--- title: CX AI Features slug: cx-ai-features description: Discover ElevateAI, the advanced CX AI framework designed to analyze interactions at scale. With voice activity analysis, pre-built behavioral models, and NICE Enlighten scores, it offers quick insights into customer-agent interactions. Assess agent behav docTags: createdAt: 2022-06-14T14:53:18.000Z --- # // Overview Modern consumers expect **seamless, efficient digital experience**s, and every interaction contributes to overall customer satisfaction. To meet these experiences, organizations need tools that deliver **deep, actionable insights** into customer conversations. Powered by **NiCE's CX AI models** – and backed by decades of contact center research and billions of real-world interactions – ElevateAI provides **immediate analysis of voice and chat data at scale** using pre-built behavioral models. What's New With [ElevateAI v1.11](https://docs.elevateai.com/changelog), all declared interactions – [audio](docId\:PqLj1vqO7gTf0iqj1LBVN), [chat](docId\:iD41ifX7uC0RZDlb7RbtO), and [transcripts](docId:8OMp8XUfyRY9gaB1YIvul) – are now automatically processed through the full range of **CX AI services**. This release expands coverage across all communication types and ensures **consistent, end-to-end customer experience analysis**. Key Capabilities ElevateAI's **pre-built CX AI models** make it easy to integrate intelligence into applications and workflows, unlocking faster time-to-value (TTV): - **Automated interaction analysis** at enterprise scale - **Predictive and prescriptive insights **powered by behavioral models - **Purpose-built models** for immediate deployment and measurable impact *** # Voice Activity :::::VerticalSplit ::::VerticalSplitItem For any audio interaction, **Voice Activity Segments** returned with [CX AI Results](docId:05yDvE34Wjax0rFhFWDxM) include **start and stop times (in milliseconds)** for both **voice** and **silence** periods. This enables immediate insight into **non-interaction time** – such as hold time or extended silence – when participants are not actively engaging. In addition, ElevateAI provides **speaker diarization** for CX AI results using both the **CX** and **Echo** transcription models. Each voice segment is automatically associated with **participantOne **or **participantTwo**, giving you visibility into: - **Total talk time** per speaker - **Cross-talk detection** (when both participants speak simultaneously) - **Initiator identification** (who spoke first during cross-talk) :::hint{type="info"} **NOTE:** The **CX model** supports diarization for both **Mono (single-channel)** and **Stereo (dual-channel)** audio. The **Echo model** currently supports diarization for **Stereo (dual-channel)** only. ::: :::: ::::VerticalSplitItem :::CodeblockTabs Sample CX AI Response ```json { "allParticipants:" { ... "voiceActivitySegments": [ { "startTimeOffset": 0, "endTimeOffset": 4920, "result": "VOICE" }, { "startTimeOffset": 4920, "endTimeOffset": 7560, "result": "SILENCE" } ... ] } ... } ``` ::: :::: ::::: *** # Enlighten AI Bundles [NiCE Enlighten Scores](https://www.elevateai.com/ai-models/) are generated for each behavioral model and provided both as an **overall score** and as **quartile-based scores** (segmented by interaction stage). This allows users to track behavioral shifts over the course of a conversation and pinpoint when key behaviors occur. ## Behaviors for Customer Satisfaction :::ExpandableHeading ## Acknowledge Loyalty The **Acknowledge Loyalty** score imeasures how effectively an agent acknowledges a caller's long-term relationship with the organization and expresses appreciation for their loyalty. - **Higher scores** indicate the agent consistently recognizes customer tenure in a natural, conversational way. - The model evaluates the **entire conversation**, using language from both the agent and the customer. - For the most accurate results, use the `allParticipants` score, which reflects the interaction as a whole. ::: :::ExpandableHeading ## Active Listening The **Active Listening** score reflects how well an agent listens and responds throughout a conversation. - **Higher scores** indicate that the agent is engaged, responsive, and avoids asking the customer to repeat information unnecessarily. - The model evaluates language from **both agent and customer** to assess listening behavior. - For the most accurate results, use the `allParticipants `score, which reflects the interaction as a whole. ::: :::ExpandableHeading ##  Be Empathetic The **Be Empathetic** score reflects how well an agent acknowledges the caller's issues and demonstrates understanding of their impact or hardship. - **Higher scores **indicate stronger predicted proficiency in showing empathy during conversations. - The model evaluates **both agent and customer language** to assess empathetic behavior. - For the most accurate results, use the `allParticipants` score, which reflects the interaction as a whole. ::: :::ExpandableHeading ## Build Rapport The **Build Rapport **score measures how effectively an agent creates a personal connection with the caller by engaging in **friendly, general conversation** outside the main issue. - **Higher scores** indicate strong proficiency in building trust and comfort through small talk or positive interaction. - The model analyzes **both agent and customer input** to assess rapport-building behavior. - For the most accurate results, use the `allParticipants` score, which reflects the interaction as a whole. ::: :::ExpandableHeading ## Demonstrate Ownership The **Demonstrate Ownership **score measures how well an agent shows understanding of the caller's issue and a commitment to resolving it. - **Higher scores** indicate strong proficiency in providing reassurance and taking responsibility. - The model evaluates **both sides of the conversation** to assess ownership behaviors. - For the most accurate results, use the `allParticipants` score, which reflects the interaction as a whole. ::: :::ExpandableHeading ## Effective Questioning The **Effective Questioning **score reflects how well an agent asks **thoughtful, purposeful questions** to better understand the caller's experience, concerns, or needs. - **Higher scores** indicate stronger proficiency in uncovering issues and opportunities through meaningful questions. - The model analyzes **agent and caller exchanges** across the full conversation. - For the most accurate results, use the **allParticipants** score, which reflects the interaction as a whole. ::: :::ExpandableHeading ## Inappropriate Action The **Inappropriate Action** score reflects how well an agent avoids unprofessional or offensive behaviors – such as refusing a reasonable transfer request, using inappropriate language, or showing disrespect. - **Higher scores** indicate stronger proficiency in consistently maintaining professional conduct. - The model analyzes **both agent and customer language** to evaluate this behavior. - For the most accurate results, use the `allParticipants` score, which reflects the interaction as a whole. ::: :::ExpandableHeading ## Promote Self-Service The **Promote Self-Service** score measures how effectively an agent encourages customers to use available self-service options – such as an IVR system, website, or mobile app – when appropriate. - **Higher scores** indicate strong proficiency in guiding customers toward tools that can streamline future interactions. - The model evaluates **language from both agent and customer** to assess this behavior. - For the most accurate results, use the `allParticipants` score, which reflects the interaction as a whole. ::: :::ExpandableHeading ## Sentiment The **Sentiment** score predicts the likelihood that a customer would give a** positive post-interaction rating** in a survey. - The score is derived from both **spoken language** (*what* was said) and **acoustic cues** (*how *it was said), including pitch, tone, laughter detection, and cross-talk. - The model analyzes the **full conversation** across both agent and customer to determine customer sentiment. - For the most accurate results, use the `allParticipants` score, which reflects the interaction as a whole. ::: :::ExpandableHeading ## Set Expectations The **Set Expectations **score measures how effectively an agent communicates **next steps**, outlines **required actions**, and ensures the customer understands what to expect after the interaction. - **Higher scores** indicate strong proficiency in providing clear, proactive guidance that builds confidence and reduces uncertainty. - The model evaluates **language from both agent and customer **to assess this behavior. - For the most accurate results, use the `allParticipants` score, which reflects the interaction as a whole. ::: ## Vulnerable Customer Behaviors A **Vulnerable Customer** is defined by the UK Financial Conduct Authority (FCA) as: *"Someone who, due to their personal circumstances, is especially susceptible to harm, particularly when a firm is not acting with appropriate levels of care."* These customers may require additional support or actions from organizations to ensure they receive **fair, equitable, and appropriate service**. Built by NiCE, ElevateAI's **Vulnerable Customer models** help identify potential signals of vulnerability in real-time interactions. :::ExpandableHeading ## Vulnerable Customer The **Vulnerable Customer **score predicts the likelihood that a caller may fall into one or more of the five subcategories commonly referenced in contact centers:** health, life events, resilience, capability, and age-related factors.** - **Higher scores** suggest customer language or behaviors that may signal potential vulnerability. - The model evaluates **both agent and customer language** across the interaction. - For the most accurate results, use the `allParticipants` score, which reflects the conversation as a whole. ::: :::ExpandableHeading ## Vulnerable Capability The **Vulnerable Capability **score predicts the likelihood that a customer has **limited knowledge, confidence, or skills** in managing financial matters. This may also reflect challenges in areas such as **financial literacy or digital skils**. - **Higher scores** indicate the customer may require extra support to navigate the interaction effectively. - The model analyzes **both agent and customer input** to identify signals of limited capability. - For the most accurate results, use the `allParticipants` score, which reflects the conversation as a whole. ::: :::ExpandableHeading ## Vulnerable Health The **Vulnerable Health** score predicts the likelihood that a customer is experiencing **physical health conditions or illnesses** that impact their ability to manage daily activities or interactions. - **Higher scores** suggest the customer may require additional assistance due to health-related challenges. - The model evaluates **both agent and customer language** throughout the interaction. - For the most accurate results, use the `allParticipants` score, which reflects the conversation as a whole. ::: :::ExpandableHeading ## Vulnerable Life Event The **Vulnerable Life Event **score predicts the likelihood that a customer is undergoing significant **life changes or hardships**, such as bereavement, job loss, or relationship breakdown. - **Higher scores** indicate the customer may need additional care and support during the interaction. - The model analyzes **language from both agent and customer** to identify potential signals. - For the most accurate results, use the `allParticipants `score, which reflects the conversation as a whole. ::: :::ExpandableHeading ## Vulnerable Mental Health The **Vulnerable Mental Health **score predicts the likelihood that a customer is experiencing **mental health challenges **that may affect their ability to manage daily activities or decision making. - **Higher scores** suggest the customer may require additional support or accomodations. - The model evaluates **both agent and customer language** to identify relevant indicators. - For the most accurate results, use the `allParticipants` score, which reflects the conversation as a whole. ::: :::ExpandableHeading ## Vulnerable Resilience The **Vulnerable Resilience** score predicts the likelihood that a customer has a **reduced ability to cope** with financial or emotional stress. - **Higher scores** indicate the customer may need extra support due to difficulty managing stress or pressure. - The model evaluates **both agent and customer language** to detect signals of reduced resilience. - For the most accurate results, use the` allParticipants `score, which reflects the conversation as a whole. ::: *** Additional models in currently in **beta** include: - **FInancial Distress** - **High-Risk Transactions** For the most up-to-date details on new models and features, [visit the ElevateAI Release Notes](https://docs.elevateai.com/changelog). **Need more help?** Contact the [ElevateAI Support team](mailto\:support@elevateai.com).