PLUM / Mobile / chatbots

Making help accessible, helping

users find answers faster

Enabled instant resolution for users and scalable support operations by reserving human assistance for judgment-based cases.

/ About Plum

Plum is an insurtech company focused on corporate insurance, expanding into healthcare and the B2C space.

/ Context

At Plum, chat is the primary way customers contact support — about 90% of users prefer messaging. We used third-party chat software to connect users with agents. As the product grew, conversations increased rapidly.

Our support demand grew quickly and the team could no longer handle the volume efficiently.

/ Problem

Users reached out to support even for simple questions. Answers existed, but they were not available at the right moment. Every query became a conversation with an agent.


Because of this:

  • Users waited for responses

  • Agents handled repetitive queries

  • Complex issues were delayed

/ Why this problem was critical to solve

Support demand was growing faster than the team. If every question required a human response, response time and service quality would drop. To scale the product, most issues needed to be resolved without contacting support.

/ pain points

  • Long wait time for simple questions

  • Repeated queries to support agents

  • Dependency on chat to understand claims and coverage

  • High operational effort

  • Agents had less time for complex problems

/ Design goals

Help users resolve common issues inside the product and involve support only when necessary.

/ Final Design

Designed an in-app support experience that helped employees resolve queries instantly using contextual guidance and AI assistance.


Bot understands user intent

  • Provides answers in context

  • Guides users through actions

  • Escalates complex cases to agents


We also introduced PolicyGPT, a GPT-based assistant trained on our policy and coverage data. Users could ask questions in natural language and understand their benefits without contacting support.


Support changed from answering every question to handling only cases that required human judgment.