Email Marketing at Scale
How Frosted manages multiple clients with AI-powered email strategies and stays sane.
Frosted manages email for 17 clients. That’s 17 different brands, 17 different audiences, 17 different goals. Without AI, we’d be drowning.
The Problem with Email at Scale
Most agencies do email the same way they did in 2010:
- Bulk send to everyone
- Generic segments (by purchase history, mostly)
- Hope the copy resonates
That worked when you had one client and 100k subscribers. It doesn’t work when you have 17 clients and 2M subscribers across all of them.
How We Use AI
Our entire operation runs on three principles:
1. Personalization at Scale
We don’t do “Hi {{first_name}}, buy this thing.” We do actual personalization:
- Each email is written for that specific subscriber
- Subject lines are generated based on their behavior
- Send times are optimized per person
This required building:
- An embeddings database of every subscriber’s behavior
- A prompt system that can iterate in <500ms
- A safety layer that catches “weird” generations before they ship
2. Testing as a First-Class Feature
We test everything:
- Subject lines (3 variants per send)
- Send times (optimal for each segment)
- Copy length (short vs. long)
- CTA placement
That’s a lot of email. But we’ve learned that the difference between a 2.1% and 2.3% open rate is $50k/month for some clients.
3. Feedback Loops
Every email generates data. We:
- Track opens, clicks, conversions
- Feed that back into the personalization model
- Update segment definitions weekly
- Retire emails that consistently underperform
The Team
We’re a 17-person team. That’s tiny for managing 2M subscribers. We’re able to do it because:
- AI writes most of the copy
- We’ve automated away the busywork
- We focus on strategy and testing
What Keeps Us Up at Night
Compliance
17 clients × regulations (GDPR, CAN-SPAM, etc.) = complexity. We have one person whose entire job is making sure we don’t accidentally violate something.
Quality Variance
AI sometimes generates… weird copy. We’ve built a multi-layer approval system:
- Automated safety checks (does this look normal?)
- Human review of a sample
- Client approval
- Final safety scan before send
Client Alignment
Not all clients want AI-generated copy. Some think it’s impersonal. We’ve learned to be transparent: “This is AI-generated, tested on your audience, and likely to outperform what a human wrote.”
The Results
Our clients see:
- 15-30% lift in open rates
- 10-25% lift in click rates
- 5-15% lift in conversion (varies by client)
That’s not magic. That’s consistent application of personalization, testing, and feedback loops.
What’s Next
Email isn’t dead. It’s evolving. The agencies winning are those who’ve figured out how to scale human-level personalization to millions of subscribers.
Frosted is just getting started.