The AI & CX Glossary.

An AI agent is a software system that uses a large language model to understand natural language, reason about intent, and take action by calling external tools or APIs. Unlike simple chatbots that only answer questions, it executes real-world tasks autonomously within boundaries defined by its designers.

An AI concierge is a customer-facing AI agent configured for service and sales, handling a full interaction from intent detection through action execution to resolution. Unlike a chatbot, it works across multiple channels simultaneously and completes tasks, such as booking, processing, or verifying, rather than only providing information.

An AI SDR is an AI agent that automates part or all of the Sales Development Representative workflow - prospecting, qualification, and meeting booking - handling BANT discovery, conversation, and CRM updates directly, so qualified opportunities reach an account executive without manual screening.

Agentic AI describes AI systems that operate with a degree of autonomy - planning multi-step sequences of actions, calling external tools and data, and adapting their next step based on results - rather than responding to a single prompt in isolation. It underlies AI agents and AI concierges that complete tasks end-to-end.

Arabic NLP refers to natural language processing systems trained specifically on Arabic text, built to handle its root-based morphology, right-to-left script, and dialectal fragmentation across Gulf, Levantine, Egyptian, and Maghrebi variants. It cannot be reliably replaced by translating Arabic into English before processing, since translation loses meaning and dialect nuance.

Automated Resolution Rate is the percentage of customer contacts an AI system resolves end-to-end without any human agent involvement. It is a key performance metric for conversational AI, typically ranging from 40-50% for basic chatbots up to 80-90% for agentic AI systems with live integrations into business systems like CRM.

BANT Qualification is a sales framework that assesses a prospect's Budget, Authority, Need, and Timeline. When applied by AI, the system gathers these four data points through natural conversation, then scores or routes the lead to the appropriate sales representative before any human sales engagement takes place.

Conversational AI refers to AI systems that engage in natural language dialogue with people, understanding intent, maintaining context across multiple turns, and generating coherent responses. The category spans simple FAQ bots that only answer questions to complex agentic systems, like AI concierges, that take real-world actions such as processing a return.

CRM Integration is a bidirectional connection between an AI system and a Customer Relationship Management platform. It lets the AI read a customer's history and account status during a live interaction, and write new contact records, cases, or notes back into the CRM automatically, without manual data entry.

CSAT, or Customer Satisfaction Score, is a post-interaction metric gathered by asking customers to rate their experience, typically on a 1-5 or 1-10 scale. In AI deployments, CSAT is collected automatically right after a conversation closes and tracked separately for AI-resolved versus human-resolved contacts.

Customer Effort Score (CES) is a metric measuring how much effort a customer had to expend to get their issue resolved, usually gathered through a short post-interaction survey. Research consistently shows reducing customer effort correlates more strongly with loyalty and churn prevention than satisfaction scores alone.

Data Sovereignty is the principle that data is subject to the laws and governance of the country where it is physically stored and processed. In the UAE, the PDPL governs data sovereignty for personal data, requiring cross-border transfers to meet UAE-level standards of protection before that data can leave the country.

Dialect NLP (Arabic) refers to natural language processing models trained specifically on Gulf (Khaleeji), Levantine (Shami), and Egyptian (Masri) dialect variants, rather than only Modern Standard Arabic. It is essential for business deployments, since accuracy gaps of up to 24 percentage points exist between MSA-only and dialect-native models.

Dynamic Action Gating is a framework that classifies every AI-executable action into one of three tiers: Fully Automated, Human Approval Required, or Never Permitted. It lets organizations grant AI systems real operational capability for routine tasks while preserving human oversight and control over consequential, high-risk, or irreversible actions.

Escalation Routing is the process of transferring a conversation from AI to a human agent when pre-defined conditions are met, such as negative sentiment, high query complexity, or an action requiring approval. Effective escalation routing carries the full conversation context to the human agent, so the customer never has to repeat information already given.

First Contact Resolution, or FCR, is the percentage of customer contacts fully resolved during the first interaction, with no follow-up contact required. It is a core customer service efficiency metric, directly correlated with customer satisfaction and inversely correlated with cost per contact, making it a standard benchmark for evaluating both human agents and AI-driven support.

Gulf Arabic, also called Khaleeji, is the Arabic dialect spoken across the UAE, Saudi Arabia, Kuwait, Qatar, Bahrain, and Oman. It is marked by distinctive phonology, frequent code-switching with English in business and commercial settings, and loanwords from Urdu and Hindi, particularly in the UAE, reflecting the region's diverse expatriate workforce.

Human-in-the-Loop, or HITL, describes an AI architecture where human oversight or approval is deliberately built into specific stages of an automated workflow, rather than removed entirely. It is applied to high-risk, high-value, or regulatory-sensitive actions, and is what makes agentic AI operationally viable in industries such as banking, healthcare, and government.

Intent detection is the natural language processing task of identifying what a customer wants from a message, regardless of wording, dialect, or language. A Gulf Arabic phrase, a Modern Standard Arabic phrase, and an English sentence asking the same underlying question should all be classified as the same intent and routed to the same workflow.

ISO 27001 is the international standard for information security management systems. Certification demonstrates that an organization has systematically identified, assessed, and treated its information security risks according to a rigorous framework, independently verified through external audit. It is commonly required in banking, government, and healthcare procurement processes worldwide as proof of vendor security maturity.

KYC automation is the use of AI to guide customers through Know Your Customer identity verification - collecting identity documents, checking submitted details, and triggering compliance review steps - without a human agent manually handling data collection. It must operate within regional regulatory frameworks such as SAMA, CBUAE, and DFSA.

A large language model, or LLM, is a deep learning model trained on vast amounts of text to understand and generate natural language. LLMs form the foundational technology behind modern conversational AI. The choice of base model, fine-tuning approach, and retrieval strategy all significantly affect how well it performs in Arabic dialects.

Levantine Arabic, also known as Shami, is the Arabic dialect spoken across Lebanon, Syria, Jordan, and Palestine. It is heavily influenced by French, particularly in Lebanese Arabic, and is distinct from Gulf Arabic in vocabulary, phonology, and grammar. It is a significant customer dialect in the UAE due to large Lebanese and Jordanian expatriate communities.

Multi-channel AI is an AI deployment that operates consistently across multiple customer touchpoints - such as WhatsApp, web chat, email, and voice - from a single underlying configuration rather than separate systems per channel. True multi-channel AI preserves full conversation context when a customer switches from one channel to another mid-interaction.

Natural language processing, or NLP, is the field of artificial intelligence that enables computers to understand, interpret, and generate human language. In customer service applications, NLP powers tasks such as intent detection, entity extraction like order numbers and dates, sentiment analysis, and the generation of natural-sounding responses to customer messages.

Omnichannel is a customer experience approach in which every channel a business uses shares one unified customer context and conversation history, so a customer can start a conversation on WhatsApp and continue it on web chat without repeating themselves. It differs from multi-channel, which lacks that shared context.

On-premise AI is a deployment model in which all AI models, customer data, and processing run entirely on infrastructure owned or controlled by the customer, rather than a shared external cloud. It is often mandated for UAE and Saudi government entities, classified-data environments, and certain regulated banking customers with strict data residency requirements.

Retrieval-Augmented Generation (RAG) is an AI architecture where a large language model's answers are grounded in facts pulled from an external knowledge base at the moment a question is asked, rather than relying only on what the model learned during training. This lets AI answer accurately about proprietary business information without retraining the underlying model.

Resolution rate is the percentage of customer contacts in which the customer's need was fully addressed within a single interaction, without follow-up contact. It is tracked separately for AI-handled and human-handled conversations so businesses can measure automated performance independently, and it is closely related to, but distinct from, automated resolution rate.

Sentiment analysis is a natural language processing technique that examines text to determine its emotional tone - positive, neutral, or negative. In customer service AI, it detects frustrated or upset customers in real time and triggers escalation to a human agent, with Arabic requiring dialect-specific training beyond standard Modern Standard Arabic models.

A Service Level Agreement (SLA) is a contractual commitment defining minimum performance standards a provider must meet, such as system uptime, response latency, and support availability. For AI customer service platforms, SLAs typically specify uptime of 99.9% or higher, maximum response times, and defined timeframes for resolving support issues raised by the customer.

SMB stands for small and medium-sized business. In the UAE, this typically means a company with under AED 250 million in annual revenue or fewer than 200 employees. The UAE SMB sector accounts for over 90% of businesses in the country and is a major segment of the economy.

UAE PDPL is Federal Decree-Law No. 45 of 2021 on the Protection of Personal Data, the UAE's federal data protection law, in effect since January 2022. It governs how personal data is collected, processed, stored, and transferred, requiring data processing agreements with third-party processors and restricting cross-border transfers without adequate safeguards.

Voice AI refers to systems that process spoken language end to end: converting speech to text, understanding the caller's intent, generating an appropriate response, and converting that response back into speech. In Gulf markets, voice AI must also reliably handle Arabic dialect speech and WhatsApp voice notes, a widely used customer input format.

WhatsApp Business API is Meta's official interface for programmatic business messaging on WhatsApp, letting verified businesses send and receive messages at scale. It is the only compliant way to run AI automation on WhatsApp, since unofficial tools breach Meta's Terms of Service and risk account termination, and it supports templated proactive messages and free-form replies.

Workflow automation is the use of AI to execute multi-step business processes - data lookups, decisions, API calls, and confirmations - in response to customer intents, without manual human initiation of each step. This end-to-end execution is what distinguishes an AI concierge from a chatbot that only answers questions.

Zero-shot learning is the capability of a machine learning model to correctly handle tasks it was not explicitly trained on, by generalising from related training data. In NLP, zero-shot classification allows AI to identify new intent categories without retraining, by reasoning about semantic relationships between the query and intent labels.