environment. Eliza automates repetitive contract review tasks including clause identification, risk flagging, and document summarization, integrating directly into the legal team's existing workflow. The internal build approach allowed the team to design governance protocols first and layer AI capabilities around them, ensuring compliance requirements were embedded into the system architecture rather than bolted on after deployment.

The deployment of Eliza delivered a 75% reduction in average contract review time, a result that substantially changes the economics of in-house legal operations. Time previously absorbed by manual document review has been reallocated to higher-value work — strategic analysis, negotiation, and counsel that requires human legal judgment. Key outcomes include:

• 75% faster average contract review cycle
• Legal professionals redirected from routine processing to strategic advisory roles
• AI governance framework validated as production-ready in a regulated enterprise context

The implementation also demonstrated that a robust internal governance model could serve as an accelerant rather than a barrier to AI adoption.

• Governance first, deployment second: designing compliance and data controls into the platform architecture — not as an afterthought — was the prerequisite for gaining internal approval in a regulated environment.
• Internal platforms outperform off-the-shelf tools for governance-sensitive use cases: a bespoke build gave BNY Mellon control over model behavior, data residency, and audit trails that commercial products couldn't match.
• Foundation model partnerships require clear scope: collaborating with OpenAI on an internal tool required clear boundaries around data access and model customization.
• Automate the repetitive, preserve the judgment: the strongest ROI comes from targeting high-volume, low-discretion tasks — leaving complex legal reasoning to attorneys.