Email Marketing AI Assistant: Breaking a 500K-Subscriber Open-Rate Plateau
Lifted open rates from 15% to 47% and CTR to 12.3% for a 500K-subscriber e-commerce list using GPT-4 personalization and send-time optimization.
THWORKS built an AI-powered email marketing assistant for an e-commerce brand with 500,000 subscribers stuck at a 15% open rate plateau despite A/B testing subject lines for over a year. Using GPT-4, FastAPI, Celery, and AWS SES, the platform personalizes every subject line, send time, and content block to individual subscriber behavior patterns — producing truly one-to-one email experiences at scale. Result: open rates jumped to 47%, click-through rates hit 12.3%, and revenue per email increased 210% within a single quarter.
The Challenge: A 500K-Subscriber List Stuck at a 15% Plateau
A mid-sized e-commerce brand had grown their email list to 500,000 subscribers through years of paid acquisition and content marketing. They used a major ESP (Klaviyo) with best-practice segmentation and had run over 200 subject line A/B tests in the past year — but their open rate was flat at 15% and click-through rate was at 1.8%. Revenue per email had been declining quarter-over-quarter for 18 months, even as list size grew. The marketing team suspected they had hit an engagement ceiling, but they didn't know how to break through it.
In 2026, email remains one of the highest-ROI marketing channels — but only if you can get subscribers to open and click. Generic 'batch-and-blast' campaigns, even with segmentation, are fundamentally limited because they send the same content to people with very different preferences, browsing history, and purchase intent. True one-to-one personalization at 500K subscribers was technically impossible with off-the-shelf ESP tools.
Our Solution: Per-Subscriber AI Personalization Engine
We built a personalization layer that sits between the product catalog and the ESP. For every send, the system pulls subscriber-level behavioral data (browsing history, past purchases, open patterns, device type, timezone), then uses GPT-4 to generate a unique subject line, preview text, and product recommendation block tailored to that individual subscriber. FastAPI handles the personalization requests, Celery distributes the workload, and the output is pushed to Klaviyo via API for final delivery.
The key insight: rather than replace Klaviyo (which handles deliverability, list management, and compliance well), we augmented it. Marketers still write campaign briefs in Klaviyo — but our system generates 500,000 unique variations at send time instead of 1 generic version. Every subscriber gets subject lines that reference things they actually care about, sent at the time they're most likely to be checking email.
Key Technical Decisions
Per-Subscriber Generation at Send Time: Instead of creating 10-20 segment variations, we generate a unique subject line and preview text for each of the 500K subscribers — GPT-4 is fast and cheap enough at scale to make this economically viable.
Send-Time Optimization: A per-subscriber model predicts the optimal send time based on historical open behavior — accounting for timezone, day of week, and typical email-checking patterns. Result: a 40% open rate lift from timing alone before personalization.
Deliverability Guardrails: Built an automated content safety layer that filters out AI-generated copy containing spam trigger words, ALL-CAPS, excessive emojis, or phrases that historically correlate with poor inbox placement — protecting the sender reputation.
Results: Open Rate Tripled, Revenue Per Email Up 210%
Before
15% open rate plateau despite 200+ A/B tests. 1.8% click-through rate. Declining revenue per email for 18 months despite list growth. Marketing team out of ideas.
After
47% open rate with per-subscriber personalization. 12.3% click-through rate (7x improvement). 210% increase in revenue per email. Marketing team repurposed to strategic work instead of A/B test fatigue.
Technology Stack
"We'd literally run out of ideas for improving email performance. We'd tested every subject line framework, every emoji combination, every day of the week. THWORKS's approach was so different — personalizing for every single subscriber instead of testing averages — that we almost didn't believe the early results. Three months in, our email channel is delivering twice the revenue on the same list size."
Frequently Asked Questions
Common questions about this project and our approach.
At send time, the system pulls each subscriber's behavioral profile (browsing history, past purchases, engagement patterns) and uses GPT-4 to generate a unique subject line, preview text, and content block tailored to them — not a segment. This means 500,000 subscribers get 500,000 unique email experiences, compared to traditional segmentation which might produce only 5-10 variants.
Related Projects
Transform Your Email Marketing
Let's discuss how we can engineer the same level of technical excellence for your project.
Transform Your Email Marketing