AI Glossary

Plain-English definitions of the terms shaping conversational AI, customer experience, and agentic AI in 2026.

A

Agentic AI

AI systems that reason, plan, and take autonomous actions to achieve goals, as opposed to responding only when prompted. Agentic AI agents can execute multi-step workflows, interact with external tools, and make decisions without human approval at each step.

Related:AI AgentAutonomous AI

AI Agent

A software entity powered by one or more AI models that performs tasks on behalf of a user. In customer experience, AI agents resolve queries, update records, process transactions, and escalate to humans when needed.

Related:Agentic AIMulti-Model AI

AI Hallucination

When an AI model generates plausible-sounding but factually incorrect information. Mitigated through retrieval-augmented generation (RAG), knowledge base grounding, and guardrails that restrict responses to verified data sources.

Autonomous AI

AI that operates without continuous human oversight. In customer support, autonomous AI resolves queries end-to-end: understanding the issue, retrieving relevant data, taking action in connected systems, and confirming resolution with the customer.

Related:Agentic AI

C

Chatbot

A software application that simulates conversation with users, typically following scripted decision trees or basic NLP rules. Traditional chatbots lack the reasoning and action-taking capabilities of agentic AI agents.

Containment Rate

The percentage of customer conversations handled entirely by AI without human intervention. Containment measures whether the AI kept the conversation, but not necessarily whether the customer's problem was resolved. Compare with Resolution Rate.

Related:Resolution Rate

Conversational AI

AI technology that enables natural language conversations between humans and machines. Modern conversational AI uses large language models to understand context, intent, and nuance, going beyond keyword matching.

Conversational Commerce

Using AI-powered conversations to drive sales, product discovery, and transactions. AI agents recommend products, answer purchase questions, add items to cart, and guide customers through checkout within the conversation.

Related:AI Agent

CSAT (Customer Satisfaction Score)

A metric measuring how satisfied customers are with a specific interaction. For AI conversations, CSAT should measure whether the problem was resolved, not just whether the bot was polite.

D

Deflection Rate

The percentage of queries redirected away from human agents. Unlike resolution rate, deflection does not measure whether the customer's issue was actually solved. High deflection without high resolution often indicates poor AI performance.

E

Embedding

A numerical representation of text, images, or other data that captures semantic meaning. Used in AI systems for similarity search, retrieval-augmented generation, and knowledge base lookups.

F

Failover

Automatically switching to an alternative AI model or provider when the primary one becomes unavailable or performs poorly. Multi-model platforms use failover to maintain uptime and response quality.

Related:Multi-Model AI

Fine-Tuning

The process of training a pre-existing AI model on domain-specific data to improve performance for a particular use case. For customer support, fine-tuning can improve accuracy on industry-specific terminology and processes.

G

GEO (Generative Engine Optimization)

The practice of optimising content and digital presence to be accurately cited by AI-powered search engines and assistants (ChatGPT, Gemini, Perplexity). GEO extends traditional SEO to the era of AI-generated answers.

Related:LLMRAG

Guardrails

Rules and constraints applied to AI agents to prevent undesirable outputs. Guardrails can restrict topics, enforce brand voice, prevent hallucinations, limit actions the agent can take, and define escalation triggers.

H

Human Handoff

The process of transferring a conversation from an AI agent to a human agent. Effective handoff preserves full conversation context so the customer does not repeat themselves. Also called escalation.

Related:Containment Rate

I

Intent Detection

Identifying what a customer wants to achieve from their message. Modern AI uses large language models for intent understanding rather than keyword matching, enabling it to handle ambiguous or multi-intent queries.

K

Knowledge Base

A structured repository of information that an AI agent uses as its source of truth. Quality knowledge bases are organised around customer intent (not internal documentation structure) and updated frequently.

L

LLM (Large Language Model)

A neural network trained on vast amounts of text data that can understand and generate human language. Examples include GPT, Claude, Gemini, and Mistral. LLMs power modern AI agents' ability to reason and converse naturally.

Related:Multi-Model AI

M

MCP (Model Context Protocol)

An open standard that allows AI agents to connect to external tools and data sources through a universal interface. MCP servers provide a standardised way for AI to interact with CRMs, databases, APIs, and other systems.

Related:AI AgentAgentic AI

Multi-Model AI

An architecture that supports multiple AI models simultaneously, allowing organisations to select the best model for each task, optimise costs by routing simple queries to lighter models, and maintain service through automatic failover.

Related:FailoverLLM

N

NPS (Net Promoter Score)

A metric measuring customer loyalty by asking how likely they are to recommend a company on a 0-10 scale. Promoters (9-10) minus Detractors (0-6) equals the NPS. AI agents can improve NPS through faster, more consistent resolution.

O

Omnichannel

Providing a seamless, consistent customer experience across all communication channels (web, WhatsApp, email, social media, SMS, phone). True omnichannel AI deploys a single agent across all channels with shared context.

P

Prompt Engineering

The practice of designing and refining instructions given to an AI model to achieve desired outputs. In AI agent deployment, prompt engineering defines the agent's behaviour, personality, knowledge scope, and response format.

R

RAG (Retrieval-Augmented Generation)

A technique that enhances AI responses by retrieving relevant information from a knowledge base before generating an answer. RAG reduces hallucinations by grounding responses in verified data rather than relying solely on the model's training.

Related:Knowledge BaseAI Hallucination

Resolution Rate

The percentage of customer queries where the issue was actually solved by the AI, confirmed by the customer or verified through outcome data. Resolution rate is a more meaningful metric than containment rate because it measures real outcomes.

Related:Containment Rate

S

Sentiment Analysis

AI's ability to detect emotional tone in customer messages (frustration, satisfaction, urgency). Real-time sentiment analysis allows AI agents to adjust their approach and escalate frustrated customers to human agents proactively.

SLA (Service Level Agreement)

A contractual commitment defining expected service levels, including uptime guarantees (e.g. 99.9%), response time targets, and remedies if targets are not met. Enterprise AI platforms typically offer SLAs with contractual credits.

T

Token

The basic unit of text that AI models process. A token is roughly 3-4 characters or about three quarters of a word. Token counts affect cost, context window size, and processing speed.

W

White-Label

A product or service that can be rebranded with a customer's own branding. White-label AI agents appear to end users as the brand's own technology, with no visible reference to the underlying platform.

See these concepts in action

Book a demo to see how Certainly puts agentic AI to work for your CX team.

Book a Demo