AI Agent That Resolves Spotify Listener Support Issues End to End

Quick Overview

A product design question about building an AI agent that runs customer support for Spotify and can act on users' accounts. It tests how you scope contact reasons, decide which billing and account actions the agent may take, design identity checks, guardrails and human handoff, and measure resolution quality.

AI Agent That Resolves Spotify Listener Support Issues End to End

Company: Sierra

Role: Product Manager

Category: Product Design & Strategy

Difficulty: medium

Interview Round: Other

Design an agentic AI system that runs customer support for Spotify. The agent should resolve users' issues and act on their behalf, not just answer questions. ```hint Sort contacts by what the agent must do Group the reasons people contact Spotify by whether the agent only has to explain something, has to read account data, or has to change something such as a charge, a plan or account access. Each group needs different tools and carries different risk. ``` ### Clarifying Questions - Which users are in scope: Free and Premium listeners only, or also artists, podcasters and advertisers, who have their own support needs? - Which channels: in-app and web chat, email, or voice? - Which actions may the agent take without a human, such as refunds, plan changes, cancellations or account recovery? - Is the main goal lower support cost, faster resolution, or keeping subscribers who contact support intending to cancel? ### What a Strong Answer Covers - A breakdown of Spotify contact reasons by volume and risk, and a reasoned choice of which ones the agent handles first. - The agent's design: the account context and knowledge it uses, the actions it can call, identity verification, and when it hands off to a person. - Guardrails for money and account changes, wrong answers and manipulation, with business rules enforced outside the model. - How the agent is evaluated before and after launch, and a staged rollout. - Success metrics centered on correct resolution, with guardrails on wrong actions, refunds and churn. ### Follow-up Questions - A user says they were charged twice and wants a refund. Walk through what the agent does and where it stops. - Resolution rate looks strong, but cancellations through the agent went up. How would you investigate? - How would you test a new version of the agent before it talks to real users?

Overview: A product design question about building an AI agent that runs customer support for Spotify and can act on users' accounts. It tests how you scope contact reasons, decide which billing and account actions the agent may take, design identity checks, guardrails and human handoff, and measure resolution quality.

Sierra logo
Sierra
May 21, 2026
mediumProduct ManagerOtherProduct Design & Strategy
0
0

Design an agentic AI system that runs customer support for Spotify. The agent should resolve users' issues and act on their behalf, not just answer questions.

Clarifying Questions Guidance

  • Which users are in scope: Free and Premium listeners only, or also artists, podcasters and advertisers, who have their own support needs?
  • Which channels: in-app and web chat, email, or voice?
  • Which actions may the agent take without a human, such as refunds, plan changes, cancellations or account recovery?
  • Is the main goal lower support cost, faster resolution, or keeping subscribers who contact support intending to cancel?

What a Strong Answer Covers Guidance

  • A breakdown of Spotify contact reasons by volume and risk, and a reasoned choice of which ones the agent handles first.
  • The agent's design: the account context and knowledge it uses, the actions it can call, identity verification, and when it hands off to a person.
  • Guardrails for money and account changes, wrong answers and manipulation, with business rules enforced outside the model.
  • How the agent is evaluated before and after launch, and a staged rollout.
  • Success metrics centered on correct resolution, with guardrails on wrong actions, refunds and churn.

Follow-up Questions Guidance

  • A user says they were charged twice and wants a refund. Walk through what the agent does and where it stops.
  • Resolution rate looks strong, but cancellations through the agent went up. How would you investigate?
  • How would you test a new version of the agent before it talks to real users?