OpenAI shipped two enterprise products on the same Thursday, and together they describe a strategy more clearly than either does alone. One is horizontal: a Data agent inside ChatGPT Work that queries a company's warehouses and builds dashboards from a conversation. The other is vertical: ChatGPT for Financial Services, a tailored edition built with Morgan Stanley and Evercore that aims squarely at the research and pitchbook work junior investment bankers have done for decades.
Key takeaways
- The Data agent connects to Snowflake, Databricks, BigQuery, Redshift, ClickHouse, MongoDB and Datadog, and enforces each user's existing table-, row- and column-level permissions.
- ChatGPT for Financial Services pairs GPT-6 Astra with licensed data from LSEG, Daloopa and PitchBook that OpenAI indexes and hosts itself, enabling traceable citations.
- OpenAI's finance chief told investors in August that enterprise revenue now exceeds consumer revenue, and product VP Nick Turley says more industry editions are coming.
What the Data agent does differently
The pitch is the removal of a queue. Questions like why renewals slipped in the largest accounts, or where spending is climbing, normally mean filing a ticket with an analytics team and waiting. The Data agent instead answers inside a single conversation, without the user writing SQL or learning a separate BI tool.
What keeps that from being a demo trick is context. A generic agent pointed at a warehouse does not know what your company means by "active user" or how it books deferred revenue. OpenAI's answer is to read those definitions out of the semantic layer a company already maintains β dbt, Databricks Genie Ontology, Snowflake Horizon, GitHub, or the BI dashboards themselves β rather than asking anyone to restate them for ChatGPT.
Governance works the same way, by inheritance. Admins gate which connections exist and who may touch them, and beyond that the agent simply runs as the connected account, so table-, row- and column-level restrictions apply without a second permissions model to maintain.
Output lands as an interactive dashboard colleagues can edit and refresh on their own, and brand guidelines can be supplied so it does not look like a chatbot's homework. For teams already standardized on a BI vendor, the agent operates dashboards inside Omni, Oracle BI, Power BI, Sigma, Tableau and ThoughtSpot.
Why OpenAI started its vertical push on Wall Street
ChatGPT for Financial Services runs on GPT-6 Astra, the company's newest and most expensive model. The differentiator is native data: financial statements, earnings transcripts and private-market records from LSEG, Daloopa and PitchBook, indexed and hosted by OpenAI rather than fetched at query time, plus automated access to a firm's existing subscriptions.
Turley described the goal to reporters as teaching ChatGPT to research like an analyst and to back up its conclusions like one. In a live demo he had the product work up a potential acquisition target, pull comparable company prices into a spreadsheet, and generate a deck formatted to a bank's house style. His framing of the hard part was pointed: making slides that look good is easy, and making slides that actually make sense is not.
Granular citations that trace a figure back to a source filing, plus administrative controls for sensitive deal material, are the features that distinguish this from generic OpenAI enterprise tooling. The product is aimed initially at investment banking and equity research. Turley declined to name banks that have signed on.
The apprenticeship question nobody answered
Asked by CNBC whether the product reduces the need to hire junior bankers, Turley reached for the spreadsheet analogy, arguing that analysts already work 100-hour weeks and that the technology raises output per employee the way Excel once did.
That answer sidesteps a real concern inside the industry. Last month a Goldman Sachs partner running one of the bank's flagship AI projects warned that automating the tasks that train junior bankers risks cognitive atrophy in the next generation, since structuring an argument is a skill built by doing the grunt work, not by reviewing it.
What comes next
OpenAI is racing Anthropic and Google for enterprise share, and Anthropic launched its own Claude for Financial Services last year. The competitive context is also financial: with enterprise now outweighing consumer revenue and a widely expected IPO ahead, industry editions are a cleaner growth story than another consumer feature. Turley said tailored solutions for a number of other sectors are planned.
FAQ
Do I need to move my data for the Data agent to work?
No. The warehouses and observability stores listed above stay where they are, and the agent reads them through connectors that inherit whatever permissions the querying account already holds. Document stores such as Google Drive and SharePoint can be folded into the same analysis.
Does ChatGPT for Financial Services include market data?
Yes, and the licensing arrangement is unusual: OpenAI hosts the premium feeds on its own infrastructure instead of calling out to a vendor per query. That is what makes figure-level citations possible, and a firm's existing subscriptions remain usable alongside it.
Who is the financial services edition for?
Investment banking and equity research desks are the first target, shaped during development by two design partners on the sell side. OpenAI has not said which institutions have actually bought the product.
