Support Copilot in Zendesk/ServiceNow: 30‑Day Plan

yaml
policy_version: v1.7
owners:
  product: "Support Ops"
  engineering: "AI Platform"
  qa: "Support QA Leads"
scope:
  platforms: ["zendesk", "servicenow"]
  regions:
    - name: "EU"
      data_residency: "eu-west"
    - name: "US"
      data_residency: "us-east"
queues:
  - name: "Device Connectivity"
    slo:
      first_response_minutes: 20
      resolution_hours: 24
    macro_intents: ["device_reset", "firmware_known_issue", "outage_update"]
    auto_apply_macro_if:
      confidence_gte: 0.82
      not_contains_tags: ["vip", "refund", "legal_review"]
    draft_reply:
      tone: "reassuring, concise"
      include_steps: true
      cite_sources: true
      max_tokens: 180
    approvals:
      required_if_tags: ["refund", "legal_review", "security"]
      approver_role: "Supervisor"
  - name: "Billing"
    slo:
      first_response_minutes: 30
      resolution_hours: 48
    macro_intents: ["invoice_copy", "payment_failed", "tax_id_update"]
    auto_apply_macro_if:
      confidence_gte: 0.86
      not_contains_tags: ["vip", "escalation"]
    draft_reply:
      tone: "empathetic, clear next step"
      include_steps: false
      cite_sources: true
      max_tokens: 160
    approvals:
      required_if_tags: ["refund", "proration", "chargeback"]
      approver_role: "Billing Lead"
retrieval:
  sources:
    - type: "kb"
      path: "knowledge_base/articles"
    - type: "tickets"
      path: "resolved_tickets/last_180_days"
    - type: "runbooks"
      path: "sn_runbooks/network"
  vector_index:
    engine: "private-vector-db"
    chunk_tokens: 800
    freshness_boost_days: 30
    dedupe_by: ["article_id", "ticket_id"]
rbac:
  roles:
    Agent: { actions: ["view_suggestion", "edit_draft", "send"], pii_access: false }
    Supervisor: { actions: ["approve", "override_macro", "label_feedback"], pii_access: true }
    Admin: { actions: ["configure", "export_logs", "feature_flag"], pii_access: true }
privacy_safety:
  pii_redaction: ["card_number", "ssn", "iban"]
  profanity_filter: true
  safe_actions:
    - "send_troubleshooting_steps"
    - "link_kb_article"
    - "apply_non_refund_macro"
logging:
  prompt_logging: true
  event_retention_days: 365
  export_targets: ["s3-like-blob-store", "audit_api"]
experiments:
  ab_tests:
    - name: "reply_drafting_v2"
      population: 0.5
      metric: "aht_minutes"
telemetry:
  metrics:
    - name: "suggestion_accept_rate"
      threshold: ">=0.55"
    - name: "edit_distance_chars"
      threshold: "<=120"
    - name: "sla_breaches"
      threshold: "<=1%"
feature_flags:
  billing_refund_autoapply: false
  device_outage_macro_autoapply: true
reviews:
  daily_qa_sample: 50
  weekly_policy_review: "Fridays 10:00 EU/US-friendly"

{
  "title": "Support Copilot in Zendesk/ServiceNow: 30‑Day Plan",
  "published_date": "2025-11-24",
  "author": {
    "name": "Alex Rivera",
    "role": "Director of AI Experiences",
    "entity": "DeepSpeed AI"
  },
  "core_concept": "AI Copilots and Workflow Assistants",
  "key_takeaways": [
    "Stand up a Zendesk/ServiceNow copilot in 30 days with human-in-loop controls.",
    "Draft replies, surface troubleshooting, and apply macros automatically—with audit trails.",
    "Expect measurable impact: lower handle time and faster first responses without CSAT risk."
  ],
  "faq": [
    {
      "question": "Will the copilot overwrite our macros or KB?",
      "answer": "No. It reads macros and articles as authoritative. Suggestions are shown to agents for approval; admins can flag any macro as non-editable and the copilot will not propose changes."
    },
    {
      "question": "How do we prevent risky replies (refunds, legal)?",
      "answer": "Approval gates route refund, proration, or legal terms to supervisors. Confidence thresholds and tags ensure sensitive actions are never auto-applied."
    },
    {
      "question": "What if agents ignore the suggestions?",
      "answer": "We track acceptance rate and edit distance. If value isn’t there, we adjust retrieval sources, tone, and macro mappings. Daily QA reviews close the loop within the pilot."
    }
  ],
  "business_impact_evidence": {
    "organization_profile": "B2B SaaS, 120-seat global support team on Zendesk with ServiceNow for problem management; EU and US regions.",
    "before_state": "Agents copied steps from old tickets and applied macros inconsistently; first-response time missed SLOs during releases; supervisors spent hours triaging duplicates.",
    "after_state": "Copilot drafted replies with macro alignment and linked troubleshooting steps; supervisors focused on QA coaching; refund scenarios routed to approval automatically.",
    "metrics": [
      "First-response time down 41% in pilot queues.",
      "Average handle time reduced 28% within 3 weeks.",
      "CSAT up 4.7 points; macro error rate down 63%.",
      "Supervisor triage time reduced by 6 hours/week."
    ],
    "governance": "Legal and Security approved due to prompt logging, RBAC aligned to Zendesk roles, regional data residency controls, PII redaction, and a human-in-the-loop approval workflow; the platform never trains shared models on client data."
  },
  "summary": "Support leaders: ship a governed copilot inside Zendesk/ServiceNow in 30 days that drafts replies, respects macros, and lifts CSAT while lowering handle time."
}