Morgan Stanley Cuts Accounting Work in Half with AI

Morgan Stanley implemented an AI agent system in profit and loss reconciliation (P&L adjustment), one of the most accuracy and speed-critical operations in banking. The result was that work which traditionally took up to 6 hours per ledger was reduced to 2-3 hours. Todd Johnson, Managing Director and leader of this initiative, revealed this achievement at a recent VB AI Impact event.

P&L reconciliation is the process of cross-checking data spanning multiple systems—finance, risk, operations, and trade records—at the end of each trading day, then correcting any discrepancies. Since each trade generates a vast array of attributes, hundreds of thousands of data mismatches occur by the close of each business day. Controllers must investigate each one individually before the next morning's deadline and determine and approve corrective actions. The time pressure is extreme, and the environment is highly prone to human error.

The solution deployed for this process is an internally developed agent system called FIXR. Once overnight P&L calculations complete, FIXR automatically analyzes data mismatches and proposes resolutions based on historical handling records. Multiple agents work in coordination internally, with each taking on roles such as interpreting past instructions to draft next-morning action plans, learning decision rules from controller behavior, and converting repetitive patterns into automated logic. For cases repeatedly resolved the same way, the system generates fixed rules and can handle them automatically going forward.

A distinctive feature is the intentional design choice to limit autonomy. FIXR requires human approval for all recommendations, and corrections and approvals by controllers are incorporated into the next cycle. Johnson stated that the system maintains "human responsibility as an element while advancing automation." He also characterized it as "closer to a colleague than a copilot."

In terms of actual results, Johnson explained that approximately 1,500 hours per week are being saved across the roughly 100 controllers engaged in P&L reconciliation. In simple terms, this amounts to roughly 15 hours of reduced work per person per week. Johnson emphasized that "building trust requires accumulated confidence," and indicated that the outlook is for an increasing number of cases that can be automatically processed. He stressed that "autonomy increases require trust to accumulate."

While many companies are leveraging AI for coding assistance and customer support, Morgan Stanley chose to apply it to a mission-critical process where precision errors directly translate to losses. This case supports the view that in AI agent implementation, "how to efficiently incorporate human judgment into the system" matters more than "how autonomous to make the system." Whether AI integration into mission-critical business processes will accelerate across financial institutions and how actual accuracy and audit compliance are evaluated will be key observation points going forward.