On-Demand Email Assistant

The simplest implementation uses a monitored draft folder in Gmail or Outlook. Create a folder named "Draft Assist" or "Draft Request." When a partner moves an email thread to this folder, n8n's email polling detects it and triggers the draft workflow. For a more seamless experience, build or install an email add-in. Gmail add-ins can be built in Google Apps Script and published for your organization. Outlook add-ins use the Office Add-ins framework. Both display a "Draft Assist" button in the email compose window that calls an n8n webhook URL with the thread content. For either approach, the trigger sends the same payload to n8n: the full email thread, the sender's email address, and any additional context the partner provides (optional instructions like "keep this formal" or "focus on the pricing objection").

After the trigger, n8n looks up the CRM record for the email recipient. Pull: contact record (communication preferences, relationship status, account tier), company record (account summary, active opportunities, recent activity), last 5 logged activities with summaries, any open service issues or complaints, and any sensitivity flags on the account. Compile this into a structured context block that the AI can reference. The more specific and current the CRM data, the more calibrated the draft will be. A draft generated from a CRM with "preferred communication: formal, no jargon; billing concern logged 3 weeks ago; relationship rated 'at risk'" will be dramatically different from one generated from a sparse record.

Pass the thread content, CRM context, and any partner instructions to the AI. Use the draft prompt below. The AI returns a complete reply draft - not a template with blanks, but a finished draft that the partner can edit or send as-is. After the draft is generated, run a second AI call for the flag check. This is a separate prompt that focuses specifically on catching problems: tone mismatches, factual inconsistencies, and red-flag patterns. Return the flag report as a separate, structured output alongside the draft. Deliver both to the partner: the draft in an editable format (inline reply in the email client if using an add-in, or forwarded to a draft-delivery email address), and the flag report as a brief summary below the draft.

You are a senior client communications specialist for a professional services firm. Your job is to write a first-draft reply to a client email, calibrated to the specific client relationship.

Email thread to reply to:
{{email_thread}}

Client CRM context:
- Contact: {{contact_name}}, {{contact_title}} at {{company_name}}
- Communication preferences: {{communication_preferences}}
- Relationship status: {{relationship_status}}
- Recent activity summary: {{recent_activity_summary}}
- Open issues: {{open_issues}}
- Sensitivity flags: {{sensitivity_flags}}

Partner instructions (if any): {{partner_instructions}}

Firm tone guidelines: {{firm_tone_guidelines}}

Write a complete reply that:
1. Opens appropriately for this client's communication style (formal or conversational based on preferences)
2. Addresses every question or point raised in the last email
3. Is accurate - only references information that is in the thread or CRM context provided
4. Uses the client's name naturally (not repeatedly)
5. Closes with a clear next step or expectation
6. Matches the relationship status - if it's "at risk," be especially careful and warm; if it's "strong," you can be more direct

Return ONLY the email body text (no subject line). Do not add any explanation or meta-commentary.

Run this as a separate AI call after draft generation. The flag check looks for specific problems, not general quality. It checks three categories: 1. Tone flags: does the draft match the client's documented communication preferences? If the client is documented as "formal" and the draft uses casual language, flag it. 2. Factual flags: does the draft reference any specific numbers, dates, invoice numbers, or client claims that can be checked against the CRM? Cross-reference any specific facts in the draft against the CRM data provided. Flag any discrepancy. 3. Red-flag patterns: scan the draft for your predefined red-flag list - internal pricing language, placeholder text that wasn't filled in, another client's name appearing in the draft (this happens when pulling from past email templates), overly promotional language that sounds like marketing copy, and any language that could create legal exposure. Return flag results as: flag type, severity (low/medium/high), the specific problematic text, and a suggested correction or note to the reviewer.

You are a quality control specialist reviewing a draft email before it's sent to a client. Check for specific problems only.

Draft to review:
{{draft_text}}

Client CRM data for fact-checking:
{{crm_context}}

Red flag patterns to check for:
{{red_flag_list}}

Client's documented communication preferences:
{{communication_preferences}}

Perform three checks:

1. TONE CHECK: Does the draft's tone, formality level, and word choice match the documented communication preferences? 

2. FACTUAL CHECK: Do any specific claims, numbers, dates, names, or invoice references in the draft conflict with the CRM data provided?

3. RED FLAG CHECK: Does the draft contain any of the defined red flag patterns?

Return ONLY a JSON object:
{
  "flags": [
    {
      "type": "tone" | "factual" | "red_flag",
      "severity": "low" | "medium" | "high",
      "description": "What the issue is",
      "problematic_text": "The exact text in the draft that's flagged",
      "suggestion": "How to fix it or what to double-check"
    }
  ],
  "overall_assessment": "clean" | "minor_flags" | "review_required",
  "flag_count": number
}

If there are no flags, return an empty flags array with overall_assessment: "clean".