What changed
OpenAI’s new customer story offers concrete examples of proactive assistance at the UK travel technology company. It reports that Liam Thompson, VP of AI Innovation, received an unsolicited app-launch performance report from his dot, Beast, which retrieved data from Google BigQuery through a ChatGPT plugin. David Lee, Director of Innovation & Growth, describes talking through upcoming meetings during his commute and requesting reports using connected company data. OpenAI also says his dot, Brian, flagged API keys that needed retiring and searched his email for people who might be affected.
This is a newly published use case, not a dots launch announcement. OpenAI introduced dots on September 29 as agents that can work between conversations, connect to authorized apps and communicate by voice or message.
Why it matters
The useful idea is not just faster answers to questions: an assistant could bring relevant information to someone before they request it. Launch monitoring and meeting preparation are recognizable workplace tasks. Holiday Extras’ account illustrates that possibility, but does not demonstrate how consistently it works or whether it improves decisions.
Who it affects
Product and operations teams exploring reporting assistants, plus workplace AI administrators assessing what information and permissions an agent would need. This is not an implementation guide or a recommendation to grant broad account access.
What you can do with this
If your organization permits dots, ask the workspace administrator to confirm access and approved data connections. Start with one bounded report, such as yesterday’s usage of a specific app feature, and define its metrics and data window. Check the result against the underlying data before using it. Record incorrect or missed flags alongside useful ones. Keep consequential actions subject to review rather than assuming an accurate report authorizes the agent to change systems. These are suggested evaluation steps, not a trial performed by AI Unpacked.
What we don’t know
This vendor-published case study contains no independent accuracy audit, error rate or controlled productivity comparison. Its examples are not proof of universal monitoring or security. OpenAI says dots can make mistakes; it describes background proactive research as read-only work through connected apps, not unrestricted autonomous action.
Access also depends on workspace eligibility and administrator settings. OpenAI’s November 4 Enterprise release notes describe general availability with an admin preview still in effect as of this November 17 review; the launch page’s older Enterprise beta wording should not be treated as current. No price or universal availability claim is made here.
SHOWING OUR WORK
Go straight to the source.
Facts are grounded in the references below. Practical implications are editorial analysis unless explicitly attributed to a source.
01See claim-level references
OpenAI published the Holiday Extras dots customer story on November 16, 2026. Source
OpenAI describes Holiday Extras as a UK travel technology company. Source
OpenAI reports that Liam Thompson’s dot, Beast, retrieved an unsolicited app-launch performance report from Google BigQuery through a ChatGPT plugin. Source
David Lee describes discussing upcoming meetings during his commute and requesting reports using connected company data. Source
OpenAI says Brian flagged API-key retirement and searched David Lee’s email to identify potentially affected people. Source
The customer story provides anecdotes rather than an independent accuracy audit or controlled productivity comparison. Source
OpenAI introduced dots on September 29, 2026, describing agents that can work between conversations, connect to authorized apps and communicate by voice or message. Source
OpenAI describes background proactive research as read-only work through connected apps and warns that dots can still make mistakes. Source
OpenAI’s November 4 release-note entry describes general availability for eligible Enterprise, Edu and Healthcare workspaces, administrator controls and an admin preview until November 18, 2026. Source