OpenAI’s September 10 release of ChatGPT for Financial Services packages financial data, connected sources, and analysis workflows together. The company is taking on setup tasks a finance team might otherwise have to assemble before using a general-purpose assistant.
That packaging deserves a workflow test. A finance team needs the right data, permission to use it, a suitable output format, and someone responsible for reviewing the result before it enters a forecast, memo, diligence process, or client recommendation.
Count the preparation work
To evaluate a tool, follow a recurring task through a normal working day. Record the effort required to find the current file, verify a figure’s source, assemble context, format the result, and explain it to the reviewer. Include that effort when comparing the AI-assisted process with the existing one.
MIT’s NANDA research reported that 95 percent of the enterprise generative AI pilots studied were failing to produce measurable return. That is a finding about the study’s pilots; it does not establish a failure rate for all AI or show that setup caused those results. It gives leaders a reason to inspect what happens between access to a tool and a usable output.
Repeated context assembly, manual source checks, and formatting can consume time an implementation was meant to save. Some checks are essential controls. The test is whether the proposed workflow supports them efficiently while producing work people can rely on.
Choose a task with an inspectable output
A narrower tool can be worth considering when its data connections and output match a recurring assignment. A broader tool requires someone to design and maintain that fit. Compare the full cost of those options, including review and exception handling.
For financial services, a trial could cover company research, market summaries, or background material for a client brief. In another business it might cover sales handoffs, contract review, support escalations, or client reporting. Choose an assignment whose current process the team understands.
Map the data required, where it lives, the output format, the review owner, and the evidence you will examine after two weeks. If the data sits in five disconnected places, include connection and permission work in the plan. Agree on the output standard before measuring the draft against it.
Use a result suited to the assignment: preparation time, handoff errors, review duration, or rework after manager review. Keep the review requirement consistent across the comparison. OpenAI’s product description identifies setup it aims to absorb; the trial needs to establish how much work remains for your team.
Sources: OpenAI, “Introducing ChatGPT for Financial Services,” September 10, 2026; MIT NANDA, “The GenAI Divide: State of AI in Business 2025.”