Chatham Financial says OpenAI's tools cut one of its slowest manual checks down to minutes.
The derivatives and capital markets advisory firm used Codex and GPT-5.6 to build internal software and redesign its workflows. The headline number: a trade validation process that used to take 30 minutes now takes under four. Codex handled the technical build; GPT-5.6 was used across the broader workflow redesign. Missing from the announcement are the basics that would let anyone judge the claim - trade volume, build cost, or how much human review still sits behind that four-minute mark.
Trade validation is a rules-heavy, repetitive task - precisely the kind of work large language models are good at, and precisely the kind vendors like to showcase because the before-and-after numbers look dramatic. A time cut that steep on one internal process says less about AI capability broadly than about how much manual busywork capital-markets back offices still run on spreadsheets and email.
It's also another entry in OpenAI's growing stack of enterprise customer stories, each pairing a steep efficiency number with thin operational detail - a format that says more about OpenAI's enterprise sales pitch than about independently verified results.