Back home

EMAIL AI / CASE STUDY

Designing Trust in Generative Email: A Phased Framework

Designing Trust in Generative Email: A Phased Framework

Email AI personalization interface and email preview
Email AI personalization interface and email preview

01 / Overview

This project explores a phased approach to AI-powered personalization that builds trust over time. Marketers can see what the system generates, understand how it works, and make the final decision before sending.

MY ROLE

Product Design Lead

Led strategy for product vision and interaction design, partnering with product and engineering to execute rollout.

CORE TEAM

Product Management, Content Tools, Generative AI, and ML Engineering

COLLABORATORS

UX Research, Product Marketing, Sales, Client Strategy

02 / The challenge

Email marketing has existed for decades, and the way brands build emails is about the same as it was 10 years ago. The build process is tedious, time consuming, and restricts marketers from unlocking the personalized messaging they desire for their customers.

An email is highly visual and speaks directly for a brand's values and voice. For marketers to hand any of that over to AI, they had to trust what shipped would be accurate, on-brand, and safe to send.

SKEPTICAL

“We’ve run several tests trying to use AI for a human’s tasks. All of those efforts failed.”

ENTERPRISE CLIENT

Somewhere in the middle

~50% of clients landed here

ENTHUSIASTIC

“I think AI is here to redesign email templates to be more effective, especially for triggered messages.”

ENTERPRISE CLIENT

SKEPTICAL

“We’ve run several tests trying to use AI for a human’s tasks. All of those efforts failed.”

ENTERPRISE CLIENT

Somewhere in the middle

~50% of clients landed here

ENTHUSIASTIC

“I think AI is here to redesign email templates to be more effective, especially for triggered messages.”

ENTERPRISE CLIENT

03 / Research & key decisions

I partnered with product and engineering leadership to align on what success actually needed to look like for the business. Our goal became increasing email program revenue through efficiency gains. We believed there were two ways efficiency could drive performance:

  • Send More. Faster creation means more campaigns.

  • Send Better. Personalization becomes easier at scale.

But our strategy meant nothing if we didn't meet our users where they were. I put together early concepts and immediately started user research. I wanted to talk to the AI-forward, innovative folks, and almost more importantly, the clients that were super skeptical of AI.

04 / The phased framework

We had a wide target. To drive clarity among complicated development work, we phased our rollout of Generative Email into digestible releases. This allowed us to nurture a relationship with our beta testers, allowing them to vet the tools and give us critical feedback.

PHASE 01

Email Brand Styles

Clients were uncomfortable with AI making sensitive brand decisions, like choosing fonts and colors, so I needed to build trust and maintain brand control. I created a manual style form that keeps clients and account managers at the wheel while improving the accuracy of our generative emails.

Email Brandkit with desktop and mobile styles. Required rows can be dragged to reorder.

PHASE 02

Seed Templates

We analyzed historically high-performing rows, filtered by technical feasibility, and used that data to build a library of seed templates. That same row-level logic became the foundation for row taxonomies: guardrails that let the model generate more dynamic templates while staying accurate and on-brand.

Breaking down the anatomy of an email, with defined purposes for each row.

PHASE 03

Testing with Private Alpha

To validate our first round of generative templates, we recruited AI-forward clients who were willing to test them with real subscribers. We helped brands set up their email styles and run A/B tests comparing generated templates to their existing templates.

In one journey sent to 50k subscribers, the AI-generated templates came within 3% of the brand’s control click-through rate. Our generative emails were keeping up with the control group’s historic, manually built emails.

“So far, the generated emails are running about 1–2% below the original. I was pleasantly surprised by how well they are doing against our existing emails.” - Enterprise marketer

That result mattered less as a performance win and more as a trust milestone. It gave us permission to keep going and gave clients their first real evidence that this was safe to build on.

PHASE 04

Unlocking 1:1 Personalization

We introduced meaningful personalization in journey emails by focusing on elements clients found valuable and easy to adopt. Personalized copy is generated through enabled tactics like offers, first names, and triggering products, with toggles that let clients experiment and adopt them at their own pace.

Toggles to enable AI-Personalization features individually, spanning copy generation and tactics to include.

Marketers can preview emails for each subscriber, seeing product copy that is dynamic and personal to them.

The trade-offs behind the design

Every project has scope changes and technical hurdles, but I believe navigating these is where a product designer shines. These are the moments we advocate for the user, while helping our product partners arrive at a solution.

01 / Navigating new territory

Challenge: In early 2024, email content generation was still new, with few established patterns to follow.

How I approached it: My engineer and I defined our own taxonomies, then prompted, tested, and refined the model. We also conducted multiple rounds of research to maintain a good pulse on customer perception.

Challenge: In early 2024, email content generation was still new, with few established patterns to follow.

How I approached it: My engineer and I defined our own taxonomies, then prompted, tested, and refined the model. We also conducted multiple rounds of research to maintain a good pulse on customer perception.

02 / Defining the right vision

Challenge: Research showed that clients wanted content generation for different use cases. Some prioritized frequent, image-heavy campaigns, while others hesitated to trust AI with product imagery. Refining our vision was challenging.

How I approached it: I advocated for trust-building first. We focused on incremental releases that demonstrated value, allowing us to improve quality before expanding the feature’s scope.

05 / Outcomes & reflection

AI-generated subject lines and content drove a 17.1% lift in open rate and a 5.0% lift in click-through rate, both statistically significant across tens of thousands of sends. Skeptical clients were now running it in production with their own subscribers. Generative Email drove performance for marketers, and our phased feature rollout nurtured trust in AI to operate on behalf of brands.

This template generation work laid the tech foundation for the next phase of our email platform. We're now building toward agentic message creation and program operations, allowing users to automate tedious tasks and unlock valuable time to accomplish their program goals.

© 2026 Katherine Homer

Back home

EMAIL AI / CASE STUDY

Designing Trust in Generative Email: A Phased Framework

Designing Trust in Generative Email: A Phased Framework

Email AI personalization interface and email preview
Email AI personalization interface and email preview

01 / Overview

This project explores a phased approach to AI-powered personalization that builds trust over time. Marketers can see what the system generates, understand how it works, and make the final decision before sending.

MY ROLE

Product Design Lead

Led strategy for product vision and interaction design, partnering with product and engineering to execute rollout.

CORE TEAM

Product Management, Content Tools, Generative AI, and ML Engineering

COLLABORATORS

UX Research, Product Marketing, Sales, Client Strategy

02 / The challenge

Email marketing has existed for decades, and the way brands build emails is about the same as it was 10 years ago. The build process is tedious, time consuming, and restricts marketers from unlocking the personalized messaging they desire for their customers.

An email is highly visual and speaks directly for a brand's values and voice. For marketers to hand any of that over to AI, they had to trust what shipped would be accurate, on-brand, and safe to send.

SKEPTICAL

“We’ve run several tests trying to use AI for a human’s tasks. All of those efforts failed.”

ENTERPRISE CLIENT

Somewhere in the middle

~50% of clients landed here

ENTHUSIASTIC

“I think AI is here to redesign email templates to be more effective, especially for triggered messages.”

ENTERPRISE CLIENT

SKEPTICAL

“We’ve run several tests trying to use AI for a human’s tasks. All of those efforts failed.”

ENTERPRISE CLIENT

Somewhere in the middle

~50% of clients landed here

ENTHUSIASTIC

“I think AI is here to redesign email templates to be more effective, especially for triggered messages.”

ENTERPRISE CLIENT

03 / Research & key decisions

I partnered with product and engineering leadership to align on what success actually needed to look like for the business. Our goal became increasing email program revenue through efficiency gains. We believed there were two ways efficiency could drive performance:

  • Send More. Faster creation means more campaigns.

  • Send Better. Personalization becomes easier at scale.

But our strategy meant nothing if we didn't meet our users where they were. I put together early concepts and immediately started user research. I wanted to talk to the AI-forward, innovative folks, and almost more importantly, the clients that were super skeptical of AI.

04 / The phased framework

We had a wide target. To drive clarity among complicated development work, we phased our rollout of Generative Email into digestible releases. This allowed us to nurture a relationship with our beta testers, allowing them to vet the tools and give us critical feedback.

PHASE 01

Email Brand Styles

Clients were uncomfortable with AI making sensitive brand decisions, like choosing fonts and colors, so I needed to build trust and maintain brand control. I created a manual style form that keeps clients and account managers at the wheel while improving the accuracy of our generative emails.

Email Brandkit with desktop and mobile styles. Required rows can be dragged to reorder.

PHASE 02

Seed Templates

We analyzed historically high-performing rows, filtered by technical feasibility, and used that data to build a library of seed templates. That same row-level logic became the foundation for row taxonomies: guardrails that let the model generate more dynamic templates while staying accurate and on-brand.

Breaking down the anatomy of an email, with defined purposes for each row.

PHASE 03

Testing with Private Alpha

To validate our first round of generative templates, we recruited AI-forward clients who were willing to test them with real subscribers. We helped brands set up their email styles and run A/B tests comparing generated templates to their existing templates.

In one journey sent to 50k subscribers, the AI-generated templates came within 3% of the brand’s control click-through rate. Our generative emails were keeping up with the control group’s historic, manually built emails.

“So far, the generated emails are running about 1–2% below the original. I was pleasantly surprised by how well they are doing against our existing emails.” - Enterprise marketer

That result mattered less as a performance win and more as a trust milestone. It gave us permission to keep going and gave clients their first real evidence that this was safe to build on.

PHASE 04

Unlocking 1:1 Personalization

We introduced meaningful personalization in journey emails by focusing on elements clients found valuable and easy to adopt. Personalized copy is generated through enabled tactics like offers, first names, and triggering products, with toggles that let clients experiment and adopt them at their own pace.

Toggles to enable AI-Personalization features individually, spanning copy generation and tactics to include.

Marketers can preview emails for each subscriber, seeing product copy that is dynamic and personal to them.

The trade-offs behind the design

Every project has scope changes and technical hurdles, but I believe navigating these is where a product designer shines. These are the moments we advocate for the user, while helping our product partners arrive at a solution.

01 / Navigating new territory

Challenge: In early 2024, email content generation was still new, with few established patterns to follow.

How I approached it: My engineer and I defined our own taxonomies, then prompted, tested, and refined the model. We also conducted multiple rounds of research to maintain a good pulse on customer perception.

Challenge: In early 2024, email content generation was still new, with few established patterns to follow.

How I approached it: My engineer and I defined our own taxonomies, then prompted, tested, and refined the model. We also conducted multiple rounds of research to maintain a good pulse on customer perception.

02 / Defining the right vision

Challenge: Research showed that clients wanted content generation for different use cases. Some prioritized frequent, image-heavy campaigns, while others hesitated to trust AI with product imagery. Refining our vision was challenging.

How I approached it: I advocated for trust-building first. We focused on incremental releases that demonstrated value, allowing us to improve quality before expanding the feature’s scope.

05 / Outcomes & reflection

AI-generated subject lines and content drove a 17.1% lift in open rate and a 5.0% lift in click-through rate, both statistically significant across tens of thousands of sends. Skeptical clients were now running it in production with their own subscribers. Generative Email drove performance for marketers, and our phased feature rollout nurtured trust in AI to operate on behalf of brands.

This template generation work laid the tech foundation for the next phase of our email platform. We're now building toward agentic message creation and program operations, allowing users to automate tedious tasks and unlock valuable time to accomplish their program goals.

© 2026 Katherine Homer