Automation
Example. Turn every customer email into a CRM record
Every customer email contains a CRM record waiting to be written. The sender is the contact. The subject line is the intent. The body is the context. Most teams retype this by hand, or skip the CRM and lose the lead. Here is the automation that does it for you, end to end.
5 min read · published May 21, 2026
The flow
Five steps. The whole thing fits on one screen in a no-code tool.
- Trigger. New email in a Gmail label called "leads."
- Extract. The email body is sent to Claude Haiku 4.5 with a prompt that returns JSON: name, company, intent, urgency, suggested next step.
- Branch. If urgency is "high," send a Slack ping. Otherwise, just write the record.
- Write. Insert a row into Airtable, HubSpot, or Pipedrive with the extracted fields. Reply-to email goes in the contact field.
- Notify. A digest email at 9am summarizing yesterday's new leads.
Tools and prices
Cheapest path: Make.com. Around $15 a month for the operations volume a small team uses. Works with Gmail, Anthropic via the HTTP module, and most CRMs out of the box.
Slightly more powerful: n8n. About $24 a month on n8n Cloud, or self-hosted for $5 a month on Hetzner plus your own monitoring. Better for branching and retries.
Custom: a Next.js route on Vercel reading via the Gmail API. Costs nothing extra if you already have a Vercel project. Best when this flow is one of several that share infrastructure.
The Claude prompt
A short system prompt and a user message with the email body. Output forced to JSON with a strict schema. Haiku 4.5 is fast and accurate enough for extraction. We default to it and only move to Sonnet for emails with attachments or unusual phrasing.
Time saved
Five to ten minutes per email when a human handles it well. Closer to zero when they skip it because they were busy. A team of three that handles 30 customer emails a day saves about ten hours a week. The CRM stays current. The morning digest replaces the "where are we" Slack thread.
When to add a human review step
When the extracted record will fire a downstream action: an outbound email, a Stripe charge, a sales rep assignment. We pause the flow at the extract step, drop the parsed JSON in Slack with Approve and Edit buttons, and only write to the CRM after a click. Turns a fully automatic pipe into a one-second supervised one.