AI Service Desk Glossary
32+ Terms of Art for Enterprise Buyers
Agentic AI
AI that can take multi-step actions across tools and systems, not just answer questions. An agentic AI ITSM system can look up a ticket, open the relevant system, execute a fix (password reset via Okta, access provisioning via Entra ID), post an update to the user, and close the ticket. This is distinct from a chatbot (scripted menus) or a RAG-answering agent (answers only). ServiceNow Now Assist, Moveworks, and Aisera are most positioned at the agentic end of the spectrum in April 2026.
Auto-resolution
A ticket or query fully resolved by the AI without any human agent involvement. Strict definition: AI provides a complete resolution, no human agent touches the ticket, and the user does not escalate or re-open within 72 hours. Zendesk uses this term specifically for its outcome-based pricing model: $1.50 per committed automated resolution. Not all vendors define it the same way; compare against the strict definition when evaluating claims.
Automated resolution (Zendesk)
Zendesk's specific pricing unit: a ticket or conversation fully resolved by the Zendesk AI agent without human agent involvement. Priced at $1.50/resolution (committed) or $2.00/resolution (PAYG). Includes AI answering via email, messaging, or chat with no human escalation; AI surfacing a KB article that resolves the query; AI executing a defined procedure and closing the ticket.
CAB (Change Advisory Board)
A committee responsible for reviewing and approving proposed changes to IT infrastructure and services. Required by ITIL change management for high-risk and significant changes. AI-assisted change risk scoring (ServiceNow Change Risk Prediction, Atlassian equivalent) provides risk scores to inform CAB decisions but does not replace human sign-off. CAB oversight is required for compliance (SOC 2, HIPAA, FedRAMP) regardless of AI deployment.
Change risk prediction
ML-based scoring of incoming change requests for blast radius, implementation complexity, and historical recurrence risk. ServiceNow ships Change Risk Prediction generally available in Pro Plus and above. Atlassian has comparable capabilities emerging in JSM Premium. Risk scores inform CAB decisions; they do not authorise changes autonomously.
Cost per ticket
The fully-loaded cost to resolve a single IT support ticket. Includes agent salary and benefits, tooling overhead, management allocation, and overhead. HDI/MetricNet benchmarks put cost-per-ticket at $6-$40+ in North American enterprises, with an internal IT median of approximately $22. This is the primary economic lever for AI service desk ROI calculation.
Deflection rate
The percentage of tickets or queries resolved by the AI without human agent involvement. Not consistently defined across vendors. Some count any auto-reply; others count only full resolution without human touch. The strictest credible definition: AI-only resolution with no human escalation within 72 hours. Gartner benchmarks: 20-30% industry average, 40-60% best-in-class. Vendor-published figures (50-75%) use looser definitions. See the deflection rates page for the full reconciled benchmark.
Escalation rate
The percentage of tickets that the AI cannot resolve and routes to a human agent. Deflection rate + escalation rate = 100% (all tickets go one way or the other). High escalation rate means the AI is routing correctly rather than guessing; inappropriately low escalation rate may mean the AI is resolving tickets it should not be (hallucinating answers or closing tickets without confirmation).
First Contact Resolution (FCR)
The percentage of tickets resolved on the first contact, without requiring a return interaction. AI service desks improve FCR by ensuring agents arrive at tickets with pre-populated context, suggested actions, and KB article references. FCR is a distinct metric from deflection; a deflected ticket is resolved by AI on first contact; a routed ticket may or may not be resolved by the agent on first contact.
Freddy AI (Freshservice)
Freshworks' AI platform for Freshservice. Comprises three products: Freddy AI Copilot (agent-facing, ticket summarisation, suggested responses, $29/agent/month add-on on Pro+), Freddy AI Agent (end-user-facing autonomous self-service, Enterprise tier only, 1,200 sessions/user/year cap), and Freddy AI Insights (analytics, Enterprise only). Freddy AI Agent is the product relevant to AI service desk deflection.
Hallucination
AI-generated content that is factually incorrect, plausibly stated. In AI service desk, hallucination occurs when the LLM generates an answer not grounded in the knowledge base, leading to confident wrong guidance (e.g., incorrect password reset procedures, wrong access provisioning steps). RAG reduces hallucination rates 42-68% compared to pure-LLM baselines when source documents are accurate and current. KB hygiene is the primary control.
Intent classification
The process of identifying what a user's ticket or query actually means, independent of how they phrased it. An intent classifier maps natural language to a predefined intent category (e.g., 'I can't log in' maps to 'password reset' intent). Accuracy ranges 70-95% depending on intent library maturity and training data volume. A well-configured intent library with 12+ months of ticket history typically achieves 85-95% accuracy.
Intent library
The collection of defined intent categories and example phrases that train the AI service desk's intent classifier. A well-built intent library covers 50-200 distinct ticket types with 20-50 example phrases each. Intent library maturity is a primary driver of deflection rate: sparse libraries produce 70% classification accuracy; mature libraries with 12 months of ticket history produce 85-95%. Building the intent library is the primary implementation effort for most AI ITSM deployments.
ITIL
IT Infrastructure Library. The dominant framework for enterprise IT service management, maintained by Axelos. ITIL 4 (current) organises IT services into practices including incident management, problem management, change enablement, service request management, and release management. ITIL compliance is a procurement requirement for most large enterprises and regulated industries. AI ITSM platforms augment ITIL practice execution; they do not replace the ITIL framework.
ITSM
IT Service Management. The practice of managing IT services to meet business needs, aligned to ITIL or similar frameworks. ITSM platforms (ServiceNow, Jira Service Management, Freshservice, Zendesk, etc.) provide the ticketing, workflow, and reporting infrastructure. AI ITSM is the category of tools that add AI capabilities (classification, KB answering, auto-resolution, agent assist) to existing ITSM platforms.
Knowledge base (KB)
The collection of articles, runbooks, FAQs, and documented procedures that the AI service desk draws from via RAG. KB quality is the single largest determinant of AI ITSM deflection rate. Stale, contradictory, or incomplete KB articles produce hallucinated answers at scale. KB hygiene (regular review, gap identification, stale article removal) is a continuous operational requirement, not a one-time setup task.
L1 / L2 / L3 support
Tiers of IT support by complexity. L1: first-line, high-volume, standard requests (password resets, access requests, status queries, basic troubleshooting). L2: second-line, requires technical expertise (application issues, network problems, complex configuration). L3: specialist, rare and complex (root-cause analysis, infrastructure changes, security incidents). AI service desks target L1 automation. L2 and L3 remain human-agent territory in most deployments through 2026.
LLM (Large Language Model)
The AI foundation model that powers natural language understanding, generation, and reasoning in AI service desk platforms. Vendors use GPT-4-class models (OpenAI, Azure OpenAI), Claude (Anthropic), or their own domain-specific fine-tuned models. The choice of base LLM affects response quality and hallucination rate; vendor-specific fine-tuning on IT and HR support data (Aisera's UniversalGPT) improves accuracy on domain-specific requests compared to general-purpose models.
Mean time to resolution (MTTR)
Average time from ticket creation to ticket resolution. A primary ITSM KPI. AI service desks reduce MTTR by: (a) deflecting L1 tickets in seconds instead of minutes; (b) pre-populating L2 tickets with context, reducing agent time-to-resolution by 20-40%; (c) enabling 24/7 AI availability where human agents are off-shift. MTTR reduction is a secondary ROI driver after deflection rate.
NLU (Natural Language Understanding)
The AI capability to understand meaning, intent, and context from unstructured text. NLU is the foundation of intent classification in AI service desks. Modern NLU combines transformer-based language models (the LLM) with intent classification training data to map free-form ticket language to structured intent categories. NLU accuracy is context-dependent: IT-domain NLU trained on IT support tickets outperforms general-purpose NLU on the same domain.
Now Assist (ServiceNow)
ServiceNow's AI platform layered on top of the ServiceNow ITSM suite. Available on Pro Plus and Enterprise tiers. Includes: auto-summarise incident and case records, draft resolution notes, Now Assist Virtual Agent (end-user self-service), agent-assist sidebar, change-risk prediction, and flow generation. Structured in Foundation/Advanced/Prime capability tiers. Pricing: 25-60% uplift on Pro Plus seats ($150-$200/user/month base). See /vs-servicenow for full analysis.
RAG (Retrieval-Augmented Generation)
A technique for grounding LLM responses in specific documents rather than the model's training data alone. In AI service desk, RAG retrieves the most relevant KB articles, past tickets, and runbook sections and includes them in the LLM's context before generating a response. RAG reduces hallucination rates 42-68% compared to pure-LLM baselines when source documents are accurate and current. KB hygiene is the primary control for RAG accuracy.
Rovo (Atlassian)
Atlassian's AI platform available on Jira, Confluence, and JSM at Premium and Enterprise tiers. Comprises Rovo Search (AI-powered search across Atlassian and connected tools), Rovo Chat (conversational AI interface), and Rovo Agents (custom task-specific automation agents). Available at JSM Premium ($51.42/agent/month) without additional cost. Virtual Service Agent is a separate JSM-specific product for end-user self-service.
Service desk
The single point of contact between IT and users for all IT service requests, incidents, and queries. The service desk function is defined in ITIL and serves as the intake layer for all IT support. AI service desk platforms automate the service desk function, handling L1 requests autonomously and routing L2/L3 requests to human agents with pre-populated context.
SLA (Service-Level Agreement)
A contractual commitment between IT and the business on service quality. Common ITSM SLA metrics: first response time, resolution time, and availability. AI service desks improve SLA compliance by reducing resolution time for L1 tickets (auto-resolved in seconds) and by pre-populating L2 tickets so agents respond faster. SLA breach notifications and escalation paths remain managed by the ITSM platform, not the AI.
Session (Freshservice metric)
Freshservice's unit for metering Freddy AI Agent usage. A session is any Freddy AI Agent interaction within a 24-hour window. Multiple interactions by the same user on the same day count as one session. The Freddy AI Agent session cap is 1,200 sessions per user per year (included in Enterprise plan). At 260 working days per year, this allows approximately 4.6 sessions per working day per user.
Shift-left
The ITSM strategy of resolving tickets at the earliest, lowest-cost tier possible. Traditional ITSM shift-left uses KB articles and self-service portals to reduce L1 call volume. AI ITSM shift-left extends this by automating L1 resolution entirely, reducing L2 handle time via agent assist, and predictively identifying problem patterns before they generate ticket volume. AI is the current state-of-the-art in shift-left strategy.
SSO / SCIM
Single Sign-On (SSO) and System for Cross-domain Identity Management (SCIM). SSO allows users to authenticate once across multiple applications. SCIM is the protocol for automated provisioning and deprovisioning of user accounts. AI service desk password reset and access provisioning automation requires SSO and SCIM integration with your identity provider (Okta, Entra ID, Ping). Both are table stakes in enterprise AI ITSM deployments.
Ticket lifecycle
The end-to-end flow of an IT support ticket from creation to closure. Stages: (1) intake/creation across any channel; (2) classification and routing (AI or human); (3) triage and initial response; (4) investigation and resolution; (5) closure and knowledge capture. AI service desks automate stages 1-4 for L1 ticket types; stages 3-5 for L2 benefit from AI assist without full automation.
UniversalGPT (Aisera)
Aisera's domain-specific large language model, fine-tuned on IT and HR support patterns. Unlike general-purpose models, UniversalGPT is trained specifically on enterprise IT ticket corpora, which Aisera claims produces higher intent classification accuracy on IT-specific language and compound requests. Domain-specific fine-tuning reduces hallucination on uncommon IT scenarios compared to general-purpose models with system prompting.
Virtual agent / Virtual Service Agent
A conversational AI endpoint for end-user self-service. In the AI ITSM context: Atlassian Virtual Service Agent (JSM), Freshservice Freddy AI Agent, ServiceNow Virtual Agent, Moveworks employee experience platform. Each provides a chat interface where users submit requests and the AI attempts to resolve them without human agent involvement. The quality of the virtual agent experience is determined by the intent library, KB quality, and action framework depth.
Workflow automation
Rules-based or AI-driven automation of IT workflows, such as provisioning, ticket routing, approval flows, and notifications. AI service desk platforms integrate with workflow automation tools (Freshservice Workflow Automator, ServiceNow Flow Designer, Atlassian automation rules) to execute actions beyond answering: provisioning access, creating cross-ITSM tickets, updating CMDB records, triggering runbooks.