The core principle: your job in the first 10 minutes is to understand their reality, not demonstrate yours. Interviewers explicitly grade whether you rush to architecture. Don't.
They say something vague. You play it back in their words — not your interpretation, theirs.
Why: It signals you heard them. It surfaces whether your interpretation matches theirs. It buys you a second to think.
Example:
Customer: "We want to add AI to our customer support." You: "So if I'm hearing you right — the support team is dealing with something that feels like it could be improved, and AI has come up as a possible answer. Before I jump to anything, I want to make sure I understand what's actually painful today. Is that fair?"
Notice: you didn't say "great, we can build a RAG chatbot." You reflected and created space.
This is the longest step. You're building a picture across 4 dimensions:
Who is affected?
"Is this about the agents handling tickets, the customers waiting for responses, or leadership looking at costs — or all three?"
Current state:
"Walk me through what happens today when a customer submits a ticket. What's the typical resolution time? What percentage get resolved on first contact?"
What's been tried:
"Have you explored any solutions already — macros, knowledge base, a previous chatbot? What happened?"
Hard constraints:
"Any compliance requirements I should know about — HIPAA, SOC 2, data residency? Any on-prem requirements?"
Success metric in 6 months: