OpenAI launches ChatGPT for Financial Services aimed at Wall Street tasks

The system pulled an M&A target into view and surfaced peer comparisons. Within minutes the same window had assembled a valuation model, flagged a recent selloff and rebound, and dropped the full set into a slide deck. The process took roughly ten minutes.

Turley watched the output refresh and spoke over the screen. “It’s very easy to make slides that look good, but it’s much harder to make slides that actually make sense,” he said. The model had chosen the peers, pulled live prices into a spreadsheet, verified the chart against the underlying data, and written the explanatory text that linked the price movement to the reported events. Each step left a traceable line back to the original records.

The datasets that fed those citations arrived already indexed and stored inside OpenAI’s own systems. Daloopa, PitchBook and LSEG News supplied the underlying records on earnings transcripts, financial statements, company fundamentals and private-company filings. Once loaded, the material required no separate licensing agreements or custom connectors; an analyst simply opened the interface and the sources were present.

Three distinct routes supplied the rest of the information. The first route kept the three named premium sets bundled and resident on OpenAI hardware. The second allowed users to sign in with their existing credentials at S&P Capital IQ, LSEG, MSCI, Factiva or Moody’s so that any entitlements already held by the bank remained active inside the session. The third opened more than fifty additional MCP connectors, among them FactSet and Datasite, each optimized for the same retrieval layer.

Because the selected records lived on OpenAI infrastructure rather than being fetched at query time, the model could surface figures with lower latency and attach citations that pointed straight back to the original filings.

OpenAI shaped the final form of the product through repeated sessions with Morgan Stanley and Evercore, the two design partners. The partners singled out reliable data access and the creation of usable artifacts as the two problems that consumed the most hours.

Those sessions produced the current architecture. GPT-6 Astra, the newest release in OpenAI’s model sequence, anchors the work and is paired with templates that each firm uploads and controls so every slide, model and note already follows its own formatting rules and section order. Enterprise controls let compliance teams set role-based permissions, retain audit logs and keep all internal data from training the model by default.

Turley said the difference between a clean demo and an output that holds up in review came down to relying on the experts who would actually use it. The same controls also let administrators publish approved templates in advance so a new analyst opens the interface with the firm’s layout already in place.

PitchBook opened its first MCP connector to OpenAI in November 2025. LSEG added its own the following month. Both links let the model reach financial statements and transcripts, yet each still required banks to negotiate separate access and maintain live calls to outside servers.

Financial data moved to the center of the platform with the March 2026 release of GPT-5.4 and the accompanying ChatGPT for Excel. That launch demonstrated how successive model versions could pull figures into spreadsheets while preserving citations back to source filings. GPT-6 Astra extends the same line of work inside the new financial-services product, now running against records already indexed on OpenAI infrastructure.

The shift removed the need for repeated handoffs between systems. The current arrangement keeps the selected datasets resident where the model operates, so the same verification steps shown in the briefing room run without additional setup or latency.

The product is built to handle the research, data reconciliation and deck assembly that have long fallen to the newest analysts and associates on Wall Street. Those junior bankers have traditionally moved between filings, spreadsheets and presentation files to produce the first drafts of pitchbooks and models. ChatGPT for Financial Services collapses those steps into a single session that starts with approved data sources and ends with slides formatted to the firm’s rules.

Nick Turley described the scale of the shift during the briefing. “If you study the life of an analyst or of a banker, depending on the industry, they’re working 100-hour weeks,” he said. He compared the change to the arrival of Microsoft Excel decades earlier, when the spreadsheet replaced hours of manual work and let the same teams produce deeper analysis in less time.

The automation leaves the verification layer intact. Every figure pulled into the model or deck still carries a direct link to its source, so a senior banker can trace the output without rebuilding the work from scratch.

ChatGPT for Financial Services reached eligible institutions the same week it was shown in the briefing room. Banks contacted OpenAI directly or worked through their existing account teams to gain access, with no public pricing or seat minimums attached to the rollout.

Turley told reporters the company intended to build similar versions for other industries. He named no specific sectors or timelines, only that the financial-services release served as the first in a planned series of tailored offerings.