Automating Complex Reconciliation Workflows with FIXR
Morgan Stanley has successfully deployed a production-grade multi-agent artificial intelligence system called FIXR to automate one of banking’s most demanding daily workflows: profit and loss (P&L) reconciliation. The system has cut the time required for this task from six hours down to two to three hours per day, saving approximately 1,500 hours per week across around 100 controllers. The development was detailed publicly by Todd Johnson, Managing Director at Morgan Stanley, and represents one of the most credible, well-documented examples of agentic AI operating in a genuinely high-stakes enterprise environment.
What makes this case study particularly instructive is the counterintuitive design philosophy driving it. Rather than maximising the autonomy of its AI agents, Morgan Stanley deliberately constrained them. The system uses large language models (LLMs) to analyse patterns and suggest resolutions, but those suggestions are validated by human operators before being converted into fixed, deterministic rules. It is the rule engine, not the AI, that executes repeating tasks going forward. This distinction between probabilistic AI assistance and deterministic rule-based execution is the central technical insight behind FIXR’s success.
For professional services firms operating in regulated, accuracy-critical environments, including environmental consultancies, engineering practices, legal firms, and financial advisers, this case provides a practical and replicable blueprint. The challenge of deploying AI in contexts where errors carry serious legal, financial, or regulatory consequences is directly analogous to what many consultancies face when considering AI-assisted reporting, data review, or compliance workflows.
Key details of the FIXR multi-agent architecture
At the end of every trading day, Morgan Stanley’s finance controllers must reconcile P&L figures across four separate internal systems: Finance, Risk, Operations, and Trade Capture. Discrepancies between these systems, referred to internally as “breaks,” can occur across hundreds of thousands of individual data attributes. Historically, identifying and resolving these breaks required controllers to manually investigate each mismatch and apply corrective adjustments under tight morning deadlines, a process that consumed up to six hours per book. The combination of high data volume, strict accuracy requirements, and hard deadlines made this workflow a significant operational bottleneck.
FIXR addresses this through three specialised agents working in sequence. The first, the Interpreter, analyses historical guidance and past resolution records to propose resolutions at the start of each day. The second, the Observer, monitors how human controllers actually respond to breaks in real time, documenting the specific adjustment logic they apply. The third, the Builder, takes those observed and validated patterns and converts them into durable, automated rules that execute deterministically in future cycles. The system does not replace the controller’s judgement. It captures that judgement, formalises it, and automates its repetition.
The economic rationale for this architecture was stated directly by Todd Johnson at a recent industry event. His explanation was: “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.” This framing is significant because it rejects the common assumption that more capable or more autonomous AI agents are necessarily better. In a compliance-driven context, predictability and auditability outweigh flexibility.
The operational outcome is measurable and specific. Across approximately 100 controllers, FIXR has freed roughly 1,500 hours per week that was previously consumed by manual reconciliation tasks. Controllers now spend that reclaimed time on higher-order work including deeper risk analysis and exception management, activities that genuinely require human judgement and cannot be systematised. The token-cost efficiency of the deterministic rule approach also means the system scales without proportionally increasing AI inference costs, which is a material consideration for any enterprise deploying LLMs at volume.
Australian professional services context and implications for regulated industries
The professional services sector in Australia is navigating the same fundamental tension that Morgan Stanley has resolved through FIXR: how to deploy AI in workflows where errors carry real consequences. In environmental consulting, those consequences can include regulatory non-compliance, personal liability for practitioners, invalid site audit statements, or contaminated land transactions that later unravel. In legal practice, they can mean professional indemnity claims. In engineering, they can affect public safety outcomes and development approvals. The Australian regulatory environment across these sectors imposes a level of accountability that makes unconstrained AI autonomy genuinely untenable.
Australian professional and licensing frameworks reinforce this constraint. Environmental auditors accredited under state EPA schemes, for example, carry personal liability for their audit conclusions. A Certified Environmental Practitioner (CEnvP) providing expert opinion must be able to demonstrate the basis for every conclusion. Introducing a probabilistic AI into that chain of reasoning without rigorous human verification and a documented decision trail creates a professional liability exposure that most practitioners would not accept, and that most regulators would not recognise as an adequate basis for a statutory deliverable. The Morgan Stanley model, where AI observes and suggests but humans validate and rules execute.
References and related sources
- Primary source: venturebeat.com
- healthbitintel.com
How iEnvi can help
iEnvi integrates technology and data-driven approaches into environmental consulting. We monitor AI and technology developments that affect how environmental professionals deliver services to clients.
This is an iEnvi Machete news summary. Prepared by iEnvi to summarise the source article for environmental professionals tracking AI, data, and technology developments that affect consulting and project delivery.
Published: 03 Jul 2026
Need advice on this topic? Speak to an iEnvi expert at info@ienvi.com.au or 1300 043 684, or contact us online.
Need advice on this issue? iEnvi provides practical, senior-led environmental consulting across contaminated land, remediation, ecology and environmental risk.