Morgan Stanley halves P&L reconciliation work with AI system emphasizing human oversight

Morgan Stanley has significantly improved its accuracy-critical, deadline-driven profit and loss (P&L) reconciliation workflow with an internal AI system called FIXR. The bank reports that this has cut the work in half, achieved not by increasing the system's autonomy, but by strengthening human oversight and iterative rule-building.

Traditionally, P&L reconciliation involves extensive manual investigation of discrepancies across hundreds of thousands of data attributes from various systems. This process often took up to six hours per book before a morning deadline. FIXR now completes this task in two to three hours daily, saving approximately 1,500 controller hours per week across the roughly 100 employees involved.

The FIXR system automatically analyzes identified differences, known as "breaks," and proposes resolutions. Specialized agents interpret past guidance, learn from human controller actions, and convert recurring patterns into durable, automated logic. The system can auto-clear familiar breaks, suggest solutions for less common ones, or flag items for human review when uncertain. Crucially, all recommendations are reviewed and approved by humans, whose decisions feed back into the system to improve its performance daily.

According to Morgan Stanley, this approach maintains human accountability while enabling controlled automation. The strategy prioritizes establishing robust processes before AI implementation and focuses on extensibility across global operations. The system is designed to be deterministic where possible, relying on fixed rules derived from human feedback rather than full AI judgment for complex tasks, thus ensuring control and efficiency.