Cc Email Drafter

Email Drafter

Drafts customer-facing emails for routine communications across the lifecycle. Personalises with account-specific data, calibrates tone by segment and relationship context, and outputs review-ready drafts for CSM approval before sending. Part of the Communication & Content Production pillar.

This is a content generation skill. It produces drafts. It never sends emails autonomously -- every customer-facing communication passes through the CSM for review, edit, and approval. The human decides what goes out.

When to Run

• Triggered: By lifecycle orchestration skills when a communication is due:
  • lo-onboarding-orchestrator: welcome emails, onboarding check-ins, milestone congratulations
  • lo-check-in-scheduler: cadence-driven check-in emails
  • lo-renewal-manager: renewal conversation openers, renewal confirmations
  • lo-milestone-tracker: milestone achievement notifications
  • lo-qbr-orchestrator: QBR scheduling and follow-up emails
• Triggered: By cc-call-summariser after a call (post-call follow-up email)
• On-demand: When a CSM requests a draft for a specific purpose

Core Execution Logic

Step 1: Determine Email Context

From the triggering event or CSM request, assemble:

Step 2: Select and Populate Template

Each email category has a template with required and optional elements:

For each email:

1. Select the template matching the category
2. Populate personalisation fields: account name, contact name, specific metrics, open items, milestone details
3. Insert context-specific content: the observation, question, or data point that makes this email relevant to this customer right now
4. Apply tone calibration (see below)

Step 3: Calibrate Tone

Situation-based overrides (these override the segment default):

Step 4: Apply Quality Gates

Before presenting the draft:

Step 5: Output Draft for Review

{
  "draft_id": "email-2026-0342",
  "account_id": "string",
  "recipient": {
    "name": "Tom Chen",
    "email": "tom@acme.com",
    "role": "VP Engineering"
  },
  "category": "post_call_follow_up",
  "trigger_source": "cc-call-summariser",
  "subject": "Follow-up: API docs and April onboarding",
  "body": "Hi Tom,\n\nThanks for the time today. Great to hear the API latency fix is holding.\n\nQuick recap:\n- API documentation for Lisa's team: I will send this by Friday\n- Latency monitoring: we will confirm resolution after the 7-day window (by March 17)\n- April onboarding: you will share the start dates and headcount plan by March 21\n\nI will also start pulling together a training plan for the new engineers once we have the headcount confirmed.\n\nTalk next week.\n\nBest,\nJane",
  "personalisation_fields_used": ["contact_name", "api_docs_commitment", "latency_ticket", "onboarding_headcount"],
  "tone": "warm_professional",
  "word_count": 98,
  "quality_gates": {
    "length": "pass",
    "single_cta": "pass -- no explicit CTA needed, follow-up is a recap",
    "subject_specificity": "pass",
    "personalisation": "pass",
    "tone_match": "pass -- mid-market, warm professional",
    "recency_check": "pass -- post-call, same day is expected",
    "sensitivity_check": "pass -- no risk signals"
  },
  "suggested_send": "2026-03-10T16:00:00Z",
  "requires_csm_review": true
}

Handoff to Human

Confidence and Limitations

• High confidence for routine lifecycle emails (check-ins, follow-ups, scheduling, milestone notifications) -- these are pattern-based with data personalisation. The templates are well-defined and the data insertion is deterministic
• High confidence for quality gate application -- the checks are objective (word count, subject length, recency)
• Medium confidence for tone calibration -- segment-based rules are a sensible default, but individual relationships have their own register. A CSM who has built informal rapport with an enterprise contact should not send a formal email just because the segment says so
• Medium confidence for personalisation relevance -- the skill includes the most recent and significant data points, but the CSM may know that a specific metric is sensitive or that the customer will not care about a milestone the system thinks is important
• Low confidence for sensitive situations. Churn risk emails, escalation responses, bad news, and relationship repair communications should be heavily edited or drafted from scratch by the CSM. The skill can provide a starting structure but the human layer is critical for anything that carries relationship risk
• Cannot assess whether the email will be well-received. The draft may be perfectly constructed and still land poorly because the customer is having a bad day, is frustrated about something unlogged, or simply does not like email
• Cannot determine optimal send time with precision. Timezone-based scheduling is a heuristic. The CSM knows whether this contact reads email in the morning or evening, and whether they respond better to mid-week or Monday outreach

Dependencies

Required:

• CRM API (contact data, activity history for personalisation and recency checks)
• bi-account-brief (account context for personalisation)

Strongly recommended:

• lo-check-in-scheduler (cadence-driven email triggers)
• lo-onboarding-orchestrator (onboarding email triggers)
• lo-renewal-manager (renewal communication triggers)
• lo-milestone-tracker (milestone notification triggers)
• cc-call-summariser (post-call follow-up triggers with summary content)
• ra-engagement-tracker (recency check and engagement pattern data)
• bi-risk-detector (sensitivity check -- flag at-risk accounts)
• Calendar integration (send time optimisation based on timezone)
• Email integration (thread context for follow-ups, delivery tracking)

Downstream consumers:

• CSMs (primary consumer -- review and approve before sending)
• CRM activity record (via cc-crm-updater -- email logged as a touchpoint)
• ra-engagement-tracker (sent email as an outbound touchpoint)
• lo-sla-monitor (commitments made in the email become tracked items)

References

• references/email-templates.md -- Template library for all email categories with personalisation field definitions and example outputs