Chatham Financial Scales Capital Markets Advisory with OpenAI: Codex Cuts Trade Validation From 30 Minutes to Under 4

What was announced

On October 2, 2026, OpenAI published a customer story about Chatham Financial, a firm that advises clients on complex capital markets decisions. In that line of work, accuracy and auditability come first. The headline number is simple: a trade validation application built with Codex cut review time from about 30 minutes to under 4 minutes per trade.

One caveat up front. This is a vendor-published case study. The figure comes from Chatham's early measurement, not from an independent evaluation. Chatham itself says it is still comparing the application's results with those of experienced reviewers, and that it will validate performance on real transactions before expanding automation. Read the 30-to-4 figure as a promising early signal, not a settled result.

Three workstreams, one method

Chatham's AI work falls into three parts. The first is employee-built applications, created on an internal platform called Chatham Vibes. The second is Chatham Onyx, the firm's next-generation capital markets operating system. The third is Process Zero, a reengineering consulting service for clients. All three share one idea: start from the outcome you want, find where human judgment matters, and then design the best way to deliver the work using the capabilities now available.

OpenAI plays two roles. Codex is the development tool used to build internal and client-facing software. The GPT-5.6 family powers the AI features inside those applications.

CEO Matt Henry put the attitude plainly. The firm is not treating AI as a faster way to run every process. It starts with the outcome, identifies where judgment matters, and designs delivery around that.

How Process Zero and trade validation work

Process Zero reengineers workflows around outcomes. For each workflow, Chatham identifies the minimum inputs and evidence required, determines where human judgment is essential, and decides how AI and AI-built tools should support the work. Trade validation is the first example. Chatham's Controls and Data Integrity team protects the accuracy of transaction data. It checks that each system record reflects what the client authorized and what was actually executed. Before, a person did this trade by trade, at roughly 30 minutes each.

Using Codex, Chatham built an application that does three things. It gathers supporting transaction evidence. It compares key terms. It flags discrepancies for review. The last step matters. The application flags, and a human reviews. That keeps the audit trail clean, because each finding can be traced back to specific evidence and a specific comparison. Alex Nordlinger, co-head of Chatham's AI Advisory practice, said early measurement showed review time falling from about 30 minutes to under 4. He added that validating the application against real transactions and experienced reviewers is just as important, and that expansion will wait for that evidence. Chatham plans to extend the application to more trade types and automate more of the workflow, while keeping appropriate controls and professional oversight.

Employee-built applications on Chatham Vibes

Chatham's client-facing AI work grew out of its own operating experience. Employees use ChatGPT and Codex in research, analysis, drafting, software development and other daily tasks. Through Chatham Vibes, an internal platform for creating applications tailored to their work, employees have also become builders.

People can build with a range of models. The AI features inside Vibes applications run on GPT-5.6 Terra by default, with GPT-5.6 Sol as an optional upgrade set per application. This is a common tiering pattern: a cost-appropriate default, with a stronger model when the task needs it.

These applications increasingly support client-facing workflows. The case study lists several. One helps teams review maturing-cap trades and prepare pricing workbooks and client communications. Others produce fixed-income rate sheets, prepare hedging dashboards and review trade confirmations. Chatham professionals still bring the market context, client understanding and judgment. They evaluate the output, refine it where needed, and decide what reaches the client.

Model routing inside Chatham Onyx

Chatham Onyx brings assets, debt and derivatives into one environment. Clients and advisors work from connected, governed data and can use AI without losing traceability to the underlying source. Chatham uses Codex across the Onyx development lifecycle. Teams use it to plan, build, test, document and review software, which shortens the path from concept to durable product capability. CTO John DeGuenther said Codex helps turn product vision into working capabilities more quickly, while the standards for accuracy, security and accountability stay the same.

The Onyx platform uses GPT-5.6 Sol, GPT-5.6 Terra, GPT-5.4 and GPT-4.1. Simple analysis and non-production testing go to the most cost-effective models. GPT-5.6 handles complex tasks where accuracy and value must be maximized. One example is ChatFIN. It summarizes patterns in historical market data, helps users understand their portfolios, and locates and links to legal documents for debt, derivative and lease terms. For enterprise developers, this routing is the useful part. Model choice becomes an engineering decision based on task risk and cost, not a single company-wide setting.

What it means for enterprises and developers

First, regulated and audit-heavy industries now have a reference pattern. Chatham does not treat AI as a black box that replaces experts. It uses AI for structured comparison, evidence organization and exception spotting, and keeps people in the judgment step. Co-COO Mike Noonan framed the constraint well: it has never been expertise, but the hours experts spend getting to the point where they can apply it.

Second, code generation tools are entering the full lifecycle of financial software. Codex is used for planning, testing, documentation and review, not only for writing code. Engineering teams need to keep review standards, test coverage and accountability steady even as the tools get faster.

Third, employee-built applications raise governance questions. A platform like Vibes speeds up delivery, but it also needs rules for model choice, data access and release review. The case study does not describe Chatham's specific controls here, so readers should note that gap.

Limits and outlook

Several points call for care. The 30-to-4-minute result is an early measurement. The sample size, trade types and comparison method are not public. The case study gives no error rate, miss rate or cost figure. Greater auditability and accuracy are stated goals, not published outcomes.

Next, Chatham plans to expand trade validation across more products, keep refining employee-built applications, and use Codex to help develop new Onyx capabilities. Two things to watch: when validation data is published, and how disagreements between human reviewers and the automated result are handled as automation grows.

For the wider industry, the case reflects a shift in the question. Firms are moving from asking how much faster an existing process can run to asking how the process should be designed if you start from the outcome. Whether that approach carries over to other financial workflows will depend on data quality, control frameworks and a clear line around where human judgment stays.