Generative AI Customer Service Glossary: 80 Key Terms Defined

The definitive reference for AI, contact center, and customer experience terminology — from agentic AI to zero-shot learning. Every term used in this ebook, and every term you'll encounter in AI CX strategy discussions, defined with precision.

A

Agentic AI

An AI system capable of autonomously executing multi-step tasks across connected systems — going beyond generating responses to actually taking actions: retrieving data, making decisions, calling APIs, completing transactions, and managing workflows end-to-end without continuous human direction. In customer service, agentic AI resolves service requests that require backend system interactions, not just information retrieval. Distinguished from generative AI (which produces content) by its action-taking capability. See Chapter 5.

Related: Generative AI, LLM, CXone Orchestrator, CXone Autopilot

After-Call Work (ACW)

The administrative tasks agents complete after a customer interaction ends — writing call summaries, logging disposition codes, updating CRM records, and scheduling follow-up tasks. ACW typically accounts for 15–20% of total agent handle time. AI automation of ACW (via tools like CXone Copilot) eliminates this cost, contributing significantly to AHT reduction. Also called “wrap time” or “post-call work.”

AI Bias

Systematic errors in AI outputs that produce unfair or inconsistent results across customer demographic groups — resulting from biased training data, proxy variables correlated with protected characteristics, or optimization objectives that don’t include fairness constraints. AI bias in customer service can manifest as different containment rates, escalation rates, or quality of service by customer segment. Requires active monitoring and mitigation. See Chapter 11.

AI Containment Rate

The percentage of customer contacts that AI self-service resolves end-to-end without transfer to a human agent. The primary operational metric for AI self-service performance. NiCE CXone deployments achieve 70–88% containment in production. Containment rate is driven primarily by knowledge base quality and coverage, intent recognition accuracy, and backend system integration depth. See Chapter 3.

Related: Deflection Rate, Escalation Rate, CXone Autopilot

AI Hallucination

When an AI language model generates plausible-sounding but factually incorrect content — inventing facts, policies, or details not present in its knowledge base. A critical risk in customer service AI where inaccurate policy or product information creates customer harm and liability. Mitigated through RAG architecture, confidence thresholds, knowledge base curation, and output guardrails. See Chapter 11.

AI-Native Contact Center

A contact center built from the ground up on AI-first architecture — where AI self-service, agent assist, quality management, and analytics are first-class platform capabilities, not add-ons to legacy infrastructure. Contrasted with “AI overlay” approaches that attempt to add AI capabilities to legacy telephony and IVR systems. AI-native platforms like NiCE CXone extract substantially more AI value than AI-overlay deployments. See Chapter 12.

AI Orchestration

The coordination of multiple AI agents, tools, and systems to execute complex multi-step workflows — routing tasks between specialized AI components, managing state across workflow steps, handling exceptions, and ensuring coherent end-to-end execution. In NiCE CXone, orchestration is provided by CXone Orchestrator. See Chapter 5.

Related: CXone Orchestrator, Agentic AI, Workflow Automation

Average Handle Time (AHT)

The average duration of a customer interaction, measured from the moment an agent answers to the end of after-call work. A primary contact center efficiency metric. AI agent assist reduces AHT by surfacing knowledge instantly (eliminating hold-and-search time), suggesting responses, and automating ACW. NiCE CXone deployments achieve 55% AHT reduction through CXone Copilot. See Chapter 4.

B

Bot

An automated software system that handles customer interactions without human involvement. “Bot” is often used interchangeably with “chatbot” and “virtual agent,” though in contemporary usage, “virtual agent” or “AI agent” typically implies generative AI capability, while “bot” may imply older rule-based or scripted automation. See also: Virtual Agent, Chatbot.

Business Process Automation (BPA)

The use of technology to automate repetitive business processes that previously required human execution. In contact centers, BPA increasingly incorporates AI to handle variable, judgment-requiring tasks (not just rule-based workflows). Agentic AI represents the highest form of BPA — capable of automating processes that require natural language understanding, contextual decision-making, and multi-system coordination.

C

CCaaS (Contact Center as a Service)

Cloud-delivered contact center technology — providing omnichannel routing, workforce management, analytics, and increasingly AI capabilities as a subscription service rather than on-premise infrastructure. CCaaS platforms eliminate the capital expense and maintenance burden of legacy contact center hardware and software. NiCE CXone is the leading CCaaS platform with integrated AI capabilities. See Chapter 2.

Chatbot

An automated conversational interface that handles text-based customer interactions. Traditional chatbots use rule-based decision trees and pattern matching; AI-powered chatbots use large language models for natural language understanding and generation. In customer service, the term is increasingly supplanted by “virtual agent” or “AI agent” when referring to generative AI-powered systems with transactional capabilities beyond simple Q&A.

Churn Prediction

AI-based analysis of customer behavioral signals, interaction history, and account data to identify customers at elevated risk of cancellation or attrition before they make a churn decision. Enables proactive retention intervention — outreach, offers, or service recovery — at the moment of highest intervention value. A key capability in predictive CX and proactive AI agent systems.

Confidence Score

A numerical measure (typically 0–1 or 0–100%) indicating an AI model’s certainty in its output — whether an intent classification, a retrieved knowledge item, or a generated response. In customer service AI, confidence scores drive routing decisions: high-confidence responses are delivered to customers; low-confidence responses trigger escalation to human agents or clarification requests. Calibrating confidence thresholds is a key lever in containment rate optimization.

Contact Center AI

The broad category of AI technologies applied to contact center operations — including AI self-service (virtual agents), AI agent assist, AI quality management, AI workforce management, and AI analytics. Contact center AI is the application context for generative AI in customer service operations. See Chapter 1 and Chapter 2.

Conversational AI

AI systems designed for natural language dialogue with humans — encompassing intent recognition, context management across conversation turns, response generation, and dialogue flow management. Conversational AI is the foundation of AI virtual agents and chatbots. NiCE CXone’s conversational AI is powered by CXone and operates across voice and digital channels. See NiCE Conversational AI Platform.

Cost Per Interaction (CPI)

The fully-loaded cost of handling a single customer interaction — including agent labor, technology, overhead, and quality assurance costs. The primary economic metric in contact center AI ROI calculations. Human agent interactions cost approximately $8.01 on average; AI-handled interactions cost approximately $0.25 — a 97% reduction that drives the economics of AI self-service investment. See Chapter 7.

Customer Experience (CX)

The totality of a customer’s perceptions, emotions, and interactions across all touchpoints with a brand — from initial awareness through purchase, service, and ongoing relationship. In AI context, CX refers specifically to the quality and consistency of service experiences delivered by AI and human agents. See What Is Customer Experience and What Is a CX Platform.

Customer Satisfaction Score (CSAT)

A survey-based metric measuring customer satisfaction with a specific interaction, typically collected immediately post-interaction via a 1–5 or 1–10 scale question. AI impacts CSAT both directly (through resolution quality and speed) and indirectly (through agent experience improvement and reduced AHT). Leading AI deployments achieve AI interaction CSAT parity with or exceeding human agent CSAT.

CXone AI Studio (NiCE)

NiCE CXone’s AI configuration and customization platform — enabling CX architects and operations teams to build AI workflows, configure knowledge retrieval, integrate backend systems, and tune AI performance without AI engineering expertise. Provides visual workflow builders, integration connectors, testing environments, and performance dashboards. The primary tool for ongoing AI optimization post-deployment. See Chapter 2.

CXone Agents (NiCE)

The omnichannel routing and agent desktop component of NiCE CXone — providing unified handling of voice, chat, email, social, and messaging channels in a single agent workspace, with AI-powered routing, real-time assistance from CXone Copilot, and integrated context from CRM and interaction history. See Chapter 2.

CXone Autopilot (NiCE)

NiCE CXone’s AI self-service platform — the virtual agent that handles customer interactions end-to-end without human agents. Powered by CXone, Autopilot achieves 97.8% intent recognition accuracy and 70–88% containment in production. Operates across voice, chat, and digital channels with intelligent escalation to CXone Agents when resolution is outside AI scope. See Chapter 3.

CXone Copilot (NiCE)

NiCE CXone’s real-time AI agent assist product — analyzing live conversations and surfacing knowledge, suggested responses, compliance guidance, next-best-action recommendations, and sentiment alerts in real time. Post-interaction, Copilot automates summary generation and ACW. Delivers 55% AHT reduction and drives agent attrition reduction from 42% to 19%. See Chapter 4.

CXone Orchestrator (NiCE)

NiCE CXone’s agentic AI workflow engine — coordinating multi-step, multi-system service workflows that require integration across CRM, order management, billing, and other backend systems. Enables AI to complete complex service transactions (not just answer questions) by orchestrating sequences of system actions with appropriate decision logic and exception handling. See Chapter 5.

D

Deflection

The redirection of a customer contact from a higher-cost channel or human agents to a lower-cost AI self-service channel, where the contact is resolved without human involvement. Deflection rate is a key ROI metric — each deflected interaction saves approximately $7.76 (the difference between $8.01 human cost and $0.25 AI cost). Distinguished from containment: deflection typically refers to contacts diverted before reaching agents; containment refers to contacts resolved within an AI channel after initiation.

Digital-First CX

A customer experience strategy that prioritizes digital channel resolution (chat, messaging, web, app) over phone/voice as the primary service channel. Generative AI enables effective digital-first CX by giving digital channels the resolution capability that previously only human voice agents could provide — making digital deflection viable for complex interactions, not just simple FAQ queries.

E

Embeddings

Dense vector representations of text that capture semantic meaning in mathematical form — enabling AI systems to measure the conceptual similarity between a customer’s question and knowledge base content. Embeddings power the retrieval component of RAG systems: when a customer asks a question, the system converts it to an embedding and finds the knowledge items with the most similar embeddings. The quality of embeddings directly affects knowledge retrieval accuracy and, therefore, response quality.

Escalation

The transfer of a customer interaction from AI self-service to a human agent — triggered when AI reaches its resolution boundary (outside knowledge scope, customer requests agent, confidence threshold not met, or regulatory requirement). In AI-native platforms like NiCE CXone, escalation transfers full interaction context — conversation history, resolved intent, customer data, sentiment analysis — to the receiving agent, enabling immediate problem-solving without customer repetition.

Experience Memory (NiCE)

NiCE CXone’s persistent cross-channel customer context layer — storing and retrieving interaction history, stated preferences, resolved issues, and relationship signals across all channels and contacts. Enables AI personalization by providing accumulated customer context to inform each interaction. The technical foundation of relationship continuity in AI customer service. See Chapter 8.

F

Fine-Tuning

The process of further training a pre-trained language model on domain-specific data to improve its performance on specialized tasks. In customer service AI, fine-tuning adapts a general-purpose LLM to understand industry-specific terminology, brand voice, product specifics, and customer service interaction patterns. Distinguished from RAG: fine-tuning updates model weights through additional training; RAG augments model responses with retrieved external knowledge without retraining.

First Contact Resolution (FCR)

The percentage of customer contacts fully resolved in a single interaction — without callback, follow-up, or escalation to another tier. A primary quality metric in contact center operations. Generative AI improves FCR by equipping virtual agents and human agents with comprehensive knowledge access and next-best-action guidance, reducing the “I’ll need to call you back” outcomes that degrade FCR rates and drive repeat contact costs.

Foundation Model

A large AI model trained on broad data at scale, designed to be adapted to a wide range of downstream tasks — the “foundation” on which specialized AI applications are built. GPT-4, Claude, Gemini, and LLaMA are examples of foundation models. In customer service AI, foundation models are typically fine-tuned or augmented with domain-specific knowledge (via RAG) for deployment. NiCE CXone’s AI is trained specifically on customer service data, giving it domain advantages over general foundation models.

G

Generative AI

A category of AI that generates new content — text, images, audio, code, or other outputs — rather than classifying or predicting from existing categories. In customer service, generative AI uses large language models to generate natural-language responses to customer queries, create interaction summaries, produce coaching content, and draft communications. The transformative property: generative AI can respond to novel inputs it has never seen before, unlike rule-based systems that can only handle scripted scenarios. See Chapter 1.

Guardrails

Constraints placed on AI systems to prevent unsafe, inaccurate, or off-brand outputs. In customer service AI, guardrails include: topic scope restrictions (AI only discusses relevant topics), PII handling rules (AI doesn’t repeat sensitive data in outputs), prohibited language filters, confidence thresholds (AI escalates rather than guessing), and brand voice enforcement. Guardrails are a core governance requirement for production customer service AI. See Chapter 11.

H

Human-in-the-Loop (HITL)

An AI system design that keeps humans involved in decision-making at defined points in an AI workflow — either for approval before consequential actions, exception handling when AI confidence is low, or quality review of AI outputs. HITL is an important governance mechanism for high-stakes customer service AI use cases (high-value transactions, sensitive escalations, regulated contexts). Represents a middle path between full automation and full human handling.

I

Intent Recognition

The AI process of identifying what a customer is trying to accomplish from their natural language input — the “why” behind a contact. Intent recognition is the foundation of all downstream AI performance: accurate intent recognition enables correct knowledge retrieval, appropriate workflow routing, and relevant response generation. NiCE CXone’s AI achieves 97.8% intent recognition accuracy, compared to 85–92% for general-purpose LLMs in customer service contexts. See Chapter 1.

IVR (Interactive Voice Response)

Traditional touch-tone or basic voice-recognition phone systems that route customers through menu trees to self-serve or reach the right agent. Legacy IVR systems use rigid, scripted menus — creating the frustrating “press 1 for... press 2 for...” customer experience. Conversational AI and generative AI virtual agents replace IVR menus with natural language voice interaction, eliminating menu navigation and enabling genuine voice self-service. See Chapter 10.

J

Journey Analytics

AI-powered analysis of the complete customer journey across all touchpoints and channels — identifying friction points, drop-off moments, channel switching patterns, and resolution path efficiency. In NiCE CXone, journey analytics uses CXone to analyze interaction data at scale, identifying systemic improvement opportunities across the entire customer service operation rather than individual interaction quality. See Chapter 6.

K

Knowledge Base

The structured repository of information — product details, policies, procedures, FAQs, troubleshooting guides — that AI retrieves from to generate responses. Knowledge base quality is the primary determinant of AI containment rate and response accuracy. Well-curated, current, clearly-structured knowledge bases enable 80–88% containment; poorly maintained knowledge bases result in low containment regardless of AI capability. Knowledge base preparation is the highest-leverage pre-deployment activity. See Chapter 9.

Knowledge Graph

A structured representation of entities, their attributes, and the relationships between them — enabling AI to reason about complex, interconnected information rather than just retrieving isolated documents. In customer service AI, knowledge graphs enable understanding of product relationships, customer account structures, and service dependencies that flat knowledge bases cannot represent.

L

Large Language Model (LLM)

An AI model trained on massive text datasets to understand and generate natural language at scale. LLMs learn statistical patterns across language — enabling them to understand context, disambiguate meaning, and generate coherent, contextually appropriate text. The foundation of all generative AI in customer service. Key characteristics relevant to customer service: context window size (how much conversation history the model can consider), latency (response speed for real-time interactions), and domain-specific training (customer service vs general text). See Chapter 1.

Latency

The time delay between a customer’s input and the AI’s response. Critical in customer service AI — high latency creates frustrating interaction experiences and degrades containment rates (customers abandon high-latency AI channels). Production customer service AI requires sub-2-second response latency for acceptable customer experience; sub-1-second is the standard for leading deployments. Latency is affected by model size, inference infrastructure, RAG retrieval time, and network performance.

M

Multimodal AI

AI systems that process and generate multiple types of data — text, voice, images, and video — within unified workflows. In customer service, multimodal AI enables image-in-conversation support (customers share photos of damaged products), video-assisted interactions, and document processing during contacts. Moving from pilot to production in 2026 as contact centers deploy use cases that text-only AI could not address. See Chapter 12.

Multi-Turn Conversation

A dialogue between a customer and AI that spans multiple exchanges — maintaining context across turns rather than treating each customer message as an isolated query. Multi-turn capability is essential for realistic customer service interactions, which typically require 3–8 exchanges to reach resolution. Generative AI systems maintain conversation context across turns; legacy chatbots often lose context between messages, forcing customers to repeat information.

N

Natural Language Processing (NLP)

The field of AI concerned with enabling computers to understand, interpret, and generate human language. NLP encompasses intent recognition, entity extraction, sentiment analysis, summarization, and language generation — all foundational capabilities in customer service AI. LLMs represent the current state of the art in NLP, having supplanted earlier rule-based and statistical NLP approaches for conversational AI applications.

Natural Language Understanding (NLU)

The AI capability of comprehending the meaning and intent of human language — distinct from merely pattern-matching or classifying text. NLU enables AI to understand that “I want to cancel my plan,” “I’d like to stop my subscription,” and “how do I get out of my contract?” all express the same customer intent despite different phrasing. Strong NLU is the foundation of high containment rates and customer satisfaction with AI interactions.

Next-Best Action (NBA)

AI-generated recommendations for the optimal next step in a customer interaction — surfaced to agents in real time to guide resolution, compliance, or upsell decisions. Next-best action AI analyzes the current interaction context, customer history, and available resolution options to recommend the specific action most likely to achieve the desired outcome. A core capability of CXone Copilot and AI agent assist platforms generally. See Chapter 4.

NiCE CXone (NiCE)

The world’s leading AI-native contact center platform — a cloud-based CCaaS solution combining omnichannel routing, AI self-service (CXone Autopilot), AI agent assist (CXone Copilot), AI quality management, workforce management, and analytics on a unified platform powered by CXone. Gartner Magic Quadrant Leader for 12 consecutive years. See Chapter 2 and NiCE CXone product page.

O

Omnichannel

A customer service approach that provides consistent, integrated service across all communication channels — phone, chat, email, SMS, social media, messaging apps — with unified customer context across channels. AI-native omnichannel platforms maintain consistent AI capability and customer context regardless of channel, enabling customers to switch channels mid-resolution without losing context or repeating information. See Chapter 2.

P

Personally Identifiable Information (PII)

Data that can be used to identify a specific individual — including names, addresses, account numbers, Social Security numbers, payment card data, and other personal identifiers. In customer service AI, PII handling requires: detection and masking in AI inputs and outputs, access control limiting AI access to minimum necessary data, retention policies governing PII storage in interaction logs, and audit logging of PII access. Improper PII handling is a primary AI compliance risk. See Chapter 11.

Predictive CX

A customer service model in which AI monitors behavioral signals, interaction history, and operational data to anticipate customer needs and initiate proactive service before customers contact support. Transforms service from reactive (customers call when they have a problem) to anticipatory (AI reaches out before problems escalate). Implemented via NiCE CXone’s Proactive AI Agent and CXone’s predictive analytics capabilities. See Chapter 12.

Prompt Engineering

The craft of designing instructions and context (prompts) that guide LLM behavior toward desired outputs — affecting response quality, tone, scope, format, and accuracy. In customer service AI configuration, prompt engineering determines how AI systems respond to customer queries, what information they include or exclude, and how they handle edge cases. CXone AI Studio provides tools for iterative prompt engineering and testing without requiring AI engineering expertise.

Proactive AI Agent (NiCE)

NiCE CXone’s outbound AI capability — an AI system that monitors trigger conditions across connected operational systems and initiates customer outreach proactively, before customers need to contact support. Use cases include shipment delay notifications, billing alerts, appointment reminders, churn-risk intervention, and service incident communications. See Chapter 5.

Q

Quality Management (QM)

The practice of evaluating customer interactions for quality, compliance, and coaching purposes. Traditional QM samples 1–5% of interactions for manual review. AI-powered QM evaluates 100% of interactions automatically — providing complete coverage, consistent scoring criteria, and eliminating evaluator subjectivity. AI QM in NiCE CXone scores interactions using the same CXone engine that powers intent recognition, maintaining scoring consistency across the entire interaction dataset.

R

Retrieval-Augmented Generation (RAG)

An AI architecture that combines knowledge retrieval with language generation — first retrieving the most relevant documents from a knowledge base, then generating a response grounded in those retrieved documents. RAG prevents hallucination by anchoring generation to retrieved facts, enables real-time knowledge updates without model retraining, and allows organizations to control AI knowledge by curating the retrieval corpus. The standard architecture for production customer service AI. See Chapter 11.

Return on Investment (ROI)

The financial return on an AI investment, calculated as (net benefit / total investment cost) × 100%. NiCE CXone customers achieve 320–650% ROI over three years from generative AI deployments, driven by containment cost savings, AHT reduction, attrition reduction, and quality improvement. ROI typically turns positive within 6–12 months of go-live. See Chapter 7 and the NiCE AI Value Calculator.

S

Self-Service AI

AI that enables customers to resolve their own service needs — finding information, completing transactions, and solving problems — without human agent involvement. Generative AI self-service goes beyond FAQ-matching to handle complex natural language queries, multi-step transactions, and contextually adaptive dialogue. The primary source of AI ROI in customer service through cost reduction and 24/7 availability. See Chapter 3.

Sentiment Analysis

AI analysis of customer language, tone, and behavior to assess emotional state — identifying frustration, satisfaction, confusion, or urgency in real time during interactions. In CXone Copilot, sentiment analysis triggers alerts when customer sentiment deteriorates, enabling agents to shift approach before situations escalate. In AI quality management, sentiment analysis scores emotional trajectory across the interaction arc — an input to overall quality scoring.

Speech-to-Text (STT)

AI conversion of spoken audio to text transcription — the prerequisite for applying NLP and LLM capabilities to voice interactions. STT quality directly affects all downstream voice AI capabilities: poor transcription degrades intent recognition, sentiment analysis, and response quality. Enterprise-grade STT in NiCE CXone is optimized for contact center audio conditions — background noise, varied accents, telephony compression — producing higher accuracy than general-purpose STT models.

System Prompt

Instructions provided to an LLM that define its behavior, role, scope, and constraints — set by the deploying organization rather than the end customer. In customer service AI, the system prompt defines the AI’s persona, knowledge scope, escalation triggers, response style, prohibited topics, and handling rules. System prompt quality is a critical determinant of AI behavior and consistency. System prompts are configured in CXone AI Studio.

T

Token

The basic unit of text processed by LLMs — roughly equivalent to 3/4 of a word in English. LLM pricing, context window limits, and latency are typically measured in tokens. In customer service AI, token considerations affect: conversation context limits (how much history the AI can consider), response length constraints, and API cost management for high-volume production deployments. Customer service platforms abstract token management from operators, but architects should understand token economics for high-volume use cases.

Transfer Learning

The practice of adapting a model trained on one task or domain to perform well on a different but related task — the foundation of fine-tuning. NiCE CXone uses transfer learning from general language models, further adapted on billions of customer service interactions, producing a model with both broad language capability and deep customer service domain expertise.

U

Unified Desktop

A single agent workspace that consolidates all customer interaction channels, CRM data, knowledge base access, and AI assist capabilities in one interface — eliminating the context-switching between multiple applications that degrades agent efficiency and AHT. CXone Agents provides a unified desktop with embedded CXone Copilot AI assist, reducing the tool fragmentation that is a leading driver of AHT and agent frustration. See Chapter 4.

V

Vector Database

A database optimized for storing and querying vector embeddings — the mathematical representations of text used in RAG systems. Vector databases enable fast semantic similarity search across large knowledge bases, finding the most conceptually relevant documents for a given customer query. Production customer service AI systems require vector databases that combine retrieval speed (sub-100ms) with large-scale knowledge storage. NiCE CXone’s knowledge infrastructure includes purpose-built vector retrieval optimized for customer service knowledge bases.

Virtual Agent

An AI system that conducts customer service interactions autonomously — understanding customer needs, retrieving information, executing transactions, and resolving contacts without human agent involvement. Contemporary usage implies generative AI capability (distinguishing from older scripted chatbots). NiCE CXone Autopilot is NiCE’s virtual agent product. See Chapter 3. Also see: NiCE AI Virtual Agent Platform.

Voice AI

AI applied to voice/telephone customer interactions — combining speech-to-text, natural language understanding, response generation, and text-to-speech to deliver conversational voice self-service. Voice AI replaces or augments traditional IVR, enabling customers to speak naturally rather than navigate touchtone menus. NiCE CXone supports voice AI through CXone Autopilot’s voice channel capability, with CXone providing the NLU layer across both voice and digital. See Chapter 3.

W

Workforce Management (WFM)

The discipline of forecasting contact volume, scheduling agents, and managing real-time adherence to ensure the right staffing at the right time. AI enhances WFM through more accurate volume forecasting (using ML on historical patterns and external signals), automated schedule optimization, and real-time reforecast capabilities. In NiCE CXone, WFM is integrated with AI analytics — enabling the full interaction dataset to inform staffing models with higher accuracy than traditional forecasting methods.

Workflow Automation

The technology-driven execution of defined process sequences without human intervention for each step. In contact centers, workflow automation ranges from simple rule-based routing to complex agentic AI workflows that span multiple systems and handle decision branches. NiCE CXone Orchestrator provides AI-powered workflow automation — executing multi-system service processes with the judgment and adaptability that rule-based RPA automation cannot handle. See Chapter 5.

Z

Zero-Shot Learning

An AI model’s ability to perform a task or recognize a category it has never encountered in training — extrapolating from general language understanding rather than requiring labeled examples for every possible case. Zero-shot capability is significant in customer service AI: a well-trained model can handle novel customer questions outside its explicit training data by reasoning from related knowledge. Contrasted with few-shot learning (a small number of examples provided in the prompt) and fine-tuning (extensive domain-specific training).