How to Build an AI Email Triage System That Sorts Your Inbox
A founder or executive receiving 100+ emails per day spends 2 to 3 hours just processing their inbox. AI can read, classify, prioritise, and draft responses to the majority of those emails — so you spend your inbox time on the 10% that genuinely needs your attention.
How the System Classifies Your Email
| Category | Description | AI Action | Your Action |
|---|---|---|---|
| Urgent and requires you | Client escalation, legal, board, time-sensitive decision | Flagged in priority inbox with summary | Read and respond today |
| Needs response — standard | Client queries, partner comms, proposals | AI drafts response for your review | Review, edit if needed, send |
| Needs response — delegate | Internal questions, team requests, admin | AI drafts response + suggests who should handle | Approve delegate or send directly |
| FYI — no action needed | Newsletters, reports, CC chains, notifications | AI summarises in daily digest | Read digest, follow up if needed |
| Spam or irrelevant | Unsolicited outreach, marketing, automated system emails | AI moves to trash or archive | Nothing |
With Make.com and Gmail
Set up the Gmail trigger in Make.com
Create a new Make.com scenario. Add a Gmail trigger: Watch Emails. Set it to watch your primary inbox, running every 15 minutes. Connect your Gmail account with OAuth. Configure a filter to only process emails in your Primary inbox tab (not Promotions or Social — these are already pre-sorted by Gmail). Test by sending yourself a test email and confirming Make.com detects it within the polling interval.
Build the AI classification step
Add an HTTP module calling Claude. Prompt: You are an executive email assistant. Classify this email and return a JSON object with: category (one of: urgent_action, needs_response, delegate, fyi, spam), priority (high/medium/low), summary (one sentence describing the email content and required action), suggested_response (if category is needs_response or delegate, draft a professional 3-5 sentence response), and delegate_to (if category is delegate, suggest the role or person who should handle this). Email details: From: [sender name and email], Subject: [subject], Body: [first 500 characters of body]. My context: I am [your role] at [company name]. My direct reports are [names and roles]. My EA/assistant is [name]. Parse the JSON response — Make.com’s JSON Parse module extracts each field.
Apply the classification actions
Add a Router module with routes for each category. Urgent: add a Gmail label ‘URGENT’, send yourself a Slack or SMS notification with the AI summary, and create a GoHighLevel or CRM task if sender is a known client. Needs Response: add label ‘RESPOND’, create a draft reply in Gmail using the AI-generated suggested_response (Gmail module: Create Draft Reply), and add the email to a daily review list. Delegate: add label ‘DELEGATE’, forward the email to the relevant team member with the AI-generated context and suggested response, add to your delegation log. FYI: add label ‘FYI’, skip to inbox processing — no urgent action needed. Spam: archive or trash automatically.
Build the daily digest
A separate Make.com scenario runs every day at 5pm. It retrieves all emails labelled ‘FYI’ from the day, passes them to Claude: Summarise these emails in a daily digest. For each email, provide: sender, subject, and a one-sentence summary of the content and any implicit action needed. Group by topic if multiple emails cover the same subject. Total digest should be readable in 3 minutes. Email the digest to yourself. You receive one consolidated email covering everything that did not need immediate attention, rather than being interrupted by each one throughout the day.
Train and refine the classification over 2 weeks
The classification will not be perfect from day one. For the first 2 weeks, review the classifications daily: note any emails miscategorised (an urgent client email classified as FYI, or a newsletter classified as Needs Response). Build a correction prompt: also update the system prompt with specific rules based on patterns you observe: always classify emails from [important client domain] as urgent_action, never classify automated notification emails from [system@] as needs_response. After 2 weeks of refinement, the classification accuracy should exceed 90%.
Is it safe to give Make.com access to my Gmail?
Make.com is a legitimate automation platform used by millions of businesses. The Gmail OAuth connection grants Make.com the specific permissions you approve — typically read and write access to your email. The connection is revocable at any time from your Google account settings. Review Make.com’s privacy policy and data processing agreement if you handle sensitive client data. For high-sensitivity situations, consider running Make.com with a dedicated business email rather than your personal Gmail account.
Can I build this for Outlook instead of Gmail?
Yes — Make.com has a Microsoft 365 / Outlook module with equivalent functionality to the Gmail module. The trigger (Watch Emails), the label/category system (Outlook categories rather than Gmail labels), the draft reply creation, and the folder management are all supported. The Claude classification step is identical regardless of email platform — the classification logic does not depend on the email provider.
Want Your Inbox Triage System Built?
SA Solutions builds Make.com email intelligence systems — AI classification, automated responses, delegation workflows, and daily digests for founders and executives who need their inbox under control.
