SDSignal Desk

Chatham Financial cuts trade validation from 30 minutes to under 4 with a Codex-built app

· Updated · Signal Desk newsroom

Analysis of reporting by OpenAI. How we work

Image: OpenAI

In an OpenAI customer story, Chatham Financial says a Codex-built trade validation app cut review from about 30 minutes to under 4, with results still checked against experienced reviewers. GPT-5.6 powers features across its Onyx platform and employee app builder.

Key takeaways

  • A Codex-built app gathers evidence, compares terms and flags discrepancies in trades.
  • Early measurement shows review falling from about 30 minutes to under 4.
  • Results are compared against experienced reviewers before broader automation.
  • Onyx routes simple work to cheaper models and heavier tasks to GPT-5.6.
  • Employees build apps in Chatham Vibes, with GPT-5.6 Terra as default.

What Chatham built

OpenAI’s customer story says Chatham Financial, which advises on complex capital-markets decisions, uses Codex to build internal and client-facing tools and GPT-5.6 to power AI features across employee apps, its Chatham Onyx capital-markets platform and its Process Zero consulting practice. A Codex-built trade validation app gathers evidence, compares key terms and flags discrepancies. Early measurement shows review falling from about 30 minutes to under 4, with results still checked against experienced reviewers before broader automation.

Employees build tools through Chatham Vibes, with GPT-5.6 Terra as the default and Sol optional. Onyx routes simple work to cheaper models, including GPT-5.4 and GPT-4.1, and heavier tasks to GPT-5.6; a feature called ChatFIN summarizes market patterns and links legal documents. CEO Matt Henry describes starting from outcomes and where judgment matters rather than speeding every process indiscriminately.

Assembly versus judgment

The pattern is sound: shrink assembly work, keep people on judgment and measure against real transactions before widening automation. Capital markets depend on auditability, and comparing the app with experienced reviewers is what turns a demo into a control. Chatham’s Mike Noonan says the constraint was never expertise but the hours needed to reach the point of applying it.

Codex as a software factory for Process Zero and Onyx is the quieter story. If domain firms can turn proven checklists into governed apps for reviews, rate sheets, hedging dashboards and confirmations, experts can spend more time advising. OpenAI has similar finance stories and products (ChatGPT for Financial Services, Jump Trading).

Risks and what would prove it

Fast validation that drifts from reviewer standards becomes fast error with a clean interface. Employee-built apps can multiply shadow IT unless security and model routing stay centralized. Traceability in Onyx and ChatFIN matters only if links to source documents survive every summary step.

This is a vendor case study with early, company-reported metrics. Published error rates, override frequency and audit findings as automation expands to more trade types would show whether the approach holds. If those numbers stay good, the template is worth copying by other advisory firms. More in our finance coverage, OpenAI coverage and enterprise coverage.

Who feels it

Capital markets and risk teams
Outcome-first redesign with human checkpoints is a reusable playbook.
Internal platform builders
Employee app tools need default tiers, review standards and audit hooks.
OpenAI enterprise buyers
Ask for time saved plus validation against experts.

What to watch

  1. Expansion to more trade types with published accuracy and override data.
  2. How Onyx preserves document traceability.
  3. Whether other firms copy the evidence-and-judgment template.

FAQ

What did Chatham build?

A Codex-built trade validation app that compares terms and flags discrepancies.

How much time does it save?

Review fell from about 30 minutes to under 4 in early measurement.

Is it fully automated?

No, results are checked against experienced reviewers.

Which models are used?

GPT-5.6 for heavier tasks, with cheaper models like GPT-5.4 and GPT-4.1 for simple work.

Read the original

Continue at the source.

OpenAI

Companies: OpenAI

More on OpenAI

All OpenAI coverage →