--- name: agent-assistance description: "Provide real-time AI assistance to support agents including suggested responses, knowledge base articles, macros, and conversation summarization. Use when building agent copilot features, configuring response suggestions, implementing knowledge surface integrations, or setting up conversation summarization. Triggers on phrases like 'agent copilot', 'suggested responses', 'agent assistant', 'real-time help', 'knowledge suggestions', 'response templates', 'macros', 'conversation summary', 'agent productivity', 'handle time reduction'." --- # Agent Assistance & Copilot Equip support agents with real-time AI-powered assistance including response suggestions, relevant knowledge articles, macro templates, and post-conversation summarization to reduce handle time and improve quality. ## Workflow ### Phase 1: Response Suggestion Engine 1. **Context analysis**: - Parse incoming customer message in real-time - Extract: intent, product area, sentiment, urgency, keywords - Review conversation history (last 10 messages) for context - Check customer profile (tier, history, known issues) 2. **Response generation**: - Match against approved response library (5,000+ pre-approved templates) - Generate AI-suggested response using fine-tuned LLM (constrained to brand voice) - Rank suggestions by: relevance, approval rate, resolution likelihood - Surface top 3 suggestions with confidence scores 3. **Presentation to agent**: - Inline suggestions within agent desktop (one-click insert) - Highlight key information: resolution steps, policy references, empathy cues - Allow agent to edit, accept, reject, or request alternative - Track acceptance/rejection rate for model improvement ### Phase 2: Knowledge Surface Integration 1. **Real-time article surfacing**: - As agent types or reads customer message, query knowledge base - Rank articles by relevance to current conversation context - Surface top 3 articles as clickable cards in agent sidebar - Include article freshness score and last-reviewed date 2. **Internal wiki & playbook access**: - Surface internal SOPs, troubleshooting playbooks, escalation guides - Contextual hints: "This matches known issue KB-4821 — follow playbook B" - Quick-links to related tickets and resolved cases 3. **Policy & compliance checks**: - Flag responses that violate policy (refund limits, data sharing, SLA commitments) - Require manager approval for policy exceptions - Surface compliance reminders at point of action ### Phase 3: Post-Conversation Summarization 1. **Auto-generated summary**: - Extract: issue description, root cause, resolution, next steps - Generate concise summary (3-5 sentences) - Extract key data points: order numbers, error codes, account IDs - Tag with categories, resolution code, satisfaction indicator 2. **Knowledge gap detection**: - If agent created new solution not in KB, flag for article creation - Suggest KB improvements based on conversation patterns - Identify recurring issues missing documentation 3. **Quality scoring**: - Auto-score conversation against quality rubric (empathy, accuracy, completeness, speed) - Flag conversations for QA review (low score, policy risk, negative sentiment) - Generate coaching recommendations ## Templates ### Agent Copilot Dashboard Layout ``` AGENT COPILIT WORKSPACE — Ticket #48320 ========================================= CUSTOMER MESSAGE (received 2 min ago): "I've been trying to export my quarterly report for 3 hours now. Every time I click export, I get error code EXP-4042. This is blocking my board meeting tomorrow morning. I need this resolved ASAP." ───────────────────────────────────────── 🤖 SUGGESTED RESPONSE (confidence: 87%): "I understand how urgent this is with your board meeting tomorrow. Let me get this resolved for you right away. Error EXP-4042 typically occurs when the report includes data from a deprecated data source. Could you share which report template you're using? In the meantime, I can generate the export from our backend and have it in your account within 10 minutes." [✓ Accept] [✎ Edit] [↻ Alternative] [✗ Reject] ───────────────────────────────────────── 📚 RELEVANT KNOWLEDGE ARTICLES: 1. [KB-4821] EXP-4042: Deprecated Data Source in Report Templates Relevance: 96% | Last updated: 3 days ago Resolution: 89% success rate 2. [KB-3901] Backend Report Generation for Blocked Customers Relevance: 82% | Last updated: 2 weeks ago Resolution: 94% success rate 3. [SOP-112] VIP Customer Urgent Request Protocol Relevance: 78% | Last updated: 1 month ago Note: Enterprise tier customer — expedited handling required ───────────────────────────────────────── ⚠ POLICY ALERTS: • Customer tier: Enterprise (response SLA: 4 hours) • Promise constraint: Do not commit to <5 min resolution without manager approval • Empathy score of customer message: HIGH frustration — acknowledge before troubleshooting ``` ### Post-Conversation Summary Template ``` CONVERSATION SUMMARY — Ticket #48320 ===================================== Generated: 2025-01-15 11:05 UTC | Handle Time: 14.2 minutes ISSUE SUMMARY: Customer unable to export quarterly report — receiving error EXP-4042. Blocking board meeting preparation (urgent business impact). Enterprise tier customer with Platinum contract. ROOT CAUSE: Report template "Q4 Executive Summary" references deprecated data source "Legacy Analytics v2" which was sunset on 2024-12-01. RESOLUTION: Agent generated backend export of requested report. File delivered to customer account within 8 minutes. Customer confirmed receipt and file opens correctly. NEXT STEPS: None — issue resolved. Ticket ready for closure. KEY DATA POINTS: Customer: Acme Corp | Account ID: AC-10847 Report: Q4 Executive Summary (template ID: TMPL-0042) Error: EXP-4042 | Backend export ID: BE-88291 CATEGORIZATION: Primary: Technical Issue → Bug/Error Resolution Code: Backend Workaround Applied Satisfaction Indicator: Positive (customer thanked agent) QUALITY SCORE: 87/100 Empathy: 9/10 | Accuracy: 10/10 | Completeness: 8/10 Speed: 8/10 | Policy Adherence: 9/10 Coaching note: Consider offering template migration proactively KNOWLEDGE GAP FLAG: Suggest creating KB article: "Migrating Reports from Legacy Analytics v2" 12 similar tickets in past 30 days — recurring issue without documented solution. ``` ## Integration Points - **Ticketing systems**: Zendesk, Freshdesk, Intercom, Jira Service Desk - **Knowledge base**: Confluence, Notion, Zendesk Guide, Document360 - **LLM platforms**: OpenAI GPT-4, Anthropic Claude, Google Gemini, custom fine-tuned models - **Agent desktop**: Zendesk Agent Workspace, Front, HubSpot Service Hub - **Analytics**: Internal dashboards (acceptance rates, handle time impact) - **QA systems**: Maestra, Playwright, Quality Manager (quality scoring) - **CRM**: Salesforce, HubSpot (customer context enrichment) - **Communication**: Slack, Teams (agent notifications, escalation) ## Edge Cases | Scenario | Handling | |----------|----------| | Suggestion is inaccurate or harmful | Agent rejection logged; immediate model feedback; human review of rejected suggestions | | Customer message in unsupported language | Language detection → translate → suggest response in source language | | Policy-sensitive topic (refund, legal, security) | Restrict suggestions to approved templates only; no free-form AI generation | | Agent is a trainee (<30 days) | Surface more detailed suggestions with explanations; additional guidance | | Conversation context exceeds token limit | Truncate to last 20 messages; preserve key entities and resolution progress | | AI suggestion acceptance rate drops below 40% | Alert ML team; trigger model re-evaluation; fallback to template library | | Agent overrides suggestion with incorrect response | Policy check catches violations; post-call QA flag; coaching recommendation | | Multiple customers with same issue (bulk) | Detect pattern; suggest bulk-response macro; auto-generate KB article | ## Output ### Agent Productivity Impact Report ``` AGENT ASSISTANCE IMPACT — Monthly Report ========================================= Period: January 2025 | Agents Active: 35 PRODUCTIVITY METRICS: Average handle time: 12.8 min (↓ from 15.2 min — 15.8% improvement) Suggestion acceptance rate: 67.3% (target: >60%) First contact resolution: 74.1% (↑ from 68.5% — 5.6 pts improvement) Tickets per agent/day: 42.3 (↑ from 37.1 — 14.0% improvement) SUGGESTION PERFORMANCE BY CATEGORY: Technical troubleshooting: 72.1% acceptance [███████████████░░░] Billing questions: 78.4% acceptance [█████████████████░] How-to / guidance: 81.2% acceptance [██████████████████] Escalation handling: 45.6% acceptance [███████████░░░░░░░] (needs improvement) Empathy / de-escalation: 63.8% acceptance [█████████████░░░░░] QUALITY IMPACT: QA score (with assistance): 86.4/100 QA score (without assistance): 81.2/100 Policy violation rate: 1.2% (↓ from 2.8%) Customer CSAT: 4.3/5.0 (↑ from 4.1/5.0) KNOWLEDGE GAPS IDENTIFIED: 23 recurring issues without KB articles 8 suggested new articles approved by content team 15 article updates recommended based on conversation patterns ```