OpenAI introduced a new Data agent in ChatGPT Work on September 10, 2026. It can connect to approved company data, investigate a business question, and build an interactive dashboard. For a small business, the right first move is not to connect everything. Test one question where the answer can change a real decision.
In its official announcement, OpenAI says the Data agent can work with approved sources such as Google BigQuery, Databricks, MongoDB, Snowflake, Google Drive, and SharePoint. It also says admins control connections and existing permissions continue to apply. Those are product claims from OpenAI, not an independent accuracy audit.
What the Data agent changes
The useful shift is from waiting for a report to asking a focused question in plain language. OpenAI says the agent can investigate changes, show evidence, build, edit, share, and refresh dashboards, then take approved actions through connected tools.
That does not mean a small company needs a data warehouse. If your source is inconsistent, the agent can only work with what it can access and how your team defines its metrics. That is an inference from the setup OpenAI describes.
Our guide to Google Sheets Canvas for small business covers a simpler path when most of the answer already lives in one Sheet. The Data agent makes more sense when the question crosses sources or needs follow-up.
Pick one decision, not a dashboard
- Choose one decision, such as which service generated the most qualified inquiries last month.
- Name the source and define the metric in words your team already uses.
- Ask for the evidence before you ask for a recommendation.
- Compare the answer with a report or calculation you already trust.
- Decide which action, if any, a person may approve next.
A dashboard can look useful before the underlying question is clear. Start with the decision. The view should help someone make it, not create another screen to check.
A practical first prompt
Prompt: Use only our approved sales and lead data. Compare qualified inquiries from the last 30 days with the previous 30 days. Show the source fields, assumptions, and any missing data. Do not recommend an action until I approve the numbers.
Replace “qualified inquiry” with your real definition. If you need to tighten that definition first, read our guide to what makes a lead qualified. Clear rules make the result easier to review.
What to check before you trust the result
Source access
Confirm that the connection is approved, the account has the right permissions, and the data is current enough for the decision.
Metric definition
Write down what counts as a qualified lead, revenue, or booked appointment. Do not assume the agent will infer the rule your team has never documented.
Evidence
Ask for the fields, filters, date range, and assumptions behind the answer. If a person cannot explain the result, it is not ready to drive an action.
Action boundary
Keep the first test in analysis mode. A human should approve customer messages, CRM changes, budget changes, and other actions that affect the business.
Where it fits with automation
Our guide to AI workflow automation starts with one job, one trigger, one review step, and one measure. The same discipline applies here. A clear question and a short review path are more useful than a large connected system.
If the next step is an agent that can use business tools, rather than only inspect data, read our guide to the OpenAI Agents API for small business. It covers the tests to run before you build around real actions.
If the question includes caller or lead records, read our guide to AI receptionist CRM integration before connecting transcripts or contact details. Decide what should be saved before you ask an AI system to analyze it.
What it does not replace
The Data agent does not replace clean definitions, access decisions, or owner judgment. OpenAI's page describes context, permissions, and safeguards as part of its internal use of data agents. A small team should treat those as setup, not an afterthought.
Before you connect sensitive customer information, use our questions about AI data privacy as a starting point for your own review.
Bottom line
OpenAI's Data agent is worth a focused test when a real business question is spread across approved sources. Start with one decision, show the evidence, compare the result with a known answer, and keep the final action with a person.
Want help choosing a practical AI workflow for your business? Book a free Leadspa consultation.