Morgan Stanley cut its riskiest reconciliation job in half — by making its agents less autonomous

Morgan Stanley has deployed an internal production agentic system, FIXR, to automate profit and loss (P&L) reconciliation, cutting the work in half by making the system less autonomous and keeping humans tightly in the loop. FIXR uses a human-agent feedback loop to learn from controllers and codify their decisions into repeatable rules, preserving human accountability while automating tasks. The system consists of multiple agents that work together to analyze breaks, propose resolutions, and learn from controller behavior. This approach has resulted in significant time savings, with approximately 1,500 hours saved per week across 100 controllers. The practical implication for engineers building AI systems is that a process-first approach, focusing on establishing processes before getting AI involved, can lead to more effective and efficient automation.

⚡ Key Takeaways

• FIXR, Morgan Stanley's internal production agentic system, has cut P&L reconciliation work in half.
• The system uses a human-agent feedback loop to learn from controllers and codify their decisions into repeatable rules.
• Multiple agents work together to analyze breaks, propose resolutions, and learn from controller behavior.
• The system preserves human accountability while automating tasks, with humans reviewing and approving every recommendation.
• Establishing processes first, before getting AI involved, was critical to the success of the system.

The deployment of FIXR by Morgan Stanley demonstrates the potential for AI to automate complex, deadline-driven workflows in the financial industry, while also highlighting the importance of human oversight and accountability. This approach can serve as a model for other enterprises looking to leverage AI in similar contexts.

✅ Practical Steps

1. Establish processes first, before getting AI involved, to ensure a solid foundation for automation.
2. Implement a human-agent feedback loop to learn from human decisions and codify them into repeatable rules.
3. Use multiple agents to work together to analyze breaks, propose resolutions, and learn from human behavior.

Want the full story? Read the original article.