Morgan Stanley has rolled out an internal AI system called FIXR to automate profit‑and‑loss (P&L) reconciliation, halving the time required for each book from up to six hours to two‑to‑three hours. The speedup translates to roughly 1,500 hours saved each week across the bank’s 100 controllers.
FIXR does not operate as a fully autonomous copilot. Instead, several specialized agents analyze nightly P&L “breaks,” propose resolutions, and generate repeatable rules. Controllers review every recommendation, approve or correct it, and feed the outcome back into the agents, which then codify the decision into durable logic. Over time the system can auto‑clear recurring breaks, suggest solutions for new patterns, or flag items for human investigation.
Johnson stressed that the human‑in‑the‑loop design preserves accountability and builds trust, a prerequisite for scaling automation in a regulated environment. The team first mapped the end‑to‑end workflow, deciding where agents, traditional automation, or process redesign would add the most value before introducing AI. Morgan Stanley’s approach mirrors VentureBeat’s survey findings that many enterprises see limited ROI from custom model fine‑tuning and struggle with governance, suggesting that a process‑first, rule‑driven strategy may be more sustainable.
