Generative and Agentic AI Recommendations
While working with Gusto’s 60-person design team, I helped design a templated, AI-powered recommendation system that replaced static homepage banners with contextual, in-flow prompts. The solution used generative and agentic AI to deliver personalized messages and actions, optimized through A/B testing.
My Role
Strategy
UI/UX Design
Visual Design
Timeline
3 Months
Every test variant showed positive impact. In the most successful case, we saw a 250% lift in revenue from a single product recommendation sequence.
The Opportunity
Gusto’s growth team wanted to evolve their recommendation strategy beyond homepage banners and into in-flow messaging that felt helpful, timely, and relevant. The goal: surface upgrade opportunities and new product add-ons directly at the user’s pain point. Generative AI would tailor the message based on user behavior and account configuration, while agentic AI would proactively take helpful action when appropriate.
Endemic recommendations solution View in full resolution
The Process
Our design process was fast-paced and experiment-driven. We relied heavily on A/B testing to validate concepts, supported by qualitative insights from UX Research. Design decisions were made collaboratively through brainstorming sessions, async critiques, and design system office hours. We worked closely with product managers and engineering to ensure feasibility and alignment.
A/B Testing and Success KPIs
We defined clear success criteria at the start of each test, tracking:
- Click-through rate
- Impression-to-interaction rate
- Conversion rate
- Dismiss rate
- Customer Lifetime Value (CLTV)
Each iteration aimed for incremental revenue lift and improved product engagement.
Conversation Design
In alignment with Gusto leadership’s broader AI mandate, we designed the first touchpoint as a lightweight, conversational UI. This allowed us to frame the recommendation as helpful guidance rather than a marketing prompt, increasing trust and engagement.
AI-Powered Recommendations Engine
The system leveraged generative AI to create personalized messages for each user scenario, while agentic AI enabled direct action—such as auto-adding a free trial of a relevant feature without friction. All templates were designed to be modular, brand-aligned, and easy to extend across other product areas.
The Solution
The solution was a modular, multi-touch system that delivered personalized, in-flow product recommendations using generative AI. Agentic AI activated trial features when appropriate, and all components were designed for consistency across products. Message timing and frequency were optimized to balance engagement and user experience.
Endemic recommendations solution See the prototype
The Results
Every test variant showed positive impact. In the most successful case, we saw a 250% lift in revenue from a single product recommendation sequence. The framework is now being extended to other product lines within Gusto’s platform.