Autonomous Email Marketing Engine

The Challenge

Email marketing platforms like Klaviyo and Mailchimp give merchants tools — but they still require significant manual effort. Someone has to define customer segments, write email copy, decide which products to feature, set discount strategies, and optimize send timing. For a store with tens of thousands of customers, doing this well is a full-time job.

The merchant wanted something different: an email system that could make intelligent decisions autonomously, learning from customer behavior and adapting its strategy without constant human intervention.

The Approach

I built a full-stack email marketing engine deeply integrated with Shopify — not a plugin or template, but a custom platform that handles the entire campaign lifecycle from segmentation to send.

ML-Powered Customer Intelligence

The system builds a comprehensive profile of every customer using multiple ML models:

• RFM segmentation — Recency, Frequency, and Monetary scoring places each customer into segments like “Champions,” “At Risk,” or “Lost,” using KMeans clustering
• Churn prediction — a Random Forest model predicts which customers are likely to stop purchasing, allowing preemptive re-engagement
• Behavioral signals — a Shopify web pixel captures storefront browsing events (product views, collection browsing, cart activity), feeding real-time intent signals into customer profiles

AI Decision Engine

For each potential email send, the system uses Claude with Structured Outputs to make a series of decisions:

1. Should this customer receive an email? — based on engagement history, recent purchases, and unsubscribe risk
2. What type of campaign? — re-engagement, upsell, winback, product recommendation, or welcome sequence
3. What tone and content? — professional, friendly, urgent, or conversational, matched to the customer’s segment
4. Should a discount be offered, and how much? — with configurable guardrails to prevent over-discounting

If the AI decision fails (network error, model timeout), the system falls back to rule-based defaults rather than skipping the customer entirely.

Journey Engine

Beyond one-off campaigns, the system orchestrates multi-step email sequences:

• Welcome series — triggered automatically when Shopify creates a new customer
• Scheduled sequences — recurring campaigns at configurable intervals
• Per-customer state tracking — knows exactly where each customer is in each journey
• Reply handling — a SendGrid inbound webhook captures customer replies, matches them to the original campaign, and surfaces them in a Reply Inbox

Post-Purchase Upsell

A Shopify checkout extension presents personalized product offers in the 3-second post-purchase window. The offer selection uses product affinity scoring (based on co-purchase patterns), excludes previously offered items, and signs the offer payload with JWT to prevent tampering. Conversion is tracked end-to-end.

Compliance Built In

Email compliance isn’t optional. The system implements RFC 8058 one-click unsubscribe headers, maintains a consent ledger with full audit trail, handles GDPR deletion workflows via Shopify webhooks, and enforces CAN-SPAM requirements on every send.

The Results

The platform runs autonomously — segmenting customers, generating personalized campaigns, and making send decisions without merchant intervention. The merchant can review and approve campaigns at any autonomy level: manual approval for every send, one-click approval with confidence thresholds, or fully automatic (gated behind a minimum number of approved campaigns to build trust).

The system is built as a white-label platform with multi-tenant architecture, deployed on Railway with the Shopify app embedded directly in the admin sidebar. The intelligence engine (Python/FastAPI) communicates with the Shopify app (Remix) over private networking, with 54 database migrations tracking the evolution from basic segmentation to full journey orchestration.