Prioritizing the First Build When an Enterprise Adopts Sierra's AI Customer Agent

Sierra builds AI agents that hold customer-service conversations on a business's behalf. A large enterprise customer wants to integrate Sierra into its operations. As the PM, what do you build first, and why does it come before everything else? ```hint Find what blocks go-live List what a large enterprise must have in place before it lets an AI agent talk to its customers and act in its systems, then ask which missing piece delays the first real, fully resolved conversation the most. ``` ### Clarifying Questions - Which customer journeys does the enterprise want the agent to handle first (for example order status, returns, account changes or billing questions), and over chat, voice or both? - Which systems must the agent read from or take actions in, such as the CRM, order management or the contact-center platform, and do they expose usable APIs today? - Is this a signed contract with a launch date, or a pilot that has to prove value before a wider rollout? - Should the first build be reusable for other enterprise customers, or is one-off work for this account acceptable? ### What a Strong Answer Covers - Discovery of what this enterprise needs before go-live: target journeys, systems of record, security and compliance review, and handoff to its human agents. - A single recommended first build, justified by time to the first resolved conversation and by reuse across other enterprise accounts. - What is explicitly deferred and why, with a phased path from a limited launch to broader coverage. - Launch metrics (resolution, customer satisfaction, escalation quality) with guardrails for wrong actions and policy breaches. - Risks such as the agent taking incorrect actions in live systems, customer data exposure, and custom work that never generalizes. ### Follow-up Questions - The customer insists on a connector to an in-house legacy system that no other customer uses. Do you build it, and how do you protect the roadmap? - Two months after launch, most conversations are resolved, but satisfaction is low for the ones escalated to humans. What do you look at and change? - How do you decide when a request from this one account should become a platform capability for every customer?