Talk track

1. This scenario proves a different kind of reuse than every other one in the kit: instead of one schema re-skinned across industries, it's one agent config re-skinned across interview types. Switch the theme dropdown from candidate screening to exit interview live, and the topic list — and therefore the questions the agent actually asks — changes with it, with zero code touched.
2. Point at the voice picker: that's the live ElevenLabs voice library, not a fixed shortlist. Search for something specific on stage — an accent, a name — and start the call in that voice. It's the same override mechanism a production integration would use to let an end customer choose how their own AI interviewer sounds.
3. This is deliberately the shallowest scenario in the kit — live transcript and captured details, no AI-generated executive summary. That's a scope choice, not a limitation: it's the fastest thing an SE could stand up in an afternoon, which is exactly the 'what could you build in 2 hours' story the deck calls out separately.
4. The interviewee details form is the enablement point as much as the agent is — a seller can walk a prospect through exactly what context gets handed to the agent before it ever opens its mouth, which is the same 'agent opens with context, not "how can I help?"' story as every scenario in the kit, just visible in a form instead of buried in a signed-URL payload.
5. This scenario now runs at two deliberately different depths, and that's the point, not an inconsistency. The interview mode stays the shallowest scenario in the kit on purpose — live transcript, no AI-generated insight — because for that use case, an unedited transcript is the trustworthy artefact. Switch the theme dropdown to one of the seller-practice personas and it's the opposite: closed-loop, AI-graded, coaching generated automatically. Different jobs get different amounts of AI, deliberately — that's a maturity signal for an audience that already knows more AI isn't automatically better AI.
6. The seller-practice personas are the clearest 'AI-first enablement' moment in the whole kit: a seller doesn't just watch a demo of ElevenLabs talking to a customer, they get talked back at by a skeptical one, live, as many times as they want to rehearse, and get scored on how well they actually positioned Agents Platform tool-calling, deterministic gating, and value quantification — not generic sales technique. That's the enablement rollout story made concrete: this is what 'certify sellers on the product' looks like when the certification tool is the product itself.

Discovery questions

• Where do you run structured one-on-one conversations today — candidate screens, customer discovery calls, exit interviews, user research — and who actually conducts them?
• How much of a scheduler's or researcher's time goes into running the conversation itself versus writing it up afterwards?
• If you could adapt the same interviewer to five different conversation types without retraining anyone, which one would you start with?
• What would make a synthetic voice feel right for a candidate or customer conversation in your brand — and what would make it feel wrong?
• How do your sellers currently rehearse a pitch before a real discovery call — role-play with a manager, a call-review programme, nothing formal?
• If a seller could get specific, structured feedback on how well they positioned your platform after every practice call — not just 'good job' — what would that change about ramp time for a new hire?

Objection handling

If a tool call fails or times out

1. The call fails to connect or drops mid-interview: switch the Configuration/Interview subtabs are force-mounted specifically so this is recoverable — flip to Configuration and back rather than re-navigating, which would tear down the claimed session. If it still won't reconnect, fall back to narrating the theme/topics live instead of running the call.
2. The voice override doesn't take (the agent speaks in its default voice instead of the picked one): don't stop to debug it live — note it, keep going, and mention afterward that per-call TTS overrides require 'Enable overrides' checked on the agent, which is a one-time setup step, not a demo-time fix.
3. The transcript panel shows nothing during the call: the flush route is shared with report-it's, so if this scenario's transcript is empty, treat it as a shared-infrastructure issue worth flagging rather than assuming this scenario specifically is broken.