Technical architecture for a Spotify support chatbot running on Sierra's agent platform

Suppose Spotify decided to run the chatbot behind its customer support on Sierra's AI agent platform. What would the technical system for that integration look like? ```hint Follow one real request Pick a common support case, such as a user who was charged twice, and trace it from the chat window through every system it touches. That shows which parts the vendor platform should own and which must stay inside Spotify. ``` ### Clarifying Questions - Which channels are in scope: chat in the app and on the support website only, or voice and email as well? - Should the bot only answer questions from help content, or also take account actions such as changing plans, issuing refunds and troubleshooting playback? - When a conversation needs a person, which human-support system should receive the handoff, and with what context? - Which markets and languages must work at launch? ### What a Strong Answer Covers - The split of responsibilities between Sierra's agent platform and Spotify's own systems (identity, accounts, billing, help content, human support). - The path of one conversation: authentication, retrieval from help content, calls into Spotify services, the reply, and handoff to a person. - Controls on account and payment actions: narrow access, policy limits, confirmation and audit logging. - Data handling: what user data leaves Spotify, retention, regional rules, and family-plan permissions. - Evaluation and operations: testing before launch, resolution metrics, monitoring, load peaks, and fallback when either side fails. ### Follow-up Questions - A billing-policy change takes effect tomorrow. How does the bot learn it, and how do you confirm it behaves correctly before then? - Some users pay for Premium through a mobile carrier or an app store. How does the system handle their refund requests? - The agent platform goes down during a large Spotify service incident. What do users see, and what keeps support working?