Allianz annuity demo¶

Status: Current | Last reviewed: 2026-09-30

Use the Allianz demo to show a human representative receiving bereavement guidance and a voice assistant working with a human specialist. All policy, claim and commission records are synthetic.

Both Vanguard and Macau have the published Beneficiary Copilot (4446), Producer Voice Agent (4444) and Producer Specialist (4445). Open the related conversations in the same environment.

For the DevSDO console and +1 833-799-5477, use the phone and Salesforce walkthrough.

Beneficiary call¶

Open Allianz Beneficiary Assist and start a Copilot test with separate customer and representative channels.

Act as Elena Martinez: “My husband Robert passed away last week. I do not know where to start.” The Copilot shows brief empathy and pacing suggestions. As Jordan, acknowledge the loss and offer one step at a time.

Ask which documents are needed and how long review takes. The knowledge card shows the approved demo explanation and source. The intake form fills only details actually stated. Suggestions are guidance you can adapt; the Copilot does not speak to the caller.

Policy facts remain hidden until the host supplies trusted verification. The verified demo record shows a $312,500 estimated gross benefit that already includes the $27,500 rider enhancement. This is not a final payment quote.

Optional clarity reminder¶

As Elena, say “Could you slow down please?” A separate coaching card suggests: “Pause briefly. Use short sentences and check the caller can follow.” As Jordan, acknowledge the request and explain one step at a time.

The reminder disappears after 15 seconds. A fresh clarity warning restarts that timer on the same card. Ordinary dialogue does not extend it. Empathy and claims guidance continue in their own cards.

The agent's transcript alone can trigger the card. One clearly garbled segment with essential words missing, implausible substitutions or meaning obscured by broken syntax can prompt “Pause and check understanding.” There does not need to be a customer complaint or a second bad segment. Borderline wording needs more evidence. The reminder acknowledges that transcription may be the cause. Minor grammar mistakes, ordinary fillers or fragments, understandable nonstandard grammar and clear self-corrections should not trigger it. It does not assess accent or nationality.

Use the saved controlled transcript examples to demonstrate the second signal; speaking quickly does not guarantee a bad transcript. Text alone cannot establish actual speaking speed or pronunciation.

Producer call and human supervision¶

Open Allianz Producer Services and start a voice preview. Call as David Chen of Chen Financial Group, producer ID PR-004781, for Helen Carter’s policy VA-5830194. Complete the secure verification step when offered.

Ask for policy values and an in-force illustration using the current premium schedule and no additional withdrawals. Taylor confirms the assumptions and requests specialist review. The visual companion shows retrieved policy values and the illustration assumptions.

Open the connected Allianz Producer Specialist conversation to follow the live call. Ask the assistant to relay a clarification question if useful.

The specialist request goes to Salesforce. Open that request and its connected Allianz Producer Specialist chat, which must be linked to the same live call. The workflow finishes after sending the request; it does not wait for approval.

In the latest Vanguard version, a queued review has the short demo tracking reference REV-291640. Taylor should read that reference clearly, rather than the workflow's long internal ID. It means the illustration is awaiting review, not that the illustration has been submitted.

Type a question, an explicit approval or a rejection in the supervisor chat and ask it to relay the message. Taylor receives the guidance during the call. The caller’s reply returns through the live transcript. You do not need the old review desk or a workflow approval button; /allianz-demo now explains this flow.

Approval is separate from completed submission. For the script’s synthetic successful ending, the human specialist explicitly confirms: “I submitted the illustration request using the current premium schedule and no additional withdrawals, reference IFR-291640. The document has not yet been generated.” Use that example only for the matching synthetic request. Taylor should attribute the confirmation to the specialist. Approval alone must not produce a reference or a claim that submission is complete.

If you finish while approval or submission remains pending, Taylor should say so in the recap. Rejection should be reported as not approved and not submitted. Start a fresh preview after the supervision correction; earlier calls retain their original configuration.

Ask about commission timing and whether it is guaranteed. The displayed $1,248.60 and October 15, 2026 cycle are subject to eligibility, adjustments and final processing. Ask Taylor to record that explanation in the synthetic service notes.

Taylor should read monetary amounts in full dollars and cents, answer follow-up value questions without repeatedly asking about withdrawals, and give a short recap focused on outcomes and next steps. Previously explained figures stay out of the default recap; ask for a detailed recap to hear them again. One short reference may be included for follow-up. The internal summary, recorded notes and conversation history retain the servicing details.

“Can you recap what we did?” should keep the call open for another question. “That is everything, goodbye” closes it. If Taylor just gave a recap, the goodbye should not repeat it. A note-save confirmation should also be brief. Taylor should let you finish an incomplete thought and use one short acknowledgement while checking, without starting an explanatory sentence and restarting it.

Reading latency results¶

Common empathy and knowledge cues use approved phrase matches. Verified facts display the data tool's exact result; form extraction and wrap-up use a model. Test reports separate these timings; they do not promise that every suggestion arrives within one second.

Start the Copilot before the caller begins speaking. In the local preconnected test, twelve cues took 405 ms median and 479 ms p95 from detected speech end to server publication. An earlier startup test had a three-second first cue. These small-sample results exclude browser rendering and telephone-network delay.

The local tests do not exercise a production telephone carrier or change Allianz systems.

September 30 voice and visual update¶

In Vanguard, the reference and recap changes are published in version 4. Start a fresh console preview of Allianz Producer Services to use the refined conversation. Taylor should give brief acknowledgements while checking a request, ask only for missing identity details, and confirm illustration assumptions once. The default recap is now about 35–55 words, including anything still pending; a detailed recap is available on request.

Accept the visual companion to see Allianz-branded policy, illustration, commission and service-summary cards. They use the Allianz logo, fonts and blue, with layouts suited to a phone. You can change the withdrawal assumption on the illustration card and confirm it there or by voice. A changed assumption needs fresh review; a pending review does not show a submission reference.

The specialist responds through the connected chat. Taylor should not promise a review deadline or that a specialist will join the phone call. For the successful scripted ending, the specialist must explicitly confirm completed submission and its reference.

The latest three fresh voice probes measured a 2.04-second median for the first audible response, ranging from 1.40 to 5.64 seconds. This is a small sample of acknowledgements or initial answers, not a subsecond guarantee or the time to complete a policy lookup. Occasional repeated wording remains, and a specialist status update can still interrupt an answer.

For the clarity reminder, three approved customer requests took 399–418 ms from detected speech end to server publication in a preconnected local test. The card disappeared 15 seconds after its last warning. Separate tests of the contextual model took 733 ms median and 1.30 seconds p95 from final text to a card-or-no-card decision. Those model timings exclude transcription and browser display, so subtle transcript-based cues are not guaranteed within one second.