# Make.com: AI-Personalized Abandoned Cart Emails for Shopify

> Beyond '10% off — come back!' — a Make scenario that writes a personal, per-cart abandoned email using the products left behind and the customer's history.

**Platform:** Make  
**Category:** marketing  
**Difficulty:** Advanced  
**Tags:** make, shopify, email, personalization, openai  
**Updated:** 2026-05-22  
**Source:** https://promptfoo.tech/workflows/make-shopify-abandoned-cart-personalization

## Overview
Abandoned cart emails have collapsed to the same 10%-off template. This Make scenario writes a per-cart email that references the specific product left behind, ties it to a real use case, and — for repeat customers — nods to previous purchases. Conversion lifts because it doesn't feel like an autoresponder.

## How it works
1. Trigger: Shopify webhook on checkout abandonment (1 hour after last activity).
2. Enrichment: pull the customer's past orders and any tags.
3. OpenAI prompt writes a short, personal email referencing product name(s), a use case tied to past purchases, and a soft CTA.
4. The draft is sent via Klaviyo (or SendGrid) with UTM tagging.
5. A separate reporting scenario tracks open, click, and conversion per generated email.

## Benefits
- Higher open + reply rates than templated abandoned cart flows.
- Feels human because it references specifics only your store would know.
- Preserves brand voice via a shared voice guide.

## Use cases
- Mid-market DTC brands with AOV > $60.
- Stores with strong repeat-purchase behavior.
- Any category where the product needs context (skincare, cookware, hobbies).

## Step-by-step
### Step 1: Whitelist the products worth personalizing
For low-AOV items the ROI is negative. Filter carts to >$40 and to products where you have clear use-case copy.

### Step 2: Author the voice guide
Same rules as your blog voice, plus: no discount language, no fake urgency, no emojis. Personal, not salesy.

### Step 3: Build the enrichment step
Past orders + customer tags. This is what turns 'come back' into 'you loved X, this is its category upgrade'.

### Step 4: Constrain the model's output length
80-120 words. Longer feels like a newsletter; shorter feels like a bot.

### Step 5: A/B against your current flow
50/50 split for 2 weeks. Measure reply rate, not just click rate — replies indicate the email felt personal.

## Example
```
Email body: 'Hey Amy — you left the ceramic braiser in your cart. Since you bought the Dutch oven in March, I thought you'd want to know these two are our team's most-recommended pair for winter braising. No rush — the cart holds for 14 days.'
```

## FAQs
**Is this GDPR-compliant?**

Yes if you have marketing consent. Never send AI-personalized email to opted-out customers even if the automation could.

**Cost?**

About $0.01-0.03 per email. Trivial next to a $60+ AOV recovery.

**Won't customers get suspicious it's AI?**

Only if the copy feels off. Voice guide + human sampling every week keeps it under the uncanny threshold.

---
Licensed under CC-BY-4.0. Attribution: PromptFoo.tech