Hey! Context HubSpot’s Customer Agent works well in live chat, but when deployed on email channels the stakes are higher. Email replies are often more formal, more binding, and in some industries (like banking and finance) a wrong or imprecise answer can have legal consequences. Because of this, we cannot allow the Customer Agent to automatically answer every incoming email. Problem Today, when the Customer Agent is enabled for email, it can respond automatically to all messages that meet its confidence threshold. However, this creates risk in cases where: The inquiry is legally sensitive, The customer’s request requires manual investigation, The reply could be interpreted as a binding statement, Or internal policy requires a human approval step. Without guardrails, we cannot safely use the Customer Agent on email. Proposed Feature: Guardrails & Human-in-the-Loop Workflow We would like a way to control which types of emails the Customer Agent is allowed to answer automatically — and which ones should require human review and approval before sending. Key components could include: Content-based Guardrails Allow admins to create rules (keywords, categories, sender-address, intent types) that determine when: Customer Agent may auto-send a response Customer Agent may draft but must wait for approval Customer Agent should not respond at all Approval Queue / Human-in-the-Loop If an email falls into a higher-risk category, Customer Agent should generate a draft and place it in a review queue for a human to approve, edit, or reject before sending. Confidence Threshold Settings per Channel Ability to set a stricter confidence threshold for email than for chat. Impact These controls would allow regulated industries (like banks, insurance companies, healthcare, or legal services) to safely use Customer Agent in email channels without risking compliance breaches or legally binding miscommunication. This would significantly increase adoption and trust in Customer Agent across enterprise and compliance-focused teams. Regards, Marte
...read more