Recipe No. 43 · Customer support
Support email triage with AI labels
Classify incoming support emails (refund, shipping, product question, other) and label/route them automatically.
- Serves
- E-commerce
- Made with
- Make
- Prep (setup)
- ~45 min est.
- Yield
- 3–5 hrs/week saved, est.
- Level
- Intermediate
- Plan needed
- Make: Core from $12/mo per Axis Consulting; a free tier may cover low volume. Check your plan.
- Est. SMS cost
- Via Twilio: ~$0.0083 per US text segment plus carrier fees, and $1.15/mo per local number (Oct 2026). Other SMS apps price differently, so check your plan.
- TriggerIn Make, create a scenario
- Step 2Add OpenAI 'Create a Completion'
- Step 3Add a Router with one route per category
- Step 4On each route add Gmail 'Add a Label'
- Step 5Add a fallback route labeling…
- Step 6Run once with sample emails, review labels,…
Ingredients
- Gmail or Microsoft 365 mailbox
- Make.com account
- OpenAI API key (pay-as-you-go)
Steps reviewed against current tool UIs by the editor; not yet run end-to-end in a live account. Time saved is an estimate; results vary. Costs from Axis Consulting and Twilio pricing; check current pricing.
SMS rules: only text people who gave you their number for this purpose, identify your business, and include “Reply STOP to opt out”. US business texting needs A2P 10DLC registration.
Method
- In Make, create a scenario. First module: Gmail 'Watch Emails' on your support inbox (criteria: unread).
- Add OpenAI 'Create a Completion' (chat). System prompt: 'Classify this email into exactly one of: refund, shipping, product, other. Reply with the single word only.' Pass subject + body text.
- Add a Router with one route per category; set a filter on each route matching the AI output.
- On each route add Gmail 'Add a Label' (create labels Support/Refund etc. first). Optionally Slack-notify for refunds.
- Add a fallback route labeling 'Support/Review' if output is unexpected.
- Run once with sample emails, review labels, then schedule every 15 minutes. Spot-check weekly — AI can misclassify.