+25% engagement among younger members · +35% discovery of emerging verticals · 4.5/5 helpfulness rating
Members trusted Credit Karma with their credit data but kept their financial questions to themselves. Financial literacy varied widely, and personal finance carried enough shame that most people kept it private even from a partner or close family member. Static help content treated every member the same, in a format anyone browsing could see.
The question: how do we help members get answers while keeping their finances private?
Led content for Penny's acquisition integration into Credit Karma's core product, partnering with the founder, engineers, and a data scientist on the Penny team to define what the conversation should do and how it should feel.
Each conversation was built individually and refined over time, with updates shipped almost daily as member data showed what worked.
Credit Score Education was one conversation among dozens. Penny grew into nine conversation areas — credit health, managing financial accounts, getting credit cards, personal loans, autos, home, tax, identity, and personal banking, plus a few smaller ones — each with at least four conversations, and each conversation logged at least 100 interactions.
Many conversations adjusted to what a member already knew and used their own data as examples, going deeper only when asked. A memory system let Penny recall a member's stated goals and build on them across sessions rather than starting over each time.
Penny launched fast and ran in real time, pulling live member data directly into the conversation. It skewed toward younger members, which is where the engagement lift shows up. Dropoff points in the transcripts surfaced UX problems that only became visible in real conversation data.
Conversation Design Before Conversation Design Was Cool