--- name: real-time-agent-assistance description: Provide agents with contextual, real-time suggestions and guidance while handling support tickets. Use when building agent assistance tools, implementing real-time ticket suggestions, surfacing KB articles to agents, providing troubleshooting step guides, detecting upsell opportunities during support conversations, or implementing compliance alerts for agents. Triggers on phrases like "agent assistance", "real-time suggestions", "agent sidebar", "just-in-time help", "agent recommendations", "support agent guidance", "contextual assistance". --- # Real-Time Agent Assistance & Suggestions Deliver contextual, AI-powered guidance to support agents during live ticket handling, improving resolution speed and quality. ## Workflow ### 1. Context Analysis and Signal Detection 1. **Ticket intake analysis** (triggered when agent opens a ticket): - Parse ticket subject and body using NLP: extract issue type, product area, urgency - Query customer profile from CRM: plan tier, lifetime value, account health, tenure - Pull ticket history: previous tickets, escalation history, satisfaction scores - Check product analytics: customer's current usage, recent errors, feature adoption - Scan for compliance triggers: billing, data privacy, SLA-sensitive keywords 2. **Real-time conversation monitoring** (as agent types response): - Analyze draft response for tone, completeness, accuracy - Cross-reference draft against KB articles for consistency - Detect potential compliance issues (promising refund without authorization) - Flag missing elements (no greeting, no next steps, no signature) - Suggest empathy language for frustrated customers 3. **Signal-based trigger categories**: - **Knowledge signals**: Issue matches KB article → suggest article - **Process signals**: Complex issue → suggest troubleshooting workflow - **Opportunity signals**: Customer asks about premium features → flag upsell - **Risk signals**: Negative sentiment, churn risk → suggest retention approach - **Compliance signals**: Legal/financial topic → remind of policy requirements ### 2. Suggestion Delivery and Interface 1. **Sidebar suggestion panel** (right-side panel in agent workspace): - **Top panel — Similar Resolved Tickets**: "3 similar tickets resolved in last 7 days — see solutions" - **Second panel — Recommended Articles**: "API Rate Limit Guide (92% relevance) | Rate Limit FAQ (87% relevance)" - **Third panel — Quick Actions**: "Apply $50 credit | Extend trial 14 days | Schedule callback" - **Fourth panel — Alerts and Reminders**: "⚠ Enterprise customer — 15-min SLA" | "💡 Mention feature X — customer on free tier" - **Fifth panel — Troubleshooting Steps**: Expandable checklist for complex issues 2. **Inline suggestions** (within compose area): - Auto-complete common phrases and greetings - Insert KB article links with one click - Auto-populate customer name, account tier, ticket number - Suggest code snippets or command-line fixes - Format buttons for structured responses (steps, tables, code blocks) 3. **Modal dialogs** (for high-priority alerts): - Compliance warnings (blocks send until acknowledged) - SLA breach imminent (visual countdown timer) - Upsell opportunity with one-click proposal - Escalation recommendation with reason and recipient ### 3. Adaptive Learning and Personalization 1. **Agent expertise profiling**: - Track agent knowledge areas (based on resolution history and certifications) - Junior agents (0-6 months): Detailed step-by-step guidance, more KB suggestions, compliance reminders - Mid-level agents (6-18 months): Summary-level guidance, focus on efficiency shortcuts - Senior agents (18+ months): Lightweight alerts only, assume deep knowledge, focus on edge cases - Adjust suggestion volume based on agent feedback (dismissed vs used suggestions) 2. **Suggestion performance tracking**: - Log every suggestion shown: what, when, which agent, accepted/rejected/ignored - Calculate suggestion hit rate per category: KB articles (68%), similar tickets (52%), upsell (23%), compliance (94%) - A/B test suggestion formats and positions - Retrain ML model weekly using acceptance data - Flag consistently ignored suggestions for removal or improvement 3. **Contextual adaptation**: - After-hours shifts: Emphasize self-sufficiency tools, escalation contacts - High-volume periods: Prioritize speed optimizations (macros, quick replies) - New product launches: Surface release notes, known issues, migration guides - Incident mode: Show incident status page link, approved messaging, ticket merge instructions ### 4. Compliance and Quality Guardrails 1. **Pre-send quality checks**: - Response length check: Too brief (<50 words) → suggest expansion - Tone check: Detected passive-aggressive or dismissive language → flag for review - Accuracy check: Response contradicts KB article → alert agent - Completeness check: Customer asked 3 questions, response addresses only 1 → flag - Policy compliance: Refund promised >$100 without approval → block send, require manager approval 2. **Compliance-specific triggers**: - Data privacy: Customer shares PII → auto-remind agent to redact before internal notes - Financial: Discussion of refunds, credits, pricing → surface policy and approval thresholds - SLA: Ticket approaching breach → highlight remaining time, suggest escalation path - Escalation: Customer demands manager → provide escalation template and process - Legal: "Attorney", "lawsuit", "complaint filed" → alert legal team, pause auto-responses ## Templates & Frameworks ### Agent Assistance Dashboard ``` AGENT ASSISTANCE PERFORMANCE — July 2025 ========================================= SUGGESTION DELIVERY (Last 30 Days): Total suggestions shown: 48,620 Agent acceptance rate: 64% (target: >60%) ✓ Suggestions per ticket: 3.2 avg Average time saved per accepted suggestion: 47 seconds BY SUGGESTION CATEGORY: Category | Shown | Accepted | Accept Rate | Avg Time Saved ----------------------|---------|----------|-------------|--------------- KB Article Links | 18,420 | 12,784 | 69% | 32 seconds Similar Tickets | 9,870 | 5,329 | 54% | 58 seconds Troubleshooting Steps | 7,640 | 4,887 | 64% | 91 seconds Quick Reply Macros | 6,230 | 4,984 | 80% | 45 seconds Compliance Alerts | 3,120 | 2,934 | 94% | N/A (required) Upsell Opportunities | 2,340 | 538 | 23% | N/A (revenue) Escalation Recs | 1,000 | 412 | 41% | 67 seconds AGENT ADOPTION BY TENURE: Junior (0-6 months): 78% acceptance, 5.1 suggestions/ticket, +42% faster resolution Mid-level (6-18 months): 65% acceptance, 3.4 suggestions/ticket, +28% faster resolution Senior (18+ months): 51% acceptance, 2.1 suggestions/ticket, +12% faster resolution TOP 10 MOST-USED SUGGESTIONS: 1. "Password Reset Steps" KB article — accepted 2,340 times 2. "API Rate Limit Error" troubleshooting guide — accepted 1,890 times 3. "Billing Discrepancy" quick reply macro — accepted 1,670 times 4. "How to Cancel" KB article — accepted 1,540 times 5. "SSO Configuration" guide — accepted 1,230 times BUSINESS IMPACT: Resolution time reduction: 35% (avg 12 minutes saved per ticket) First contact resolution improvement: +25% Compliance violation reduction: 82% Upsell opportunities surfaced: 2,340 (538 acted on, $47,100 potential revenue) Estimated annual time savings per agent: 312 hours (6.2 weeks) TREND ANALYSIS: Acceptance rate trend: ↑ 8% over last quarter Resolution time trend: ↓ 35% (improving) Junior agent ramp time: ↓ from 8 weeks to 5 weeks Customer satisfaction with assisted responses: 4.4/5.0 (vs 4.2 without) ``` ### Suggestion Relevance Model ``` SUGGESTION RELEVANCE SCORING MODEL =================================== Input Signals (weighted): 1. Ticket text similarity to suggestion source (30%): NLP semantic match 2. Customer context match (20%): Plan tier, industry, geography match 3. Historical effectiveness (20%): Past acceptance rate for similar tickets 4. Agent expertise match (15%): Does agent need this level of guidance? 5. Time sensitivity (10%): Urgent tickets get higher-priority suggestions 6. Product version match (5%): Suggestion matches customer's product version Relevance Score (0-100): ≥90: Show prominently, inline suggestion 70-89: Show in sidebar, top position 50-69: Show in sidebar, standard position 30-49: Show in sidebar, collapsed section <30: Don't show Feedback Loop: - Agent clicks suggestion: +relevance weight for future similar tickets - Agent dismisses suggestion: -relevance weight - Agent ignores suggestion: Neutral (may not have seen it) - Suggestion leads to resolution: +strong relevance weight - Weekly model retrain on accumulated feedback ``` ### Real-Time Assistance Agent Workspace Layout ``` ┌─────────────────────────────────────────────────────────────────────────┐ │ TICKET #12847 | api_integration_error | Enterprise | ⚠ 14:32 SLA │ ├──────────────────────────┬──────────────────────────────────────────────┤ │ CUSTOMER PROFILE │ TICKET CONTENT │ │ ┌────────────────────┐ │ ┌────────────────────────────────────────┐ │ │ │ Acme Corp │ │ │ Subject: API returning 429 errors │ │ │ │ Enterprise Plan │ │ │ │ │ │ │ MRR: $4,200 │ │ │ "Our integration is hitting rate │ │ │ │ Tenure: 18 months │ │ │ limits constantly since the update. │ │ │ │ Health: Green (82) │ │ │ We need to process 500+ requests/min │ │ │ │ Last ticket: 12d │ │ │ and getting blocked after ~200." │ │ │ │ VIP: Yes │ │ │ │ │ │ └────────────────────┘ │ └────────────────────────────────────────┘ │ │ │ │ │ USAGE INSIGHTS │ RESPONSE COMPOSE (with inline help) │ │ ┌────────────────────┐ │ ┌────────────────────────────────────────┐ │ │ │ API calls (30d): │ │ │ Hi {{First Name}}, │ │ │ │ 47,800 (↑ 230%) │ │ │ │ │ │ │ Rate limit hits: │ │ │ [INLINE: Insert KB "Rate Limits" link] │ │ │ │ 1,240 (last 7d) │ │ │ │ │ │ │ Errors: 429 errors │ │ │ [INLINE: Auto-complete response draft] │ │ │ │ Product version: v3│ │ │ │ │ │ └────────────────────┘ │ └────────────────────────────────────────┘ │ ├──────────────────────────┤ │ │ ASSISTANCE SIDEBAR │ │ │ ┌────────────────────┐ │ ⚠ COMPLIANCE: Enterprise SLA — respond │ │ │ 🔥 SIMILAR TICKETS │ │ within 15 min. 2 tickets resolved with │ │ │ • #12789 resolved │ │ rate limit increase for Enterprise plan │ │ │ • #12801 resolved │ │ │ │ │ • #12823 escalated │ │ 💡 OPPORTUNITY: Customer exceeding rate │ │ └────────────────────┘ │ limits → upsell API Premium add-on ($500/mo)│ │ ┌────────────────────┐ │ │ │ │ 📖 KB ARTICLES │ │ [✓ Apply rate limit increase] │ │ │ • Rate Limits Guide│ │ [✓ Send API upgrade proposal] │ │ │ (92% match) │ │ [✓ Escalate to API team] │ │ │ • v3 Migration Notes│ │ │ │ │ (78% match) │ │ [COMPOSE & SEND] [SAVE DRAFT] [ESCALATE] │ │ └────────────────────┘ │ │ │ ┌────────────────────┐ │ │ │ │ ✅ QUICK ACTIONS │ │ │ │ │ • Increase API limit│ │ │ │ │ • Apply $50 credit │ │ │ │ │ • Schedule callback │ │ │ │ └────────────────────┘ │ │ └──────────────────────────┴──────────────────────────────────────────────┘ ``` ## Integration Points - **Help desk platforms** (Zendesk, Freshdesk, Intercom): Ticket data source; agent workspace integration for sidebar rendering - **Knowledge base systems**: Real-time article search and relevance scoring; one-click link insertion - **CRM systems** (Salesforce, HubSpot): Customer profile data, account tier, contract details, lifetime value - **Product analytics** (Mixpanel, Amplitude, Datadog): Real-time usage data, error logs, feature adoption metrics - **AI/NLP engines** (OpenAI, Anthropic, Google NLP): Semantic search, intent detection, tone analysis, response drafting - **Compliance systems**: Policy database, approval workflows, data retention rules - **Revenue systems** (Stripe, Chargebee): Billing history, plan details, upgrade paths, credit application - **Communication platforms** (Slack, Teams): Escalation notifications, collaboration on complex tickets ## Edge Cases - **Agent overwhelmed with suggestions**: New agents or complex tickets may receive too many suggestions, causing cognitive overload: - Implement "suggestion density" control: max 3 active suggestions visible at once - Allow agents to adjust suggestion sensitivity in preferences - Collapsible suggestion categories with clear visual hierarchy - "Quiet mode" toggle during high-pressure situations - Track and optimize: agents who dismiss >60% of suggestions get reduced volume - **Conflicting suggestions**: Two KB articles suggest different solutions for same issue: - Rank by version relevance (match to customer's product version) - Surface both with clear differentiation: "For v2 users" vs "For v3 users" - Flag to content team to resolve conflicting documentation - Agent can report "conflicting suggestions" for model improvement - **Sensitive customer data in suggestions**: AI suggestions inadvertently reference PII from similar tickets: - All AI-generated text passes through PII filter before display - Customer names in examples are always genericized ("a customer reported...") - Agent workspace never shows another customer's data in suggestions - Audit logging on all suggestion content generated - **Suggestion hallucination**: AI suggests incorrect solution or fake KB article: - All suggestions grounded in real data sources (actual KB articles, real tickets) - "No matching content found" rather than hallucinated content - Confidence score displayed: only show suggestions with ≥70% confidence - Agent feedback button: "This suggestion was wrong" → immediate retraining signal - **After-hours and limited-staff scenarios**: Fewer agents available, each handling more tickets: - Priority suggestions shift to fastest resolution paths - Auto-suggest macro responses for common after-hours issues - Escalation contacts clearly displayed with on-call info - "Quick resolve" mode: surface only top-3 most likely solutions - Pre-approved credit/refund authority increased for after-hours agents ## Output ### Agent Assistance Weekly Summary ``` WEEKLY ASSISTANCE IMPACT REPORT — Week of July 14, 2025 ========================================================= TEAM OVERVIEW: Total tickets handled: 1,520 Tickets with assistance used: 1,247 (82%) Average suggestions per assisted ticket: 2.8 Acceptance rate: 67% (↑ 3% from prior week) PERFORMANCE IMPROVEMENT: Metric | Without Assistance | With Assistance | Improvement ------------------------------|--------------------|-----------------|------------ Avg resolution time | 24 min | 15 min | -35% ✓ First contact resolution | 58% | 74% | +16 pts ✓ CSAT score | 4.1/5.0 | 4.4/5.0 | +0.3 ✓ Compliance violations | 12/week | 2/week | -83% ✓ Escalation rate | 18% | 13% | -5 pts ✓ TOP PERFORMING AGENTS (by suggestion utilization): Agent | Tickets | Suggestions Used | Accept Rate | Resolution Time ----------- |---------|------------------|-------------|---------------- Maria S. | 67 | 189 | 81% | 11 min James K. | 58 | 156 | 78% | 12 min Priya R. | 72 | 168 | 73% | 13 min Tom B. | 45 | 92 | 68% | 14 min Sarah L. | 53 | 110 | 62% | 16 min RECOMMENDATIONS: 1. Maria's approach (high utilization, fast resolution) — share in team meeting 2. Tom B. could benefit from "Getting Started with Agent Assistance" training 3. KB article "Webhook Setup Guide" has only 34% acceptance — review for relevance 4. Compliance alerts are 96% accepted — consider automating compliance actions 5. Junior agent average acceptance (76%) vs senior (54%) — model working as intended ```