Hands-Free CRM

Self-hosted n8n is the recommended setup for professional services firms - you own the data and control where it goes. Provision an Ubuntu server with at least 2GB of RAM from DigitalOcean, Google Cloud, or Azure (budget $12 - 18/month). Run the server setup commands from the resource site, install n8n via Docker or npm, configure NGINX for HTTPS access, and set up a persistent process manager so n8n restarts automatically if the server reboots. If the server setup feels like too much, install OpenCode on the server and tell it to set up n8n. It handles the configuration automatically from a plain-English instruction. Either way, the full setup guide with copy-paste commands is at workforceplaybook.ai. Once n8n is running, connect your first email inbox via the Gmail or Outlook trigger node. Use OAuth authentication - n8n walks you through the authorization flow step by step. Set the polling interval to five minutes. This means n8n checks for new emails every five minutes and queues them for processing. Send a test email from an address you know is in your CRM and confirm it appears in n8n as an execution.

Before any email gets passed to the AI, you need to verify that the sender or recipient is a known contact in your CRM. This filter is what separates CRM activity logs from noise. Without it, you'll fill your CRM with vendor newsletters, internal messages, and recruiting emails. Add an HTTP Request node after the email trigger. Configure it to call your CRM's contact search API endpoint with the sender's email address as the search parameter. In HubSpot, this looks like: GET https://api.hubapi.com/crm/v3/objects/contacts/search with a filter on the email property. In Salesforce, you'll query the Contact object via SOQL. Most other CRMs have equivalent lookup endpoints - check the API documentation under "Search Contacts" or "Find Contact by Email." If the lookup returns a contact, save the contact ID and the full contact record to an n8n variable and continue to the AI node. If the lookup returns no results, route to an exception handler that checks whether the email might be a new prospect worth flagging (see Step 4 on edge cases).

HubSpot contact search request:
Method: POST
URL: https://api.hubapi.com/crm/v3/objects/contacts/search
Headers:
  Authorization: Bearer {{your_api_key}}
  Content-Type: application/json
Body:
{
  "filterGroups": [{
    "filters": [{
      "propertyName": "email",
      "operator": "EQ",
      "value": "{{$json.from.email}}"
    }]
  }],
  "properties": ["firstname", "lastname", "company", "email", "hubspot_owner_id"]
}

Add an OpenAI or Anthropic (Claude) node after the contact lookup. This node reads the full email content and extracts everything your CRM needs in a structured format. The quality of this prompt determines the quality of your CRM data - spend time on it. Configure the node with your API key. Use the "Chat" completion mode. In the User Message field, pass the full email data from the previous node using the expression {{$json}}. In the System Message, use the prompt below. The AI returns a JSON object. Map each field from that JSON to the corresponding CRM field in your write node. If the AI returns follow_up_required as true, trigger a second CRM action that creates a task on the contact record with the follow_up_notes as the task description and a due date of today.

You are a CRM activity logger and sales operations specialist for a professional services firm. Your job is to read client and prospect emails and extract the key information needed for an activity log entry.

Analyze this email carefully, then return ONLY a valid JSON object with these exact fields:

{
  "contact_name": "Full name of the primary external contact",
  "company": "Company or firm name of the external contact",
  "email_address": "Email address of the external contact",
  "communication_date": "ISO 8601 date string",
  "communication_direction": "inbound or outbound",
  "subject": "Email subject line",
  "summary": "2-3 sentence summary of the email content and its significance to the relationship",
  "action_items": ["Array of specific commitments or next steps mentioned"],
  "sentiment": "positive, neutral, or needs_attention",
  "sentiment_reason": "One sentence explaining the sentiment assessment",
  "follow_up_required": true or false,
  "follow_up_notes": "If follow_up_required is true: exactly what needs to happen, by whom, and by when if mentioned",
  "opportunity_signals": ["Any signals that indicate a new or expanded opportunity"],
  "risk_signals": ["Any signals that indicate dissatisfaction, delay, or relationship risk"]
}

Flag follow_up_required as true if: a question was asked that hasn't been answered, a deadline or deliverable was mentioned, a commitment was made by your firm, or the email tone suggests urgency or dissatisfaction. Return only the JSON object with no additional text.

After the AI extraction, write the structured data back to your CRM as an activity record. For HubSpot and Salesforce, n8n has native nodes - search for your CRM name in the node library. For industry-specific platforms (Clio, Karbon, ServiceNow, Cosential, etc.), use the HTTP Request node with your CRM's activity creation endpoint. The field mapping process is the same regardless of which node you use. Map AI output fields to CRM fields: - activity_type → "Email" (hardcoded) - activity_date → email timestamp - subject → email subject - notes → AI-generated summary + action items formatted as a bulleted list - sentiment → custom CRM property if available - contact_id → ID returned from your contact lookup in Step 2 - owner_id → the CRM owner assigned to that contact After writing the main activity, if follow_up_required is true, create a separate CRM task on the contact record with the follow_up_notes as the task description, assigned to the contact owner, due today. This creates a visible queue of items that need human attention. Test the complete workflow end-to-end: send a test email from an address that exists in your CRM, watch it flow through n8n, and verify the activity appears on the correct CRM record with the correct content. Check that field mapping is accurate and that follow-up tasks are created when appropriate before moving to the calendar workflow.

HubSpot activity creation (if using HTTP Request node):
Method: POST  
URL: https://api.hubapi.com/crm/v3/objects/notes
Headers:
  Authorization: Bearer {{your_api_key}}
  Content-Type: application/json
Body:
{
  "properties": {
    "hs_note_body": "{{$json.summary}}\n\nAction Items:\n{{$json.action_items.join('\n')}}",
    "hs_timestamp": "{{$json.communication_date}}",
    "hs_note_subject": "Email: {{$json.subject}}"
  },
  "associations": [{
    "to": {"id": "{{contact_id}}"},
    "types": [{"associationCategory": "HUBSPOT_DEFINED", "associationTypeId": 202}]
  }]
}

Create a second, separate workflow for calendar events. The trigger is a Google Calendar or Outlook Calendar node configured to fire when a new event is created or an existing event is updated. Authenticate it the same way you authenticated the email workflow. After the trigger, add the same contact lookup logic from the email workflow - cross-reference attendee email addresses against your CRM contact database. If any attendee matches a known contact, continue. If none match, stop the workflow. The calendar workflow branches based on event timing. Add an IF node that compares the event end time against the current time. For past events (meeting already occurred): pass the calendar event data to an AI node to generate a meeting activity log. For upcoming events (meeting hasn't happened yet): create a preparatory task on each matched contact record, with text like "Review account history before [Meeting Title] on [Date]." For past events, use this AI prompt structure:

You are a CRM activity logger. A client meeting has just completed.

Based on the calendar event details provided, create a CRM activity log entry.

Return ONLY a valid JSON object with these exact fields:

{
  "contact_names": ["Array of attendee names matched to CRM contacts"],
  "meeting_date": "ISO 8601 date string of when the meeting occurred",
  "meeting_duration_minutes": number,
  "meeting_title": "Title of the calendar event",
  "activity_summary": "2-3 sentences describing the meeting context, who attended, and what it was about based on the event details",
  "follow_up_required": true or false,
  "follow_up_notes": "If follow_up_required is true: what specifically should happen after this meeting",
  "next_meeting_suggested": true or false
}

If the meeting title or description contains words like 'debrief', 'review', 'proposal', 'pitch', or 'follow-up', set follow_up_required to true. Return only the JSON object with no additional text.

The contact matching logic is where most implementations fail. Three edge cases will appear in your first two weeks of production use - build the handling logic before you go live, not after you've generated bad data. **Multiple email addresses per contact**: A client might email you from their work address while their assistant schedules meetings from an assistant's address. Update your contact lookup to search all email fields on the contact record, not just the primary email. If your CRM allows multiple email addresses per contact, confirm your lookup searches all of them. **Unknown senders who might be prospects**: When the contact lookup returns no match, don't stop the workflow silently. Route these emails to your exception queue with: sender name, email address, company domain extracted from the email address, email subject, and a one-sentence AI-generated summary of what the email was about. The exception queue owner reviews these daily and decides whether to create a new CRM record. This prevents you from missing early-stage prospect conversations. **Internal and noise filtering**: Add an IF node that checks whether the sender's domain matches your own firm's domain. Internal emails between your team members should never generate CRM activity logs. Also filter emails containing: "unsubscribe," "@yourdomain.com" as sender, "resume" or "CV" in the subject, "notification" or "alert" in the subject, "invoice" or "receipt" if you don't want those logged. Build these filters before going live. The cost of cleaning up corrupted CRM data afterward is far higher than building the filters correctly upfront.

Create a third workflow - entirely separate from the email and calendar workflows - that runs on a schedule each weekday morning. This workflow delivers an AI-generated account summary to each partner, flagging accounts that have gone quiet, overdue follow-ups, and opportunities requiring attention. The trigger is a Schedule node configured for 7 AM on weekdays (or whatever time your team starts their day). After the trigger, add a CRM query that pulls all active opportunities and accounts, then for each account, retrieves all activity logged in the last 14 days. Pass this data to an AI node using the digest prompt below. After the AI generates the digest, add an email send node that delivers it to each partner whose accounts were included. Format flagged accounts at the top - "Needs Attention," "Overdue Follow-Up," "Review Recommended" - followed by routine accounts below. Test by running the workflow manually before enabling the automated schedule. Review the output yourself: are account groupings correct, are flags firing on the right accounts, is the formatting readable on mobile?

You are a client relationship manager for a professional services firm. Each morning you review account activity and prepare a brief digest for internal employees.

Here is the account activity data for the past 14 days: {{$json}}

For each account, generate a 2-3 sentence summary of recent activity. Apply these flags exactly as specified:

- "NEEDS ATTENTION" - if the account has had no logged activity in the last 14 days
- "OVERDUE FOLLOW-UP" - if the account has an open task with a due date that has already passed  
- "REVIEW RECOMMENDED" - if recent communication sentiment has been 'needs_attention' on two or more logged activities

Format the output as a clean email digest with:
1. A header line: "Account Activity Digest - [Today's Date]"
2. Flagged accounts section (NEEDS ATTENTION first, then OVERDUE FOLLOW-UP, then REVIEW RECOMMENDED)
3. Active accounts section (alphabetical by account name, one paragraph each)

Keep each account summary to 3 sentences maximum. Write in plain, direct language without corporate filler. The goal is to give each partner exactly what they need to know in under 2 minutes of reading.

Create a dedicated Slack or Teams channel - name it something unambiguous like #crm-exceptions or #crm-queue. Set the exception queue owner as the channel manager with their backup listed in the channel description. Route three categories of items to this queue: 1. Potential new contacts identified in emails or calendar events where no CRM match was found (include sender name, email, company domain, and email subject) 2. Accounts flagged "Needs Attention" in the daily digest that have been silent for more than 21 days (include account name, assigned partner, last logged activity date, current opportunity stage, and any open tasks) 3. Workflow errors where AI extraction produced a confidence issue or a CRM write failed Go live in phases. Week 1: email workflow active on one inbox, exception queue monitored daily. Week 2: add remaining inboxes one at a time after verifying each one produces clean data. Week 3: activate calendar workflow and daily digest. Week 4: full production with weekly tuning sessions based on exception queue patterns. Announce the system to your team before going live. Explain exactly what it does: it logs their client communications to the CRM automatically, so they don't have to. Frame it as removing a burden, not as surveillance. This matters for adoption.