Improve email support with AI
Role UX lead, product strategy, rapid experimentation to launch MVP
Timeline8 months
How might we double the email agent’s efficiency while improving the customer experience?
Why solve this problem now?
Email plays an outsized role yet its experience is lagging
77% of support starts via email (200M yearly)
Lower CSAT and longer issue resolution time compared to chat and phone
To capitalize on LLM advancements
Leadership wanted to invest more in elevating Google Support with AI
Zoom in 100x to define the MVP scope
Email support presented a vast and ambiguous landscape, encompassing numerous products and issue types. To navigate this complexity, I led research and created key artifacts to clarify the problem space for the team, fostering shared understanding and empathy.
User journey diagram for end-to-end email support experience
Partnering with the PM, we leveraged research insights to identify our focus: streamline the email draft experience as a first step towards automating emails. We then visualized a two-year strategy, outlining progressive automation and its impact on user experience, which was key for aligning stakeholders on the project's long-term vision.
Data from research were analyzed and mapped in our opportunity-solution tree to help us visualize how we can achieve the desired outcome
The ideal user journey when email support becomes more automated in the future
Generative or Predictive?
We considered 2 options:
Use LLM to generate the entire email draft
Use traditional ML to predict a known email template
Through rapid concept validation with agents, we learned a hybrid approach—blending LLM drafts and templates—was optimal for balancing agent trust and technical feasibility. This iterative development informed our MVP strategy: focusing on streamlined, template-based recommendations to mitigate LLM risks.
Approach 1: use generative AI to create a draft
Approach 2: use ML to predict the email template
Leveraging agent feedback, I further iterated on interactions and visuals for the templated approach to ensure accuracy and trust for maximum acceptance rates. Our MVP goal was to achieve at least 10% acceptance by agents of the recommended email templates.
MVP identified a path to greater automation
Our MVP achieved 20% acceptance rate across security and compliance cases (17M/y), measurably reducing issue resolution time. Our key learnings provided a clear path to scale AI-driven email automation and informed our overarching Google support transformation vision.
