Talk track

1. This is the same proactive-disruption shape as scenario 2 — something goes wrong, the organisation notices before the customer does, and reaches out with options — applied to fraud instead of travel. It's the strongest evidence that the pattern generalises, not just the flight-delay demo trick.
2. Watch the gate: freeze_card is only ever called after the customer has explicitly and verbally agreed to it. That's the same Workflow-style deterministic gating from scenario 1's shipping upgrade and scenario 2's rebooking, now on an action with real regulatory weight — this is the moment to point at when a buyer asks how irreversible actions are controlled.
3. The value story is fraud loss avoided, not just contact deflection — call out that this changes the ROI conversation from a cost centre (support headcount) to a risk number (fraud losses prevented), which is usually a bigger number and a different buyer.

Discovery questions

• Where does a suspected-fraud alert reach your customer today — SMS, push, app, or an outbound call queue — and how long does that typically take from detection to contact?
• What's your current escalation path when a customer doesn't recognise a transaction? How many steps/systems does an agent touch before a card is frozen?
• Is a card freeze today a human decision, a rules-engine decision, or some mix — and where would a voice agent need to sit relative to your existing fraud rules engine?
• What's your average fraud loss per confirmed case, and your average cost per fraud-related inbound contact? (Feeds directly into the value panel's numbers.)

Objection handling

If a tool call fails or times out

1. get_fraud_alert returns nothing or times out: say so plainly — "let me pull that back up" — and use the ops console to confirm the alert is still flagged rather than guessing at what the agent saw. Never let dead air stand in for an answer.
2. confirm_transaction_recognition or freeze_card fails after the customer has already said yes verbally: don't re-ask the consent question — that's the one thing that must not repeat. Reassure them the freeze is being retried, then fall back to the ops console to freeze the card manually and confirm out loud that it's done.
3. If the agent seems to freeze a card without a clear verbal yes on the recording, stop and call it out live rather than let it pass — it's the single moment this scenario's regulatory story depends on, and glossing over a gate failure undercuts the point of showing the gate at all.