The bottleneck is often not data access. It is the waiting.
A manager asks for a number. An analyst pulls it from one system, checks it against another, builds the chart, then explains why the first answer was incomplete. By the time the question is answered, the decision has moved on.
OpenAI’s Data agent points at a better pattern for that kind of work. In ChatGPT Work, people can ask business questions, investigate changes, and turn the analysis into an interactive dashboard in one conversation. It can also work with connected BI tools such as Power BI, Tableau, Sigma, and others. The important part is not the interface. It is that the analysis stays tied to the same connected account permissions already in place.
That changes the operating model. Instead of routing every question through a reporting queue, teams can let business users explore within guardrails while keeping table, row, and column access intact. In plain English: people can self-serve without being given more data than they should see.
The design choice is where the governance lives. If you want self-serve analysis, the first question is not “Can the model answer it?” It is “What data can it touch, who can share the result, and what happens when the dashboard needs to be refreshed or reviewed?” OpenAI’s business data policies and workspace controls suggest that those limits are part of the product, but they still need to be configured and managed as part of the workflow.
For operations teams, that is the useful threshold: fewer ad hoc report requests, more answerable questions, and a clearer path for review when the data is sensitive or the decision is consequential.