Glossary

Agentic workflow automation#

Agentic workflow automation is the use of AI agents to execute complete business processes end to end: interpreting a request, applying policy, verifying identity, taking the required actions across systems, and updating records, rather than answering questions and leaving the follow-through to a human. It differs from chatbots, which retrieve answers, and from traditional workflow automation, which follows fixed rules. Agents handle variation, make policy-bounded decisions, and complete the work.

At Cascade: Agents for IT and HR shared services. See the platform.

AI agent#

An AI agent is software that pursues a goal autonomously: it interprets a request, plans the steps, uses tools and systems to act, and verifies the outcome, rather than producing a single response. In employee support, an agent doesn’t just tell an employee the equipment-refresh policy. It checks eligibility, opens the order, routes the approval, and updates the record.

Agentic AI#

Agentic AI is the class of artificial intelligence built to take actions toward goals (deciding, using tools, and executing multi-step work), as distinct from generative AI used purely to produce content or answers. Most enterprise “AI assistants” are generative. Agentic systems complete transactions.

Chatbot (vs. AI agent)#

A chatbot is a conversational interface that retrieves and delivers answers, typically from a knowledge base or script. It can tell an employee what the policy says, but it cannot execute the request. The practical test: after the conversation, does a human still have to open the ticket, chase the approval, and do the work? If so, it’s a chatbot.

Copilot#

A copilot is an AI assistant that helps a person do their own work faster (drafting, summarizing, suggesting), with the person still executing. Copilots augment individual productivity; agents absorb the work itself. An IT copilot helps the technician resolve a ticket. An IT agent resolves the ticket.

Virtual agent#

A virtual agent is a conversational AI interface that handles employee or customer requests through chat. It is the term many established service platforms use for their conversation layer. Virtual agents answer, route, and open tickets. Whether the work behind the request gets executed autonomously is what separates a virtual agent from an agentic platform.

Conversational AI#

Conversational AI is the technology that lets software understand and respond to natural language in chat and voice. It is the interface layer of employee support, and it is distinct from execution: a system can be excellent at conversation and still leave every request for a person to fulfill.

AI assistant (enterprise AI assistant)#

An enterprise AI assistant is a conversational AI that employees use for questions and tasks across the organization. Assistants center on the conversation, while agents center on the completed work. In practice the terms blur, which is why the meaningful question is what percentage of requests end fully resolved.

Intent (intent library)#

An intent is a predefined category of request, such as “reset password” or “check PTO balance,” that classic conversational AI systems match what a user types against, often drawing on libraries of thousands of pre-built intents. Intent-based bots break on requests they weren’t configured for. Systems built on large language models can interpret new requests without a predefined library, which is a key dividing line when evaluating platforms.

RPA (robotic process automation)#

RPA is software that automates tasks by mimicking human clicks and keystrokes in application interfaces, following fixed scripts. It is the predecessor to agentic automation: fast for stable, rule-bound tasks, but brittle when screens or cases change. Agents differ by working through APIs and policy, interpreting the request and handling variation rather than replaying a recording.

No-code and low-code#

No-code and low-code describe platforms where workflows and agents are configured through visual interfaces rather than programming. The distinction determines whether your operations team can own and change automations, or whether every change has to go through engineering, yours or the vendor’s.

AI operating layer#

An AI operating layer is a platform that sits between employees and an organization’s systems (knowledge sources, ticketing, HRIS, identity, ERP), providing one intelligent interface through which questions get answered and work gets executed across all of them, without replacing any of them.

Agent orchestration and agent harness#

Agent orchestration is the coordination of multiple AI agents, tools, and approval steps to complete a multi-part workflow. An agent harness is the framework agents are built and governed in. It defines what each agent may access, which policies bound it, and what requires human approval, so new agents inherit security and governance rather than reimplementing them.

Human in the loop (HITL)#

Human in the loop is a control pattern in which an AI system pauses at defined points, typically before consequential or irreversible actions, for a person to review and approve before it proceeds. It’s how organizations get the speed of automation while keeping accountability for the decisions that matter.

AI guardrails#

AI guardrails are the constraints placed on an AI system’s behavior (what it may say, access, and do), including tone and scope controls, policy boundaries, permission checks, and escalation triggers. Guardrails are configured by the organization, not the vendor, and they are what make an agent’s autonomy governable.

AI governance#

AI governance is the set of policies, roles, controls, and oversight processes an organization uses to make sure its AI systems are used responsibly, safely, and in line with the law and company policy. For AI agents, governance defines what an agent may access and do, which decisions need human approval, how every action is recorded in an audit trail, and how performance is monitored over time. Guardrails are one part of governance, and standards like ISO 42001 certify that it is in place.

At Cascade: Read AI Agent Governance Starts in the Architecture.

Grounding and RAG (retrieval-augmented generation)#

Grounding is constraining an AI system’s answers to a defined body of trusted content (your policies, plan documents, and knowledge base) rather than the model’s general training data. Retrieval-augmented generation (RAG) is the standard technique: retrieve the relevant documents first, then generate the answer from them, with sources that can be cited. Grounding is why an enterprise agent can be accurate about your company’s 401(k) match.

AI knowledge base#

An AI knowledge base is a continuously maintained, machine-usable body of organizational knowledge that AI systems draw on to answer and act. It is built from documents, policies, resolved cases, and system data, and it improves over time by flagging content that is duplicated, contradictory, or outdated. Unlike a static wiki, it learns from every resolution.

At Cascade: See the AI Knowledge Base agent.

Content gap#

A content gap is a question employees ask that the knowledge base can’t answer, because the relevant policy or article is missing, outdated, or contradictory. Content gaps show up as unanswered questions, escalations, and repeat tickets on the same topic. Finding and closing them is how an AI knowledge base improves over time.

Enterprise search#

Enterprise search is the ability to find information across an organization’s systems (documents, wikis, tickets, and apps) from one query. In employee support, it is the retrieval half of an answer: search finds the knowledge, and an agent applies it, personalizes it, and acts on it.

Model Context Protocol (MCP)#

Model Context Protocol is an open standard that lets AI assistants connect securely to external tools and data sources. In practice, MCP is how an enterprise platform’s agents can be used from inside assistants like Claude or Gemini, so employees don’t have to switch tools.

At Cascade: Connect Cascade to the AI assistants your teams already use. Learn about the Cascade MCP server.

ISO 42001#

ISO/IEC 42001 is the international management-system standard for artificial intelligence, the AI counterpart of ISO 27001 for information security. Certification means an accredited auditor has verified the organization’s AI governance: risk management, oversight, transparency, and continuous improvement in how AI is developed and operated.

At Cascade: Cascade is ISO 42001 certified. See how Cascade handles security.

Shared services#

Shared services is an operating model in which high-volume, repeatable work (employee inquiries, transactions, case processing) is consolidated from individual departments into a centralized service organization with standardized processes, defined channels, and measured service levels. HR, IT, and finance are the classic shared services functions. The model exists to deliver consistent service at lower cost than scattered, duplicated teams.

Global Business Services (GBS)#

Global Business Services is the mature form of shared services: all support functions (HR, IT, finance, procurement) integrated under one global governance structure, with unified technology, processes, and metrics. Organizations typically evolve toward GBS from function-by-function shared services as they mature.

Shared services center (SSC)#

A shared services center is the physical or virtual organization that delivers shared services, often as regional hubs providing follow-the-sun coverage for a global workforce.

Center of Excellence (COE)#

A Center of Excellence is a specialist team that designs strategy, policy, and programs in its domain, such as compensation philosophy, benefits design, or talent frameworks, while shared services executes at scale and HR Business Partners connect the work to the business. COEs are the design pillar of the three-pillar model.

HR Business Partner (HRBP)#

An HR Business Partner is a senior HR professional embedded with a business unit to work on strategy: workforce planning, organizational design, and leadership support. In the standard model, HRBPs are meant to be strategic, but they are often consumed by operational work when the shared services layer beneath them can’t resolve enough on its own.

Three-pillar (Ulrich) model#

The three-pillar model is the standard HR operating design: HR Business Partners embedded in the business (strategy), Centers of Excellence (specialist design), and shared services (execution at scale). Most large enterprises run a variant of it. The model only works when the shared services pillar resolves enough volume to keep the other two pillars strategic.

Tiered support model (Tier 0–3)#

Tiered support organizes service by escalating expertise. Tier 0 is self-service: portals, knowledge bases, and AI agents resolving requests with no person involved. Tier 1 is the first human line, handling routine cases. Tier 2 is specialists, handling complex cases like multi-state payroll corrections or network faults. Tier 3 is advanced expertise: engineers in IT, and COEs and HRBPs in HR. Every request resolved at Tier 0 costs a fraction of a Tier 1 touch, which is why automation targets Tiers 0 and 1 first.

L1 and L2 support#

L1 (Level 1) and L2 are the IT names for Tiers 1 and 2. L1 handles first-contact, routine issues such as password resets, access requests, and known errors. L2 handles what L1 escalates. “Automating L1 and L2” means resolving both routine volume and a share of specialist work without a person stepping in.

Employee self-service (ESS)#

Employee self-service is the ability for employees to resolve their own needs (checking a policy, updating personal data, requesting time off, downloading an employment letter) through a portal, app, or AI agent, without contacting HR or IT. Strong self-service is what makes Tier 0 real. Weak self-service just moves the queue.

Manager self-service (MSS)#

Manager self-service extends self-service to people managers’ tasks, such as looking up team information, initiating promotions and transfers, and handling approvals, with policy and permissions enforced automatically rather than routed through HR.

One front door#

“One front door” is the service-delivery principle that employees should have a single place to bring any request, rather than needing to know which team, portal, or queue owns their problem. The front door interprets the request and routes or resolves it, so the org chart stops being the employee’s problem.

Case management#

Case management is the structured handling of employee requests that can’t be resolved instantly: tracking each case’s status, owner, history, and communications from intake to resolution, with service levels measured. It is the HR counterpart of IT ticketing.

Escalation routing#

Escalation routing is the rules-based or AI-driven transfer of a request to the right human queue when automation or a lower tier can’t resolve it, with full context attached, so the receiving specialist doesn’t have to restart the diagnosis.

Employee experience (EX)#

Employee experience is the sum of an employee’s interactions with their employer (tools, services, support, and the moments that matter), managed as deliberately as customer experience. In service delivery, it is the employee-side argument for automation: instant, around-the-clock, judgment-free help changes how willing employees are to ask for help at all.

Employee support automation#

Employee support automation is the umbrella term for using AI and workflow technology to resolve employees’ IT, HR, and operational requests without human handling. It spans self-service, virtual agents, and agentic execution, and it is the phrase buyers often use before they’ve chosen a specific category.

Shift-left#

Shift-left is the service strategy of moving resolution to lower, less costly tiers: from specialists to the service desk, from the service desk to self-service, and ultimately to Tier 0 automation. Each tier a request moves down reduces its cost, which makes agentic automation the end state of shift-left.

ITSM (IT service management)#

ITSM is the discipline of designing, delivering, and improving IT services for an organization, covering how requests, incidents, problems, and changes are handled, measured, and governed. ServiceNow, Jira Service Management, and Zendesk are examples of ITSM platforms, and ITIL is the dominant framework.

At Cascade: See the IT Service Management agent.

ITIL#

ITIL is the most widely adopted framework of ITSM best practices. It defines the core service processes (incident, service request, problem, and change management) and the vocabulary IT organizations use for them.

AI service desk (AI help desk)#

An AI service desk, or AI help desk, is an employee support function where AI is the first responder: answering questions, executing routine requests, and escalating with context when a person is needed, rather than a ticket queue with a chatbot in front of it. The distinguishing measure is how much of the queue the AI resolves outright, versus deflecting or routing it.

Incident management#

Incident management is the process for restoring normal service when something breaks: identifying, triaging, and resolving disruptions fast enough to minimize business impact. An incident is unplanned. A service request is not.

Service request management#

Service request management handles the planned, routine requests made of IT, such as access to an application, a new laptop, or a software install, through defined fulfillment workflows with approvals. Requests are the highest-volume and most automatable category of IT work.

Service catalog#

A service catalog is the defined menu of services and requests an organization offers its employees, each with its fulfillment workflow, approvals, and SLAs attached. Catalog requests are the most automatable work in IT, because the process is already standardized.

Triage#

Triage is the review of incoming requests to classify, prioritize, and route them to the right queue. Manual triage is one of the largest hidden costs in a service desk. AI triage classifies and routes instantly, and agentic systems increasingly resolve the request instead of routing it.

Problem management#

Problem management is the process of finding and eliminating the root causes behind recurring incidents, so the same disruption stops happening. It is distinct from incident management, which restores service without necessarily fixing the underlying cause.

Change management (IT)#

Change management in ITSM is the controlled process for modifying production systems: assessing risk, obtaining approval, scheduling, and documenting changes so improvements don’t cause outages. It is distinct from organizational change management in HR.

Ticket deflection#

Ticket deflection is resolving an employee’s need before a ticket is ever created, through self-service, a knowledge article, or an AI agent, so the request never enters the human queue. Precision matters: deflection counts contacts that didn’t reach a person, which includes people who gave up. It is not a synonym for resolution. See resolution rate.

Resolution rate (resolved without escalation)#

Resolution rate without escalation is the percentage of requests actually completed, answered or executed to the employee’s satisfaction, with no human intervention. It is the stricter, more honest counterpart to deflection: deflection measures what didn’t reach an agent, while resolution measures what got done. Vendors also call it “automation rate” or “AI-resolved rate.” When comparing claims, ask which of the two is being counted.

At Cascade: Up to 99.3% of cases resolved without human escalation.

First contact resolution (FCR)#

First contact resolution is the percentage of requests resolved in the employee’s first interaction, with no follow-ups, transfers, or escalations. It is a core measure of service quality, alongside speed.

MTTR (mean time to resolution)#

MTTR is the average time from a request being raised to its full resolution. Automation shortens MTTR for routine requests from hours or days to seconds, which is usually the most visible change in the employee experience.

SLA (service level agreement)#

A service level agreement is a contractual commitment to measurable service performance, such as uptime, response times, and resolution times, with defined remedies if targets are missed. For enterprise buyers, the question is often less about the exact numbers and more about whether the vendor will commit to them contractually.

CMDB (configuration management database)#

A CMDB is the database of an organization’s IT assets and configurations (devices, software, and services) and the relationships between them, used to assess impact, troubleshoot, and govern change. CMDBs are notoriously hard to keep current, which is why AI agents that update them as a side effect of resolving requests matter.

ITAM and SAM#

IT asset management (ITAM) tracks hardware and software assets across their lifecycle: ownership, entitlement, cost, and retirement. Software asset management (SAM) is its software-specific discipline, covering licenses, usage, and compliance, including reclaiming unused licenses, one of the most directly measurable automation savings.

Joiner–mover–leaver (JML)#

Joiner–mover–leaver is the identity lifecycle: provisioning access when someone joins, adjusting it when they change roles, and revoking it when they leave. JML spans HR (the triggering events live in the HRIS) and IT (the access lives everywhere else), which makes it the classic cross-functional automation use case and, when done manually, a chronic security gap.

Provisioning and deprovisioning#

Provisioning is granting a user the accounts, access, and equipment their role requires. Deprovisioning is the timely removal of all of it when they leave. Slow provisioning wastes a new hire’s first days. Slow deprovisioning is a security risk waiting to be found in an audit.

Identity provider (IdP), SSO, and MFA#

An identity provider, such as Okta or Microsoft Entra, is the system that authenticates users and governs access. Single sign-on (SSO) lets one login open many applications, and multi-factor authentication (MFA) requires a second proof of identity. In agentic automation, the identity provider is also how an agent verifies who is asking before it acts.

HR service delivery (HRSD)#

HR service delivery is the function, and software category, for how HR provides services to employees at scale: intake channels, knowledge, case management, and increasingly AI resolution, measured with the same rigor as IT support. It is the operational layer of HR, distinct from HR strategy or program design.

At Cascade: See the HR Service Delivery agent.

HRIS and HCM#

An HRIS (human resource information system) or HCM (human capital management) platform, such as Workday, SAP SuccessFactors, UKG, or ADP, is HR’s system of record: the authoritative source for employee data, jobs, compensation, and organizational structure. Everything else in the HR stack reads from it or writes to it.

Open enrollment (OE)#

Open enrollment is the annual window, typically in the fall in the U.S., when employees choose their benefits for the coming year. It compresses a year’s worth of benefits questions into a few weeks, making it the highest-volume, highest-stakes season in HR service delivery.

At Cascade: Read the 2025 Open Enrollment report.

Benefits decision support#

Benefits decision support is guidance, increasingly AI-driven, that helps each employee choose among their actual benefit options based on their own eligibility, location, family situation, and priorities, rather than parsing plan documents alone. Done well, it changes enrollment outcomes, which makes it a cost lever, not just a convenience.

At Cascade: See the Benefits Decision Support agent.

Qualifying life event (QLE)#

A qualifying life event is a change, such as marriage, birth, divorce, or loss of other coverage, that lets an employee change benefits elections outside the open enrollment window, within a deadline. These requests are deadline-bound, personal, and policy-heavy, which is exactly when employees most need guided support.

Leave of absence (LOA) and FMLA#

Leave of absence management is the administration of extended time away (parental, medical, or military) across overlapping company policies, state programs, and federal law. FMLA is the U.S. federal law guaranteeing eligible employees up to 12 weeks of job-protected leave. Automation handles eligibility, paperwork, and tracking, while people handle the judgment calls.

Total rewards#

Total rewards is the complete package an employer provides (compensation, benefits, equity, recognition, wellbeing, and development), managed as one strategy.

Employment verification#

Employment verification is the confirmation of a person’s employment and income for a third party, such as a lender, landlord, or agency. It is a natural fit for automation: high volume, urgent for the employee, entirely rule-bound, and traditionally a multi-day HR ticket for a document a system can generate in seconds.

Onboarding automation#

Onboarding automation is the coordinated execution of everything a new hire needs before and during their first day (accounts provisioned, equipment ordered, benefits enrolled, paperwork completed, orientation scheduled), triggered from the hire event rather than chased across HR and IT queues. It is a classic cross-functional workflow: the trigger lives in the HRIS, and most of the work lives in IT.

System of record (SOR)#

A system of record is the authoritative source of truth for a category of data: the HRIS for employee data, the ITSM platform for tickets, the ERP for financials. Updating the system of record is what makes an automation claim meaningful, because the work isn’t done until the authoritative system reflects it.

System of engagement#

A system of engagement is where people actually interact, such as email, Slack, Teams, or a portal, as distinct from where the data authoritatively lives. Modern service delivery meets employees in systems of engagement while executing against systems of record.

ERP (enterprise resource planning)#

An ERP system, such as SAP, Oracle, or NetSuite, is the integrated platform for an organization’s core business operations: finance, procurement, supply chain, and often HR. In shared services automation, the ERP is a common system of record that agents read from and write to.

Integration and API#

An integration is a live connection between systems that lets them exchange data and trigger actions, typically through APIs (application programming interfaces). Pre-built integrations are often the difference between deploying in weeks and a months-long integration project.

Audit trail#

An audit trail is the tamper-evident record of every action taken in a system: who requested what, what the system did, what data it touched, when, and under whose approval. For AI agents acting on sensitive data, the audit trail is what makes autonomy accountable and compliance provable. See also observability and root cause analysis.

Observability (AI observability)#

AI observability is the ability to see what an AI system is doing and why: the requests it received, the sources it used, the tools it called, the decisions it made, and the outcomes, both in real time and after the fact. Where an audit trail records actions for accountability, observability makes the system’s behavior visible for monitoring, troubleshooting, and improvement.

Root cause analysis (RCA)#

Root cause analysis is the structured process of tracing a problem back to its underlying cause rather than its symptoms. In AI-powered support, it means following a wrong or unexpected answer back through the system’s record (the question, the sources retrieved, and each step taken) to find whether the cause was outdated content, a content gap, or a configuration issue, and fixing it at the source. See also problem management.

Role-based access control (RBAC) and least privilege#

Role-based access control grants permissions by role rather than individually. Least privilege is the principle that every user, and every AI agent, gets only the minimum access its function requires. Together, they mean an agent can only see and do what the requesting employee’s role allows.

White labeling#

White labeling is deploying a vendor’s product under the customer’s own brand (name, logo, colors, and voice) so employees experience it as their company’s own service rather than a third-party tool. In employee support, branding affects adoption, because people trust a front door that looks like their employer’s.

Vendor lock-in and platform-agnostic#

Vendor lock-in is dependence on one vendor’s ecosystem, to the point where switching becomes prohibitively costly. It is a risk buyers weigh when an automation layer is owned by the same vendor as the ticketing platform beneath it. Platform-agnostic, or independent, describes automation that works on whatever systems an organization runs and moves with it, keeping ITSM, HRIS, and identity choices separate from the AI choice.