How Morgan Stanley cut P&L reconciliation time with AI agents and human oversight

Morgan Stanley’s deployment of AI agents in profit-and-loss (P&L) reconciliation turned industry assumptions on their head. While most enterprises focus AI on coding or customer service, the financial giant targeted one of banking’s most exacting, deadline-driven processes—and achieved dramatic efficiency gains not by maximizing autonomy, but by tightly integrating human oversight.

At a recent VentureBeat AI Impact event, Morgan Stanley Managing Director Todd Johnson described the internal system, FIXR, as “much more like a co-worker than a copilot.” Unlike early generative AI tools that automate simple tasks, FIXR operates within a complex, regulated workflow where accuracy and accountability are non-negotiable. “We think that's where the opportunity is to really unlock more complex work in the organization,” Johnson noted.

Inside FIXR: Balancing automation with human judgment

Every trading day at Morgan Stanley ends with a meticulous reconciliation of P&L across Finance, Risk, Operations, and Trade Capture systems. The challenge? Hundreds of thousands of data points often fail to align, forcing controllers to manually investigate mismatches, make judgment calls, and approve adjustments—all before a strict morning deadline.

Before FIXR, reconciling a single book could consume up to six hours. Now, the system completes the task in two to three hours. Across approximately 100 controllers globally, that translates to roughly 1,500 hours saved per week. The efficiency gain stems from FIXR’s layered approach: it doesn’t just automate—it learns, adapts, and codifies human decisions into repeatable rules.

• One agent analyzes overnight P&L breaks and proposes resolutions based on accumulated patterns.
• Another learns from controller actions, documenting the logic behind each adjustment.
• A third converts recurring fixes into durable automation rules.

Over time, the system can auto-resolve familiar breaks, suggest solutions for ambiguous ones, escalate uncertain cases for review, and even generate firm-wide rules when resolutions become standardized. Crucially, every recommendation is reviewed, approved, or corrected by a human before execution. Controllers then feed their feedback back into the system, enabling FIXR to refine its logic daily.

“You still preserve that element of human accountability even as you start to automate,” Johnson emphasized. “Over time you'll see more and more of those items resolved in an automatic way.”

Process-first automation: Why governance matters more than autonomy

Morgan Stanley’s approach prioritized process clarity before introducing AI. The team conducted a rigorous process intelligence assessment to identify inefficiencies ripe for automation—whether through agents, traditional scripts, or workflow redesign.

“If we can fix that first before we add agents to the problem, then we really will be transforming the opportunity,” Johnson said. The P&L sign-off process, riddled with manual steps, was an ideal candidate. By offloading repetitive tasks, controllers now focus on higher-value analysis and risk assessment.

The team also deliberately minimized reliance on model-driven judgment where deterministic logic sufficed. “If you have an opportunity to make things very prescribed and repeatable, that's cheaper in terms of token consumption, it's more repeatable in terms of controls—and have the LLM do the stuff where you don't need that kind of deterministic workflow,” Johnson explained.

As FIXR encounters repeated fixes for specific break types, it converts those patterns into fixed rules rather than deferring to the model. This hybrid design reduces variability and strengthens compliance.

Agents as digital colleagues: Redefining accountability in AI governance

A core question in the agentic AI era is whether agents are best viewed as automated code or as digital employees. Johnson argues for a balanced perspective: “They're probably a little bit of both.” This duality demands nuanced governance—technical teams must maintain infrastructure safeguards like firewalls and encryption, while human users retain responsibility for agent performance within their workflows.

For example, a senior controller guiding a junior colleague retains accountability even when using an AI assistant. “One of our strong principles in our AI governance generally is that there always has to be human accountability, even if there's a degree of automation,” Johnson stated.

This hybrid model ensures that automation scales without compromising control. By embedding human-in-the-loop discipline from the outset, Morgan Stanley transformed a labor-intensive process into a repeatable, auditable system—one that continues to evolve with each controller’s input.

As generative AI reshapes enterprise workflows, FIXR offers a blueprint: pair intelligent agents with human oversight, prioritize process clarity, and let iterative learning drive sustainable transformation. The result isn’t just faster reconciliation—it’s a new standard for responsible automation in high-stakes environments.