Auto-resolution policy

Confidence threshold to auto-resolve: 75%

This case study models a support chatbot handling 3000 tickets a week. Each ticket gets a model confidence score for auto-resolution. Above the threshold, the bot resolves it directly; below, it routes to a human agent. Auto-resolving a ticket the bot cannot actually handle well produces a poor customer experience.

Weekly outcome

Tickets deflected (auto-resolved)—

Poorly-handled auto-resolutions—

Agent hours saved—

Tickets by confidence (blue = auto-resolved, red ring = poorly handled)