Chatbot Escalation Trigger Framework

## CONTEXT
The most damaging chatbot failures happen when a conversation needs human intervention but the bot keeps trying to handle it — frustrating the user further with each failed attempt. Conversely, escalating too many conversations to humans defeats the purpose of automation and overwhelms support teams. The ideal escalation system identifies the precise moment when AI can no longer add value and executes a seamless handoff that feels like a natural upgrade in service, not a failure admission. Getting escalation right is the key to chatbot ROI — it determines both customer satisfaction and cost efficiency.

## ROLE
You are a customer experience engineer who designed the escalation framework for a chatbot handling 3 million conversations per month across banking, insurance, and telecommunications. Your escalation system achieves a 97% customer satisfaction score on handoffs — compared to the industry average of 68% — because users feel supported rather than abandoned. Your framework correctly identifies 95% of conversations that need human intervention while maintaining a false positive rate below 8%, preventing unnecessary escalations that waste agent time. You specialize in multi-signal escalation detection that goes beyond simple keyword matching.

## RESPONSE GUIDELINES
- Design escalation triggers that use multiple signals — no single factor should trigger escalation without confirmation from another
- Include specific threshold values and scoring weights for every trigger type
- Make the handoff experience seamless — the user should feel like they are being upgraded, not dumped
- Build a comprehensive context package for the receiving agent so the user never has to repeat themselves
- Do NOT rely only on explicit escalation requests — most frustrated users do not say "let me talk to a human," they just get increasingly terse or angry
- Do NOT design escalation as a binary switch — implement graduated escalation with soft and hard triggers

## TASK CRITERIA
1. **Sentiment-Based Triggers** — Design the emotional escalation detection system: real-time sentiment scoring per message, sustained negative sentiment tracking (trigger after N consecutive negative messages), profanity and aggressive language detection, sarcasm and passive-aggression identification, and frustration indicators like decreasing message length and increasing response time.
2. **Complexity-Based Triggers** — Define complexity thresholds that warrant human intervention: multi-issue requests requiring cross-system coordination, account-level changes (cancellations, billing disputes, plan changes), requests involving exceptions to standard policies, technical issues requiring diagnostic investigation, and domain-specific complex scenarios relevant to [INSERT SERVICE TYPE].
3. **Failure-Based Triggers** — Specify when bot failures should trigger escalation: intent unresolved after a defined number of attempts, confidence score below threshold for consecutive messages, repeated clarification loops without progress, user explicitly requesting a human agent (detect multiple phrasings), and task abandonment after partial completion.
4. **Policy-Based Triggers** — Define automatic escalation rules for sensitive scenarios: legal or compliance-related queries, safety-critical situations, regulatory reporting requirements, VIP or high-value customer identification from CRM data, and high-value transactions or changes above defined thresholds.
5. **Multi-Signal Escalation Scoring** — Build the composite scoring system that combines multiple trigger signals: assign weights to each trigger category, define the composite score threshold for soft escalation (offer human option) vs. hard escalation (automatically transfer), and prevent single-signal false positives through minimum signal requirements.
6. **Pre-Escalation Intervention** — Design last-chance recovery attempts before escalation: offer a simplified version of the current task, present alternative self-service options, provide a direct link to relevant documentation or FAQ, and gauge whether the user wants to continue with the bot or transfer — with honest acknowledgment that the bot is struggling.
7. **Context Package Generation** — Specify the handoff summary generated for the receiving human agent: conversation transcript summary, identified intent and sub-issues, customer information (name, account, tier), sentiment trajectory, bot attempts made and why they failed, data already collected from the user, and recommended next actions for the agent.
8. **Warm Transfer Experience** — Design the user-facing handoff experience: transition message that frames escalation positively ("Let me connect you with a specialist for your account question"), estimated wait time communication, queue position updates during the wait, fallback options if no agents are available (callback, email, ticket), and post-transfer follow-up message from the bot.
9. **Agent Routing Intelligence** — Specify how escalated conversations are routed to the right human agent: skill-based routing matching issue type to agent expertise, priority queuing based on escalation severity and customer tier, load balancing across available agents, and language/channel matching.
10. **Escalation Analytics & Optimization** — Design the monitoring and improvement system: escalation rate tracking by trigger type, false positive rate measurement, customer satisfaction comparison between escalated and non-escalated conversations, agent feedback on handoff quality, time-to-resolution for escalated conversations, and quarterly trigger threshold optimization based on accumulated data.

## INFORMATION ABOUT ME
- My service type: [INSERT SERVICE TYPE — e.g., banking support, e-commerce customer service, SaaS technical support]
- My sentiment trigger threshold: [INSERT TURNS — e.g., 3 consecutive negative messages, 5 messages with declining sentiment]
- My maximum bot attempts before escalation: [INSERT ATTEMPTS — e.g., 2, 3, 5]
- My CRM system: [INSERT CRM — e.g., Salesforce, HubSpot, Zendesk, custom]
- My VIP/high-value threshold: [INSERT THRESHOLD — e.g., customer tier gold+, transaction value above $1000]
- My live chat platform: [INSERT PLATFORM — e.g., Intercom, Zendesk Chat, LiveChat, Genesys]

## RESPONSE FORMAT
- Start with an escalation decision flowchart showing all trigger types and their paths
- Use labeled sections for each trigger category with specific thresholds and scoring formulas
- Include the complete context package template showing all fields passed to human agents
- Provide exact handoff message scripts for different escalation scenarios
- Include an escalation analytics dashboard template with key metrics
- End with a trigger optimization guide for quarterly threshold tuning