Glossary
Last updated: March 2026
A practical glossary of the terms you encounter when evaluating, deploying, or operating an AI contact center. Each definition is self-contained and uses everyday language. Cross-references link related concepts so you can jump between them.
A
An AI agent is software that perceives inbound conversations, decides what to do, and acts on behalf of a business across channels. In customer service it understands intent, retrieves knowledge, calls backend systems, and either resolves the request or escalates to a human.
Related: AI Contact Center, HITL
An AI contact center is an inbound customer-service platform where AI agents handle voice, messaging, and email conversations end-to-end. It combines language understanding, retrieval, and orchestration so most contacts are resolved automatically, with human staff focused on exceptions.
Related: Omnichannel, Voice Bot
C
Containment rate is the percentage of inbound conversations a virtual agent fully handles without transferring to a human. High containment means lower cost per contact and faster resolution; it is typically tracked per channel and per intent.
Related: Conversation Resolution Rate, Handoff
Conversation resolution rate measures the share of customer conversations whose stated problem was solved within that conversation. Unlike containment, it requires verification that the customer's goal was actually met, often via end-of-chat surveys or follow-up signals.
Related: Containment Rate, CSAT
Customer Satisfaction Score (CSAT) is the percentage of customers who rate an interaction positively, typically on a 1-5 scale collected immediately after the contact. CSAT measures interaction quality, not loyalty.
Related: NPS
E
Embeddings are numerical vector representations of text that encode meaning. AI customer service uses embeddings to find semantically related content in a knowledge base, so a question phrased in one way retrieves passages phrased differently.
Related: RAG, Vector Search
An entity is a named value extracted from a conversation, such as an order number, product code, location, or date. Entities give the AI structured inputs for backend calls, like looking up a shipment or booking an appointment.
G
The General Data Protection Regulation is the European Union law governing how personal data of EU residents is collected, processed, and stored. AI contact centers must define controller and processor roles, sign a DPA, and document data residency and sub-processors.
H
Handoff is the transfer of a live conversation from an AI agent to a human, with full context attached. A clean handoff includes the transcript, detected intent, customer identity, and any actions already taken.
Related: HITL, Containment Rate
Human-in-the-loop (HITL) is a workflow where humans review, approve, or override AI decisions before they reach the customer or affect a system. In customer service it covers escalation to live agents and supervised training of the AI.
I
An intent is the goal a customer is trying to achieve in a conversation, such as 'cancel order', 'book appointment', or 'reset password'. Intent classification lets the AI route to the right workflow or knowledge.
Interactive Voice Response is the touch-tone or speech menu that routes inbound phone calls. Modern AI voice bots replace traditional IVR trees with natural-language conversations that resolve the caller's request directly.
K
A knowledge base is the structured store of documents, articles, and product data that an AI customer service agent retrieves answers from. Indexing quality and freshness directly determine answer accuracy.
Related: RAG, Vector Search
L
N
Natural Language Understanding is the component that turns a customer's free-form message into structured meaning: intent plus entities. NLU sits between raw input and the action the AI takes.
Net Promoter Score asks customers how likely they are to recommend a company on a 0-10 scale, then subtracts detractors from promoters. NPS measures loyalty across the relationship, not individual interactions.
Related: CSAT
O
Omnichannel customer service handles inbound contacts across voice, messaging, email, and webchat with shared context, history, and policies. A true omnichannel AI uses one brain across channels rather than separate bots per channel.
Related: AI Contact Center, Voice Bot
R
Retrieval-Augmented Generation grounds an LLM's answers in retrieved source documents instead of relying only on training data. RAG reduces hallucination and lets answers update the moment the underlying knowledge changes.
Related: LLM, Embeddings, Vector Search
S
Sentiment analysis detects the emotional tone of a customer message, typically positive, negative, or neutral, sometimes with intensity. AI contact centers use it to prioritize escalations and to score interaction quality.
Related: CSAT
Session Initiation Protocol is the standard signaling protocol used to set up, manage, and tear down voice calls over IP. AI voice bots connect to traditional phone numbers through SIP trunks provided by telecom operators.
Speech-to-Text converts spoken audio into written transcript. It is the first step in an AI voice conversation; quality and language coverage of STT directly limit how well the rest of the pipeline understands the caller.
T
V
Vector search retrieves content by semantic similarity rather than keyword match, by comparing embedding vectors of the query and stored documents. It is the lookup mechanism that powers RAG-based AI customer service.
Related: Embeddings, RAG
A voice bot is an AI agent that handles phone calls end-to-end using speech recognition, language understanding, and speech synthesis. Inbound voice bots answer customer calls without an IVR menu and can perform backend actions during the call.
W
The WhatsApp Business API is Meta's programmatic interface for sending and receiving WhatsApp messages at scale on a verified business account. AI contact centers use it to deliver inbound customer support through the channel most customers already use.
Related: Omnichannel
Telecom operators, PBX vendors, software vendors and resellers can white-label, embed or resell it.