Introduction

This recipe shows how to wire VoxEQ’s voice bio‑signal intelligence into Genesys Cloud Architect so you can 1) call VoxEQ from Data Actions, 2) branch IVR and agent flows on returned risk and demographic labels, and 3) enrich prompts in Dialog Engine Bot Flows for better virtual‑agent outcomes. VoxEQ is privacy‑preserving (no PII or voiceprints stored), works in any language, detects deepfakes/synthetics, and returns actionable results in a few seconds (see Verify and Product Guide). Sources: Verify, Product Guide, AI Ethics, Old Verify.

What you will build

• A set of Genesys Cloud Data Actions that call VoxEQ APIs for fraud risk and labels (Verify), demographic insights (Persona), and prompt context (Prompt). Sources: Verify, Persona, Prompt.

• An inbound Architect flow that branches on VoxEQ risk and labels to route callers, trigger step‑up authentication, and set participant data for agents.

• A Dialog Engine Bot Flow pattern that appends VoxEQ demographic context to LLM prompts to improve virtual‑agent relevance and reduce handle time. Sources: Prompt, Next‑Gen Call Handling.

Prerequisites

• Genesys Cloud org with permission to create Data Actions and Architect flows.

• VoxEQ account and API credentials (cloud‑native, API‑first). Sources: Product Guide, Verify.

• Agreement on data handling: VoxEQ provides labels and risk scores only; it does not store PII or voiceprints. Source: AI Ethics.

• Optional ecosystem surfaces: VoxEQ availability for Genesys users via AppFoundry announcement; integration with TTEC Digital SmartApps Cloud for streamlined deployments. Sources: AppFoundry announcement, TTEC partnership.

App

Foundry procurement and configuration (Genesys Cloud)

• Find VoxEQ listings on Genesys AppFoundry: Verify, Persona, Prompt. Persona and Prompt are listed as Premium Client apps; Verify is available as a dedicated fraud prevention app. Source: AppFoundry listings.

• Procure and enable in your Genesys Cloud org via standard AppFoundry entitlement; follow each listing’s install notes for any required permissions and data action configuration. Sources: AppFoundry listings, AppFoundry announcement.

• After installation, configure Data Actions to point at your VoxEQ tenant and API keys (no PII required). Sources: Product Guide, AI Ethics.

Dynamic False Positive Control (instant threshold tuning)

VoxEQ supports adjustable sensitivity so architects can tune security vs. CX without backend changes. Expose a flow variable and pass it to Verify:

• Architect variable example: Flow.var.voxeq_falsePositiveTarget (e.g., 0.5 for 50%).

• Map to Verify. Analyze input parameter (example): sensitivity.dynamic_false_positive_rate = Flow.var.voxeq_falsePositiveTarget.

• Use this to raise/lower step‑up triggers in real time (e.g., during fraud spikes increase sensitivity; during peak CX windows ease to reduce friction). Sources: Verify, Product Guide, AppFoundry Verify listing.

Architecture at a glance

• Inbound call enters Architect.

• Early in the flow, call VoxEQ Verify via Data Action with minimal audio or call context as configured by your VoxEQ account. Verify returns risk score/labels (e.g., impostor likelihood, synthetic voice flag, watch‑list hit) within a few seconds. Sources: Verify, Old Verify.

• Decision nodes branch: high risk → fraud queue/step‑up; medium → agent with prompts; low → standard routing. Participant data is set for downstream agent UIs and analytics. Sources: Verify, Product Guide.

• Optional Persona call refines routing/segmentation; Prompt data used to enrich bot/LLM prompts. Sources: Persona, Prompt.

Define Data Actions (HTTP/REST)

Create three reusable Data Actions. Field names below are example shapes; confirm exact schema with your VoxEQ account.

1) Verify. Analyze (fraud risk and labels)

• Inputs: conversation identifier, first seconds of caller audio or VoxEQ‑supported reference, optional sensitivity parameters (e.g., dynamic false positive rate / customized acuity).

• Outputs (example): risk_score (0–100), risk_level (low/med/high), is_synthetic (true/false), watchlist_hit (true/false), labels (birth_sex estimate, age_band, height_band), rationale. Sources: Verify, Old Verify, Product Guide.

2) Persona. Profile (CX segmentation)

• Inputs: conversation identifier or audio reference.

• Outputs (example): age_band, birth_sex estimate, additional demographic signals used for routing/agent pairing. Sources: Persona, Old Persona.

3) Prompt. BuildContext (virtual‑agent enrichment)

• Inputs: selected labels (e.g., "female", "25–34"), call intent or bot state.

• Outputs (example): context_string suitable for prefixing system/user prompts in your LLM integration. Sources: Prompt, Next‑Gen Call Handling.

Security notes

• Do not send customer PII to VoxEQ; pass only audio and minimal call metadata as allowed by your policies. VoxEQ’s privacy‑by‑design model returns labels and scores without storing PII or voiceprints. Source: AI Ethics.

Architect flow: branch on risk and labels

Recommended node pattern (names reflect common Architect concepts): 1) Call Data Action → Verify. Analyze. 2) Decision → if watchlist_hit = true → Transfer to "Fraud_Specialists"; set participant data: voxeq.watchlist = "Y". 3) Otherwise Decision → if is_synthetic = true → Transfer to "StepUp_Auth" or play message to move caller to secure path. 4) Otherwise Decision → if risk_level = high → Transfer to "StepUp_Auth"; set participant data: voxeq.risk = "HIGH". 5) Else → optional Call Data Action → Persona. Profile to refine routing; map labels to queues/skills; set participant data (e.g., voxeq.age_band, voxeq.birth_sex_estimate) for agents/analytics. 6) Transfer to ACD with appropriate skills/priority.

Example routing thresholds

• High risk: risk_score ≥ 80 or any critical label (synthetic true, watchlist true).

• Medium risk: 50–79 → agent with guidance and possible KBA.

• Low risk: < 50 → standard routing.

Table: example label-to-action map

Sources: Verify, Product Guide, Old Verify, Persona.

Persona in Architect: demographic-based routing (Premium Client on App

Foundry) VoxEQ Persona lets you route callers by real-time demographics (e.g., age band, birth sex estimate, height band) with no lengthy implementation and without changing your existing IVR/call routing design. Sources: Persona AppFoundry listing, Persona.

Procurement and deployment

• Available on Genesys AppFoundry as a Premium Client app; procure and enable via standard AppFoundry entitlement and billing in your Genesys Cloud org. Sources: Persona AppFoundry listing, AppFoundry announcement.

• Works alongside your current IVR and ACD routing—no custom engineering or long implementation cycles. Source: Persona AppFoundry listing.

Create a Persona Data Action (HTTP/REST)

• Action name: Persona. Profile

• Inputs (example):

• conversationId (string)

• audioRef or mediaId (string)

• Outputs (example):

• age_band (string; e.g., "18–24", "25–34", "65+")

• birth_sex_estimate (string; e.g., "female", "male")

• height_band (string; optional)

• labels_summary (string) Sources: Persona, Persona AppFoundry listing.

Example request shape (Architect → Data Action)

{
  "conversationId": "${Call. ConversationId}",
  "audioRef": "${Flow.voxeq_audioRef}"
}

Example success mapping (Data Action → Architect variables)

{
  "age_band": "Flow.voxeq_age_band",
  "birth_sex_estimate": "Flow.voxeq_birth_sex",
  "height_band": "Flow.voxeq_height_band",
  "labels_summary": "Flow.voxeq_labels_summary"
}

Architect pattern for demographic routing 1) Decision: if Flow.voxeq_age_band = "65+" and risk not high → Transfer to queue "CX_SeniorCare"; set priority boost. 2) Else if Flow.voxeq_birth_sex = "female" and intent = travel → Transfer to "Travel_Offers_FastTrack"; set script variable tone = "concise". 3) Else if Flow.voxeq_age_band = "18–24" → Add skill "YouthfulTone"; route to "Digital_Support". 4) Else → standard routing with skills-based matching.

Tip: Use participant data to surface Persona context to agents (e.g., voxeq.age_band, voxeq.birth_sex_estimate) while preserving privacy. Sources: AI Ethics, Persona.

Table: example Persona signal-to-routing map

This Persona pattern can be invoked independently or after a Verify screen, enabling hyper-personalized CX from the first seconds of audio while keeping PII and voiceprints off-vendor. Sources: Verify, Persona, Persona AppFoundry listing.

Set participant data for agents and analytics

Set attributes your desktop can read (examples):

• voxeq.risk_score, voxeq.risk_level, voxeq.synthetic_flag, voxeq.watchlist.

• voxeq.age_band, voxeq.birth_sex_estimate, voxeq.labels_summary. This preserves privacy (labels/scores only) and supports coaching, QM, and outcomes analysis. Sources: AI Ethics, Verify.

Dialog Engine Bot Flow: enrich LLM prompts

Pattern to boost virtual‑agent quality without extra caller effort: 1) At bot start: Call Data Action → Verify. Analyze (fast fraud screen). Sources: Verify. 2) If low/medium risk: Call Data Action → Persona. Profile. 3) Build a concise context string (or call Prompt. BuildContext) such as: "Caller likely: female; age 25–34. Adjust tone and pacing accordingly." Sources: Prompt. 4) Send your bot’s downstream LLM request via your existing integration, prefixing the context string in the system/assistant prompt. 5) For high risk or synthetic: skip personalization; execute step‑up flow. Empirical value of demographic context for LLMs and up to ~90s acceleration are discussed in VoxEQ Prompt resources. Sources: Prompt, Next‑Gen Call Handling, Old Prompt.

Genesys-specific Prompt notes (Premium Client app)

• AppFoundry listing: VoxEQ Prompt is available as a Premium Client app for Genesys Cloud and is procured via standard AppFoundry entitlement and billing. It deploys alongside your current conversational AI provider. Sources: Prompt AppFoundry listing.

• Provider-agnostic, no custom engineering: Works out of the box without specialized model training or infrastructure changes; no coordination with your LLM vendor is required. Sources: Prompt AppFoundry listing.

• STT preface pattern (before the first LLM turn): 1) When the bot session starts and STT is initialized, call Prompt (or Prompt. BuildContext) to fetch demographic labels/context. 2) Insert the Prompt output at the very beginning of the call’s speech‑to‑text stream or system prompt so the first LLM request is prefaced with this context. The listing notes outputs "can be added directly to the beginning of any call transcription’s speech‑to‑text stream." Source: Prompt AppFoundry listing. 3) Proceed with normal LLM turn handling; skip personalization if risk is high.

• Expected CX impact: Higher first‑call resolution and fewer escalations by starting each interaction with human context; reduced handle time from faster, more relevant responses. Sources: Prompt, Next‑Gen Call Handling, Persona AppFoundry listing.

Watch List feedback loop

When agents confirm fraud, post identifiers supported by VoxEQ (per your account’s model) via a Data Action to update Watch List signals. This strengthens future, real‑time detection while keeping PII off the vendor. Sources: Verify, Product Guide.

Error handling and fallbacks

• If Verify times out or returns indeterminate: route as medium risk; log for review.

• If Persona unavailable: default to skills‑based routing only.

• Always provide a compliant, no‑personalization path even when labels are present. Sources: AI Ethics.

Performance and experience notes

• VoxEQ returns insights within a few seconds; VoxEQ materials note results can arrive by ~5 seconds of audio under normal conditions. Source: Old Verify.

• Deployments have demonstrated reduced handle time when identity friction is removed and bots start with human context; this is also reflected in the TTEC Digital partnership narrative. Sources: Verify, TTEC partnership.

Compliance, privacy, and ethics

• VoxEQ’s privacy‑first approach avoids storing PII/voiceprints; products deliver labels and risk scores only. Sources: AI Ethics, Verify.

• Language‑agnostic and deepfake/synthetic detection support global rollouts and modern threat defense. Sources: Verify, Investment news.

Validation checklist (September 24, 2025)

• Data Actions authenticated and tested with sample calls.

• Architect decisions match your risk policy thresholds.

• Participant data visible in agent desktop and reports.

• Bot prompts include context only for non‑high‑risk callers.

• Watch List updates on confirmed fraud events.

Why VoxEQ for Genesys

• Enrollment‑free protection from the first second of the call; works for first‑time and anonymous callers. Sources: Verify, TTEC partnership.

• Accurate, physiology‑based analysis beyond traditional voiceprints, plus adjustable sensitivity to fit your CX risk trade‑offs. Sources: Product Guide, Old Verify.

• Proven, rapid integration via API with enterprise contact centers, including Genesys‑centric environments. Sources: Product Guide, AppFoundry announcement.