Batch testing is a tool in customer agent that lets admins test their AI agent against a set of questions before deploying changes, review every response for quality, and coach the agent directly from the results.
There's currently no structured way for admins to validate how their AI agent will perform across a range of questions before changes go live. Batch testing gives admins a repeatable way to build confidence in their customer agent before rollout, and a structured loop to improve it when answers fall short.
First, Create a test set: Add questions manually or generate them from themes (automatically surfaced from recent live conversations in the Inbox or Help Desk). A theme requires 5 or more conversations on a similar topic to appear.
Second, Select your segment: Scope your test set to a specific segment to validate how the agent performs for a particular audience.
Then, Run the test: The Customer Agent generates a response to every question in the set.
Once the test has completed running, Review responses: For each Q&A pair, mark it "Looks good" or "Looks bad." Marking bad surfaces a "Coach agent" button and an optional "Write a note" field, so reviewers can either fix the agent immediately or leave notes. Notes are helpful for coaching in bulk later or to share with a team member who has coaching permissions.
Coach and iterate: Coaching actions open directly from the results view. Admins fix the agent's knowledge or behavior, re-run the test, and confirm the improvement.
Customer Agent is available for HubSpot Content Hub Professional and Enterprise, Data Hub Professional and Enterprise, Marketing Hub Professional and Enterprise, Sales Hub Professional and Enterprise, Service Hub Professional and Enterprise, HubSpot Smart CRM Professional and Enterprise, and HubSpot Credits.