OpenAI launches ChatGPT for Financial Services, pairing bank and provider data to power one AI tool for finance teams

OpenAI's own worked example tells you which desk this is aimed at. "For example, for an acquisition, they can compare the target with its peers, and test how revenue growth affects valuation, and turn the entire analysis into an editable model or pitchbook using their firm's templates," the company wrote in a blog post ^[16]. Peer financials come from a licensed feed. Whether that workflow runs at your firm depends on the terms of the licence.

Turley told reporters the offering feels similar to ChatGPT Work with a "finance-specific bend" ^[23], and the interface looks like standard ChatGPT with extra toggles for data sources ^[10]. The data itself sits in two piles. OpenAI pre-loads four sets: Crunchbase, Pitchbook, Daloopa and LSEG News ^[11][22]. Bloomberg and FactSet are in the other pile, described as sources a bank connects itself alongside its own data ^[10], with about 50 Model Context Protocol connectors covering other systems and third-party software ^[12].

That split sets the ceiling on any pilot. If the comparables your bankers actually cite live in a terminal whose contract will not allow piping into a third-party model, what you are testing is four vendor datasets plus your internal files: a useful test, and a narrower question than the pitchbook example suggests.

Turley said he worked with large financial institutions including Morgan Stanley and Evercore to work out what finance teams need ^[7]. He said there is a big gap between what looks good in a demo and what is actually a usable output, and that you rely on the experts to close it ^[8]. For anyone who has to defend a page to a managing director, the visible product of that work is sourcing: Turley said the tool provides detailed citations whenever it uses the connected data ^[13].

The ambition he described goes past speed. "We're trying to think of new ways of doing the work, rather than just making the existing ways faster," he said ^[17]. His example was a ChatGPT-coded dashboard where someone types in a condition and gets a revised forecast instead of adjusting numbers in a spreadsheet ^[25]. That is the work juniors do, and SiliconAngle reported concerns that the product might encourage banks to hire fewer junior staff ^[18].

It has been roughly 16 months since Anthropic shipped Claude for Financial Analysis ^[21]. That tells you Anthropic got there first, and nothing about how either product performs. The announcement carries nothing on how many banks run either product, or on what buyers score them against. SiliconAngle described industry-specific packaging of frontier models as a trend aimed at targeted enterprise revenue ^[19], and Turley named financial services, software engineering and cybersecurity as the verticals OpenAI is prioritising most ^[3].

So the sort before a pilot comes down to two things: whether the data a candidate workflow needs is already connected, and whether its output goes to a client. Internal work on connected data is where a month of use produces a verdict you can take to a steering committee. Client-facing work on data you cannot connect produces a demo, and then a slide about the demo. OpenAI is aiming at the whole grid, saying this will be the "one product" that banks and financial firms with thousands of workers ever need ^[6]. "This is the canonical product we are hoping the industry adopts," Turley said ^[5].