American Express Digital Referral
Experimental chatbot prototype that helped validate a new customer acquisition channel, increasing referral conversions from 0% to 13% in the first quarter.
Project Snapshot
American Express via Philosophie
Financial Services
April 2019 - May 2019
Software Engineer & Product Strategist
Google Assistant, Twilio, Google Voice, NLP, Prototype Development
Prototype Development, Technical Strategy, UX Research Support, AI/ML Implementation
2 UX researchers, 1 software engineer
What It Was
A rapid prototyping initiative to validate American Express's hypothesis for a new chatbot-based credit card referral system. The project aimed to explore whether customers would engage with conversational interfaces for financial referrals while maintaining AMEX's professional standards and security requirements.
This was part of AMEX's broader strategy to discover untapped marketing channels and simplify their customer acquisition process through digital innovation.
Chatbot prototype interface designed for AMEX referral system testing
What I Did
- Built interactive chatbot prototypes using Google Assistant and NLP tools to support UX research studies
- Developed both brand-agnostic and AMEX-branded prototypes to test user preferences without bias
- Implemented Twilio and Google Voice integrations for realistic conversation flows
- Adapted quickly when the Google Assistant responses proved inconsistent, pivoting to a Wizard of Oz methodology
- Supported the UX research team by ensuring the technical implementation didn't interfere with user testing goals
Technical implementation showing Google Assistant and Twilio integration for prototype testing
What Came Out of It
- Helped validate a new customer acquisition channel that launched with measurable success
- Contributed to increasing referral conversions from 0% to 13% in the first quarter after launch
- Provided technical insights that influenced AMEX's investment decisions in chatbot technology
- Delivered actionable prototypes that supported 9 rapid experiments with 50 target users over 16 weeks
- Enabled research findings that shaped AMEX's multi-channel customer interaction strategy
What I Learned
Building the right prototype for research means focusing on what you're actually testing. When the Google Assistant chatbot became inconsistent and hurt the user experience, switching to a Wizard of Oz approach was the right call since we weren't testing the AI itself. Sometimes the simplest solution is the best one, especially when technology complexity can distract from the core research questions.
Want to Prototype AI-Powered Customer Experiences?
Interested in building chatbots and conversational interfaces that drive real business results?