The MENA AI & CX Glossary.

A software system that uses a large language model to understand natural language inputs, reason about them, and take actions by calling external tools or APIs - going beyond simple Q&A to execute real-world tasks autonomously within defined boundaries.

In Eshal: Every Eshal deployment runs as an AI agent with live integrations to business systems, executing workflows like order tracking, return processing, and KYC collection rather than just retrieving information.

A customer-facing AI agent configured for service and sales workflows - handling the full lifecycle of a customer interaction from intent detection through action execution to resolution, across multiple channels simultaneously.

In Eshal: The Eshal product is an AI concierge - not a chatbot. It executes workflows (processes returns, books appointments, initiates KYC) rather than merely providing information.

AI systems that operate with a degree of autonomy - planning multi-step sequences of actions, using tools and external data, and adapting based on results - rather than responding to single prompts in isolation.

In Eshal: Eshal uses agentic architecture with guardrails. The AI plans and executes multi-step workflows within pre-defined action boundaries set by Dynamic Action Gating.

Natural language processing systems trained specifically on Arabic text, accounting for its root-based morphology, right-to-left script, and dialectal fragmentation across Gulf, Levantine, Egyptian, and Maghrebi variants. Cannot be reliably replaced by translating Arabic to English before processing.

In Eshal: Eshal's NLP processes Gulf Arabic, Levantine Arabic, and Egyptian Arabic natively - including code-switched Arabic/English - reducing accuracy loss by up to 24 percentage points vs MSA-only models.

The percentage of customer contacts handled end-to-end by an AI system without any human agent involvement. A key performance metric, typically ranging from 40–50% for basic chatbots to 80–90% for agentic AI with live system integrations.

In Eshal: Eshal's average automated resolution rate across deployed customers is 86%, ranging from 78% (complex healthcare bookings) to 93% (logistics tracking).

A sales qualification framework assessing Budget, Authority, Need, and Timeline. When applied by AI, the system collects these data points through natural dialogue and scores or routes leads before human sales engagement.

In Eshal: Eshal's sales module conducts BANT qualification through Arabic and English conversation, scoring leads and routing to sales reps with qualification summaries pre-populated in CRM.

AI systems that engage in natural language dialogue with humans - understanding intent, maintaining context across multiple turns, and generating coherent responses. Encompasses FAQ bots and complex agentic systems that take real-world actions.

A bidirectional connection between an AI system and a Customer Relationship Management platform, allowing the AI to read customer history and account status - and write new contact records and cases - in real time during interactions.

In Eshal: Integrates with Salesforce, HubSpot, and custom CRMs. Customer context is read at conversation start; resolved summaries are written back automatically.

A post-interaction metric collected by asking customers to rate their experience, typically 1–5 or 1–10. In AI deployments, CSAT is collected automatically at conversation close and tracked separately for AI-resolved vs human-resolved contacts.

A metric measuring how much effort a customer expended to resolve their issue. Research consistently shows effort reduction correlates more strongly with loyalty and churn prevention than satisfaction alone.

The principle under which data is subject to the laws and governance of the country where it is physically stored and processed. In the UAE context, the PDPL governs data sovereignty for personal data - requiring cross-border transfers to meet UAE-level adequacy standards.

In Eshal: All UAE deployments process and store data on OVHcloud Dubai infrastructure. A PDPL-compliant Data Processing Agreement (DPA) is provided to every UAE customer.

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

A framework classifying AI-executable actions into: Fully Automated, Human Approval Required, and Never Permitted. Allows organisations to give AI operational capability while maintaining oversight for consequential actions.

In Eshal: Every deployed workflow is classified at configuration time. Standard returns may be Fully Automated; large refunds are Human Approval Required; account closure is Never Permitted.

Transferring a conversation from AI to a human agent, triggered by pre-defined conditions such as customer sentiment, query complexity, or action type requiring approval. Effective escalation includes full conversation context so agents never ask customers to repeat themselves.

In Eshal: Agents receive the full transcript, AI's issue classification, and pre-collected data - reducing agent handle time on escalated contacts by an average of 38%.

The percentage of customer contacts resolved in the first interaction without a follow-up. A primary efficiency metric directly correlated with satisfaction and cost per contact.

The Arabic dialect spoken across UAE, Saudi Arabia, Kuwait, Qatar, Bahrain, and Oman. Characterised by distinctive phonology, heavy English code-switching in commercial contexts, and Urdu/Hindi loanwords particularly in UAE. The primary dialect for Eshal's core deployment markets.

In Eshal: Gulf Arabic is the primary training dialect. The system handles UAE, Saudi, Kuwaiti, and Qatari variations - including Arabizi (romanised Arabic) and mixed Arabic/English messages.

An AI architecture where human oversight or approval is built into specific workflow stages - for high-risk, high-value, or regulatory-sensitive actions. HITL makes agentic AI viable in regulated industries like banking, healthcare, and government.

The NLP process of classifying what a customer wants from their message, regardless of phrasing. "وين طلبيتي؟" (Gulf Arabic), "أين طلبي" (MSA), and "where is my order?" all carry the same intent - order tracking - which the AI maps to the correct workflow.

The international standard for information security management systems. ISO 27001 certification demonstrates that an organisation has systematically identified and treated information security risks to an independently audited standard. Required by banking, government, and healthcare procurement in MENA.

In Eshal: Eshal is ISO 27001 certified. Customers may request the certificate and last audit summary for vendor qualification.

Using AI to guide customers through Know Your Customer identity verification - collecting identity documents, verifying details, and triggering compliance reviews - without manual agent involvement in data-collection steps. Must operate within SAMA, CBUAE, and DFSA frameworks in MENA.

In Eshal: Eshal's banking KYC flow reduces onboarding time from 45 minutes to 3 minutes. Account activation remains a human-approval step to satisfy CBUAE requirements.

A deep learning model trained on large text corpora to understand and generate natural language. LLMs form the foundation of modern conversational AI. Choice of base LLM, fine-tuning approach, and retrieval strategy significantly affects Arabic dialect performance.

The Arabic dialect spoken across Lebanon, Syria, Jordan, and Palestine. Heavily influenced by French (particularly Lebanese Arabic) and distinct from Gulf Arabic in vocabulary, phonology, and grammar. A significant customer dialect in UAE due to large Lebanese and Jordanian expatriate communities.

Middle East and North Africa - a geopolitical and commercial region from Morocco to Iran. In AI customer service contexts, "MENA" typically refers to the GCC states (UAE, Saudi Arabia, Kuwait, Qatar, Bahrain, Oman) plus Egypt, Jordan, Lebanon, and Iraq.

An AI deployment that operates consistently across multiple customer touchpoints - WhatsApp, web chat, email, voice - from a single configuration. True multi-channel AI maintains conversation context across channel switches.

In Eshal: A single Eshal deployment serves WhatsApp, web chat, and API channels simultaneously with shared configuration - no channel-specific re-configuration required.

A field of AI enabling computers to understand, interpret, and generate human language. In customer service, NLP handles intent detection, entity extraction (order numbers, dates, product names), sentiment analysis, and response generation.

A CX approach where all channels share a unified customer context and conversation history - so a customer starting on WhatsApp can continue on web chat without repeating themselves. Distinct from multi-channel, which simply means operating on multiple channels without shared context.

An AI deployment where all models, data, and processing run on infrastructure owned or controlled by the customer - not shared cloud. Required by UAE and Saudi government entities, classified-data environments, and certain banking customers.

In Eshal: Supports on-premise deployment with the same AI capabilities running on customer-controlled infrastructure, no external API dependencies for conversation processing.

An architecture where an LLM's responses are grounded in information retrieved from an external knowledge base at query time - rather than relying solely on training data. Allows AI to answer questions about proprietary business data without retraining the model.

In Eshal: Eshal uses RAG for product catalogues, return policies, and FAQ content - combined with live API calls for real-time operational data.

The percentage of customer contacts in which the customer's need was fully met within a single interaction. Tracked separately for AI-resolved and human-resolved contacts to measure AI performance independently. See also: Automated Resolution Rate.

NLP analysis of text to determine emotional tone - positive, neutral, or negative. Used in customer service AI to detect frustrated customers and trigger escalation. Arabic sentiment analysis requires dialect-specific training; Gulf Arabic colloquialisms differ significantly from MSA negative phrasing.

A contractual commitment defining minimum performance standards - uptime, response time, and support availability. For AI customer service, SLAs specify: system availability (typically 99.9%+ uptime), maximum response latency, and support resolution timeframes.

In Eshal: Enterprise tier includes 99.95% uptime SLA with sub-4-second AI response time commitments.

Small and medium-sized business - in the UAE, typically businesses with under AED 250 million annual revenue or under 200 employees. The UAE SMB sector accounts for over 90% of businesses in the country and is Eshal's primary Starter and Growth tier market.

Federal Decree-Law No. 45 of 2021 on the Protection of Personal Data - the UAE's national data protection law, effective January 2022. Governs the collection, processing, storage, and transfer of personal data. Requires data processing agreements with third-party processors and restricts cross-border transfers without adequacy safeguards.

In Eshal: Eshal provides a PDPL-compliant DPA to all UAE customers. Data is processed on OVHcloud Dubai to satisfy residency requirements. See Security & Compliance for details.

AI systems that process spoken language - speech to text, intent understanding, response generation, and text to speech. In MENA, Voice AI must handle Arabic dialect speech recognition and WhatsApp voice notes, which are a common customer input format in the Gulf.

Meta's official API for business messaging on WhatsApp, enabling verified businesses to send and receive messages programmatically at scale. The only compliant method for deploying AI automation on WhatsApp - unofficial tools violate Terms of Service and risk account termination. Supports templated proactive messages and free-form conversational replies within 24-hour windows.

In Eshal: All Eshal WhatsApp deployments use the official Business API. Eshal manages business verification and green-tick verification as part of onboarding.

Using 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. What distinguishes an AI concierge from a chatbot.

The capability of an ML model to correctly handle tasks it was not explicitly trained on - 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.