Overview
Overview
Overview
Context & Constraints
Existing chat solutions can’t complete rentals, process payments, or handle operations—blocking revenue and trust.
Prototyping with LLMs: De-risking Feasibility
Challenge
The challenge was designing an AI that could reliably interpret storage-specific intent, execute multi-step tasks, and drive revenue, not just respond.
Testing
I treated the LLM itself as a design prototyping tool. Before creating UI, I ran conversational experiments to test:
Industry-specific intent recognition
Sentiment detection
Multi-step task completion
Upsell timing and phrasing
Outcome
By validating AI behavior first, the product foundation supported complex rental tasks, enabling a conversational system capable of completing bookings instead of simply answering questions. Targeting primary metrics:
Intent resolution rate ↑
Successful outcomes per session ↑
Chapter 2
Designing for Trust, Privacy, & Flow
Challenge
Rental transactions require users to provide sensitive information, including payment details and personal data. Directly exposing this data to an LLM created security, compliance, and trust risks.
Design Decision
I designed custom UI widgets to securely capture sensitive inputs while implementing an event-based logging system that communicated completed actions to the AI without exposing raw data.
Trade-Off
This reduced conversational flexibility and removed some “natural chat” moments. However, it increased security, trust, and scalability.
Outcome
The AI could safely support payments and reservations while maintaining user confidence and meeting production-level privacy expectations.
Business Impacts
By integrating directly with the marketplace and property management system, the AI has real-time access to inventory, pricing, and customer state; Allowing it to act as a true rental agent, not just a support tool.
Conversions
Turns high-intent conversations into completed rentals by enabling discovery, reservation, and checkout directly in chat. Positions AI as a revenue-generating system—not just a support tool (Measured by payment link opens, payments completed, and move-ins via chatbot)
Lead Capture
Captures and nurtures qualified leads during live conversations, ensuring after-hours demand is not lost.
Operational Efficiency
Automates high-frequency rental tasks, reducing operator involvement and support load.
Customer Satisfaction
Maintains a high-quality experience by allowing users to complete tasks quickly while capturing feedback for continuous improvement. (Measured by thumbs-up / thumbs-down and qualitative feedback)
Platform Differentiation
Positions the AI as a category-defining rental agent within the industry, not just a chatbot—leveraging live marketplace and PMS data to complete bookings, process payments, and handle operational tasks that horizontal chatbots can’t match
I designed a fully transactional AI rental agent, treating AI behavior as a core design material rather than a UI layer.
By prototyping directly with LLMs and designing the conversational architecture, privacy model, and checkout flow, I enabled the system to safely capture leads, complete payments, and drive conversions under real-world constraints.
Timeline
3 Weeks
My Role
Lead Designer
Team
AI Engenering Architect
Data PM
Product
web
Skills
End-to-end product design
AI interaction design
LLM prototyping
System logic + UX
Research
Results
+
Solutions
+