Recognize when to step back
A good setup treats complaint-shaped language — refund, broken, angry, unacceptable, this is ridiculous — as a signal to offer a human path immediately rather than attempting to fully resolve it in chat with a scripted response. The goal in these moments isn't to solve the problem in the widget; it's to move the person toward someone who can.
This requires the chatbot to recognize emotional or urgent language as a distinct category from routine questions, and respond differently — with acknowledgment and a path forward, rather than the same style of factual answer it would give to 'what are your hours.'
Never argue with a frustrated visitor
The bot should acknowledge the issue plainly and move toward a resolution path, not debate the visitor's account of what happened or defend the company's position the way a poorly trained script might. This is a case where a short, honest response beats a long, defensive one — 'I'm sorry that happened, let me connect you with someone who can help directly' does more good than a paragraph explaining the policy.
Visitors in this state aren't looking for a policy explanation from a bot — they're looking for confirmation that a real person will actually address the problem.
Make escalation genuinely easy
A visible email, phone number, or handoff option — not buried three menus deep, not requiring the visitor to type 'human' three times before it works — is the difference between a contained situation and an angrier one. The friction of finding a way to a person compounds an already frustrated visitor's frustration.
If you offer a live-chat handoff to a human, make sure it's genuinely available during the hours you claim — an escalation path that leads to an unmonitored inbox during off-hours isn't much better than no path at all, though it's still better than a dead end.
Follow up on what the transcript shows
After a complaint gets escalated, it's worth checking the transcript afterward — not to second-guess the visitor, but to see whether the bot's initial response made things better or worse, and to catch any pattern of complaints about the same underlying issue that might need addressing at the source, not just in chat.
