Orply.

OpenAI Pitches ChatGPT as Workflow Infrastructure for Financial Institutions

Stephanie AnaniOpenAIMonday, June 8, 20267 min read

OpenAI solutions engineer Stephanie Anani makes the case that ChatGPT should sit inside financial-services workflows rather than alongside them as a general productivity tool. Her argument is that AI can take on the search, reconciliation, modeling, compliance-checking and presentation work that consumes analysts’ time, while leaving investment and risk judgment with humans. In a QXO investment case, she shows ChatGPT moving from trusted research sources to an auditable Excel model and committee deck, using firm-specific skills and controls meant for regulated environments.

OpenAI is pitching ChatGPT as workflow infrastructure, not a productivity sidecar

Stephanie Anani frames ChatGPT’s role in financial services as a way to move regulated, information-heavy work into AI-supported workflows without removing the need for human judgment. The problem she identifies is not that employees are slow. It is that they spend too much time on work that has to be done but does not require the level of judgment firms need from them: searching policies and market data, reconciling numbers across systems, running compliance and risk checks, verifying regulatory documentation, chasing approvals, formatting reports, updating spreadsheets, and tracking regulatory changes.

The claim is that AI multiplies workforce impact by freeing employees to spend more of their time on decisions that drive business value. Anani cites a recent survey in which 75% of ChatGPT Enterprise users said AI helped them do tasks they could not do before. She distinguishes that from simple speed: the promise is not only that employees finish existing work faster, but that they operate at a higher capability level.

75%
of ChatGPT Enterprise users said AI helped them do tasks they could not do before, according to a survey Anani cites

That higher operating level depends, in OpenAI’s framing, on five listed supports: more context, deep research, shareable skills, intelligence everywhere, and governance and control. The financial-services argument is built around several of those supports appearing inside an analyst’s daily work: gathering trusted context, generating research, building models, applying firm-specific conventions, and producing decision materials that can be reviewed.

Trusted context has to meet the analyst inside the workflow

For financial services work, context starts with access to trusted, current information. OpenAI lists app connectors for Dow Jones, LSEG, and S&P as examples of sources used in the industry. The integration screen also includes Factiva, Moody’s, FactSet Research, PitchBook Data, Third Bridge, and MSCI Financial. Anani describes this connector work as “just the start.”

Placement matters as much as access. Much financial services work happens in Excel, so OpenAI has built ChatGPT into Excel, allowing users to interact with workbooks in natural language. The example shown is not a generic summary request; it asks ChatGPT to build a clean bridge from EBITDA to free cash flow conversion from 2024A to 2030E, with any assumption changes falling back onto a balance sheet tab.

The underlying model named in the source is GPT-5.5. Anani says OpenAI “went a step further” by embedding financial services processes directly into the model’s intelligence. To evaluate model performance on such work, OpenAI developed Banker Bench, a benchmark for financial services tasks. In the listed scores, GPT-5.5 registers 0.885, slightly above GPT-5.4 at 0.873 and above Opus 4.6 at 0.641. Anani describes GPT-5.5 as state of the art.

ModelOpenAI Banker Bench score
GPT-50.437
GPT-5.20.684
GPT-5.40.873
GPT-5.50.885
Opus 4.60.641
OpenAI Banker Bench scores listed in the source, where higher is better

The investment case turns research into an auditable model

Anani then applies the workflow to a QXO investment case. She plays an investment analyst at Blossom Bank who has been asked for “a fresh view on QXO” for an investment committee meeting the next day. The work begins with context gathering. In ChatGPT, Anani uses Deep Research and connects it to trusted sources and apps, including LSEG, PitchBook, Google Drive, SharePoint, and trusted sites.

The research prompt asks for an investment dossier on QXO focused on leadership, track record, strategic thesis, recent fundraising, and the Kodiak acquisition. It asks for primary sources where possible, including SEC filings, earnings transcripts, investor presentations, interviews, and shareholder letters. Before running the report, ChatGPT produces a plan that the user can edit.

The completed report is shown as taking 7 minutes, with 23 citations and 240 searches. Its summary says QXO is positioning itself as a tech-enabled consolidator in building products distribution, with an ambition to become a “tech-enabled leader” in a company-estimated roughly $800 billion industry and reach $50 billion of annual revenue within the next decade through acquisitions and organic growth. It identifies Beacon Roofing Supply, acquired in 2025, as the operating foundation for the platform strategy, and frames the current investment debate around repeated capital raises and the Kodiak acquisition.

7m
shown for the completed QXO research report, with 23 citations and 240 searches

The next step is turning research into numbers. Anani asks ChatGPT to generate a well-formatted three-statement model for QXO in Excel, incorporating recent financing, investment, and the Kodiak acquisition. She uses a “Blossom Bank 3 statement model” skill, which contains the firm’s preferred way to build three-statement models. The point of the skill is that the analyst does not have to restate the bank’s modeling conventions in every prompt.

ChatGPT generates an Excel workbook titled “QXO | Three-Statement Model,” prepared for Blossom Bank, with a model date of 2025-02-12 and a “special situations” template type. The summary describes an integrated 2025A–2030E three-statement model with explicit financing, deal, and Kodiak schedules. The output is presented as auditable rather than as a black box: ChatGPT leaves comments about assumptions and sources, and the numbers use formulas that can be inspected.

The model becomes a scenario engine for committee judgment

After research and base modeling, Anani adds management-interview context and asks ChatGPT to update the model for intrinsic value across bear, base, and bull cases. The bull case assumes QXO can double acquired companies’ revenue within five years of integration. The prompt also asks the model to reflect “platform optionality,” credit the existing business and announced deal, and treat remaining committed acquisition capital differently by case. The bear case reflects delays and partial undeployment of acquisition capital; the bull case reflects rapid deployment, integration, and stronger M&A outcomes.

ChatGPT returns an intrinsic valuation summary with common equity values of $5,881 for the bear case, $11,350 for the base case, and $15,700 for the bull case.

ScenarioCommon equity
Bear$5,881
Base$11,350
Bull$15,700
Intrinsic valuation summary generated in the QXO workflow

The explanation accompanying the output says ChatGPT mapped unpriced M&A out of the hard-coded base model and into a togglable optionality layer, made terminal multiple expansion flow through the scenarios, and converted revenue cases into explicit inputs rather than simple modifiers. It also says sources are documented and original formatting choices remain intact.

The significance, in Anani’s framing, is elapsed time and decision readiness. Work that “might have taken hours or days” can happen in minutes with ChatGPT and Excel, allowing analysts to “move almost as fast as the markets.” But the standard for the investment committee is not volume of output. The work still has to be reduced to signal and presented in a way that informs decisions.

Firm-specific skills turn analysis into a recommendation, with an audit trail

To create the investment committee deck, Anani uses a “Blossom Bank deck build” skill, which encodes how the firm likes to create presentations. She turns on extended thinking, attaches the QXO research markdown file and QXO valuation spreadsheet, and asks for a “rich and opinionated IC deck” based on the research and modeling work. Her prompt states that, based on the research, she thinks QXO is a good investment option and is optimistic that leadership will trend toward the bull scenario.

The generated deck’s title is “QXO’s right-tail is worth owning now.” The recommendation slide says: “Approve with conditions: QXO’s right-tail is worth owning now.” Its supporting sentence is more cautious: “Kodiak makes the platform credible, but the common only works if Brad Jacobs’ team converts dry powder into per share value faster than dilution.”

The deck also turns the model into committee-facing decision points. One slide frames the debate around whether today’s price already reflects the base case, with a base case of $18.30 per share versus $16 spot, a bull case of more than $33 per share, and a bear case of $7.50 per share. Another slide recommends approving a position now, adding only as execution confirms the bull path, and reducing or exiting if dilution happens faster than per-share value creation or discipline slips.

The presenter notes carry the auditability claim. They explain why the recommendation was made, state that the stock is compelling only if execution quality can push outcomes toward the bull case, and say the final recommendation reflects the uploaded research memo and valuation workbook rather than a new target generated outside the provided materials.

The point is not that AI has replaced the investment decision. It has connected the workflow: trusted context to deep research, research to financial model, model to decision points. The human role remains judgment. Anani’s closing claim is that AI changes the allocation of time: it frees up the day so people can spend more of it making those judgments.

AI does not replace the need for human judgment. It just frees up the day so that you can use the time to make those judgments.

Stephanie Anani

The frontier, in your inbox tomorrow at 08:00.

Sign up free. Pick the industry Briefs you want. Tomorrow morning, they land. No credit card.

Sign up free