AI & Automation Glossary

An AI agent is a software system that can understand a goal, decide what actions to take, use tools or data, and complete tasks with limited human input.

AI automation combines artificial intelligence with automated workflows so a system can interpret information, make context-aware decisions, and carry out business tasks.

AI governance is the set of policies, roles, controls, and review processes used to keep artificial intelligence systems lawful, secure, fair, and accountable.

An AI hallucination is a confident-looking response generated by an AI model that is inaccurate, unsupported, or invented.

An application programming interface, or API, is a defined way for software systems to exchange data and request actions from one another.

An API key is a secret credential used to identify and authorize an application when it connects to an API.

Artificial general intelligence, or AGI, is a hypothetical form of AI capable of learning and performing a wide range of intellectual tasks at a human-like level.

Artificial intelligence, or AI, is the field of building computer systems that perform tasks associated with human intelligence, such as reasoning, language, perception, and decision-making.

Backpropagation is the training method that calculates how much each neural-network parameter contributed to an error and adjusts those parameters to improve predictions.

Batch processing runs a collection of tasks or records together at a scheduled time instead of handling each item immediately.

Bias in AI is a systematic distortion in an AI system's data, design, or outputs that can create unfair or inaccurate results.

A bot is a software program designed to perform a narrow, repeatable task automatically, often through a website, messaging app, or API.

Business process automation uses software to execute repeatable operational steps such as approvals, data entry, notifications, and reporting.

A business rules engine applies explicit if-then logic to data so decisions can be made consistently inside an automated process.

A chatbot is software that communicates with people through text or voice to answer questions, collect information, or complete simple tasks.

Classification is a machine-learning task that assigns an input to one or more predefined categories, such as spam or not spam.

Cloud automation uses scripts, policies, and workflows to provision, configure, scale, and maintain cloud resources with less manual work.

Computer vision is a field of AI that enables machines to interpret images and video, including objects, text, faces, movement, and visual patterns.

A confidence score is a numeric estimate of how certain a model is about a prediction or generated result; it is not a guarantee of accuracy.

A context window is the maximum amount of text, images, or other tokenized information an AI model can consider during one interaction.

Conversational AI uses language models, speech technology, and dialogue logic to hold useful text or voice conversations with people.

CRM automation automatically updates customer records, routes leads, schedules follow-ups, triggers messages, and reports activity in a customer relationship management system.

Data annotation is the process of labeling text, images, audio, or other data so it can be used to train or evaluate AI models.

Data extraction identifies and pulls useful information from documents, databases, websites, emails, or APIs for further processing.

A data pipeline is an automated sequence that collects, transforms, validates, and moves data from one system to another.

Deep learning is a branch of machine learning that uses neural networks with many layers to learn complex patterns from large datasets.

A deterministic workflow follows predefined rules and produces the same result whenever it receives the same inputs and conditions.

A diffusion model is a generative AI model that learns to create images, audio, or other data by reversing a gradual noising process.

A digital worker is an automated software system that performs defined business activities across applications, often combining RPA, AI, and workflow tools.

Document automation creates, extracts, routes, reviews, or stores documents using templates, rules, and AI-assisted data processing.

Edge AI runs artificial intelligence models on or near the device where data is created, reducing latency and dependence on cloud connectivity.

An embedding is a numerical representation of content that captures meaning or similarity, allowing machines to compare text, images, audio, or other data.

Enterprise automation connects workflows, data, people, and governance across multiple departments and systems in a large organization.

Entity extraction identifies named items such as people, organizations, dates, products, and locations within unstructured text.

AI evaluation measures a model or system against defined criteria such as accuracy, relevance, safety, latency, cost, and task completion.

Event-driven automation starts a workflow when a specific event occurs, such as a form submission, payment, status change, or incoming webhook.

Explainable AI, or XAI, includes methods that help people understand why an AI system produced a particular prediction, recommendation, or action.

Feature engineering transforms raw data into useful variables that help a machine-learning model learn patterns and make better predictions.

Few-shot learning enables a model to perform a task after seeing only a small number of examples, often supplied in the prompt.

Fine-tuning continues training a pretrained AI model on selected examples to adapt its behavior, style, or performance for a specific use case.

A foundation model is a large AI model trained on broad data that can be adapted to many tasks, including writing, coding, vision, and analysis.

Function calling lets an AI model produce structured instructions for approved software functions, allowing it to retrieve data or take controlled actions.

A generative adversarial network, or GAN, trains a generator and a discriminator together so the generator learns to create realistic synthetic data.

Generative AI creates new content such as text, images, audio, video, or code by learning patterns from training data.

GPT stands for generative pretrained transformer, a type of language model trained to understand instructions and generate human-like text or other structured outputs.

Grounding connects an AI response to trusted source material, business data, or tool results so the answer is more relevant and verifiable.

AI guardrails are technical and operational controls that constrain inputs, outputs, tool access, and actions to reduce unsafe or unwanted behavior.

Hallucination occurs when a generative AI system produces plausible but false, fabricated, or unsupported information.

A heuristic is a practical rule or shortcut used to reach a useful result quickly when an exact calculation would be difficult or unnecessary.

Human-in-the-loop automation requires a person to review, approve, correct, or guide important decisions within an AI or automated workflow.

Hyperautomation is an organization-wide approach that combines workflow automation, AI, RPA, process mining, and integration to automate many connected processes.

Image recognition is the use of computer vision to identify objects, people, text, scenes, or patterns in digital images.

Inference is the stage when a trained AI model receives new input and produces a prediction, classification, or generated response.

Instruction tuning trains a model on examples of instructions and desired responses so it follows user requests more reliably.

An integration connects two or more applications so they can exchange data, trigger actions, and participate in the same workflow.

Intelligent automation combines rules-based workflows with AI capabilities such as language understanding, prediction, and document processing.

Intent classification determines the goal behind a message, such as requesting support, booking a meeting, or checking an order.

A job scheduler runs tasks automatically at specific times, intervals, or calendar conditions and tracks whether they succeed or fail.

Journey automation coordinates messages and actions across stages of a customer journey based on behavior, timing, and profile data.

JavaScript Object Notation, or JSON, is a lightweight text format commonly used to exchange structured data between APIs and automation tools.

K-means clustering is an unsupervised learning method that groups data points into a chosen number of clusters based on similarity.

A knowledge base is an organized collection of approved information that people, chatbots, or AI agents can search to answer questions.

A knowledge graph represents entities and the relationships between them, helping systems connect facts and reason over structured knowledge.

Knowledge retrieval finds relevant information from documents, databases, or indexes so a person or AI system can use it in a response.

A language model estimates patterns in language so it can interpret, predict, or generate sequences of text or other tokens.

A large language model, or LLM, is an AI model trained on extensive language data to understand and generate text, code, and structured information.

Latency is the time between a request and a system's response; lower latency generally makes automations and AI experiences feel faster.

Lead automation captures, enriches, scores, routes, and follows up with potential customers using connected marketing and sales workflows.

Low-code automation uses visual builders and reusable components while still allowing custom code when a workflow needs more control.

Machine learning is a branch of AI in which algorithms learn patterns from data to make predictions or decisions without every rule being manually programmed.

Marketing automation uses software to segment audiences, schedule campaigns, nurture leads, personalize messages, and measure results across channels.

Machine learning operations, or MLOps, is the practice of deploying, monitoring, versioning, and maintaining machine-learning systems reliably in production.

Model drift is a decline or change in AI performance caused by shifts in real-world data, behavior, or conditions after deployment.

Model monitoring tracks AI quality, errors, latency, cost, data changes, and safety signals after a model is deployed.

A multi-agent system uses multiple AI agents with separate roles or tools that coordinate to complete a larger task.

Multimodal AI can process or generate more than one type of information, such as text, images, audio, and video, within one system.

Named entity recognition, or NER, identifies and classifies names, places, dates, organizations, products, and other entities in text.

Natural language processing, or NLP, enables computers to analyze, understand, and generate human language.

A neural network is a machine-learning model made of connected computational layers that learn patterns by adjusting numerical weights.

No-code automation lets people build workflows through visual interfaces and prebuilt connectors without writing traditional software code.

In a workflow, a node is one step that performs an action, applies logic, transforms data, or connects to an external service.

An ontology formally defines concepts, categories, and relationships in a domain so data and AI systems share a consistent meaning.

Optical character recognition, or OCR, converts printed or handwritten text in images and scanned documents into machine-readable text.

Orchestration coordinates multiple tools, services, decisions, and workflow steps so they operate as one end-to-end process.

An output parser converts a model's response into a validated structure that software can safely store, display, or pass to another workflow step.

Overfitting happens when a model learns training data too closely and performs poorly on new, unseen examples.

A parameter is a learned numerical value inside an AI model that influences how the model transforms input into output.

Predictive analytics uses historical data, statistics, and machine learning to estimate future outcomes such as demand, churn, or sales.

Pretraining teaches a model broad patterns from a large dataset before it is adapted or prompted for a specific task.

Process automation uses technology to execute a repeatable sequence of business activities with less manual intervention.

Process mining analyzes event logs from business systems to reveal how work actually moves, where delays occur, and what can be improved or automated.

A prompt is the instruction, context, examples, or data given to an AI model to guide the response it produces.

Prompt engineering is the practice of designing and testing instructions, context, examples, and output formats to improve an AI model's results.

Q-learning is a reinforcement-learning algorithm that learns the expected value of actions in different states to choose a useful policy.

Quantization reduces the numerical precision of a model's weights or calculations to lower memory use and speed up inference, sometimes with a small quality trade-off.

Query expansion adds related words or reformulates a search request to improve the chance of retrieving relevant information.

A queue temporarily stores tasks or messages so systems can process them reliably in order or as capacity becomes available.

A reasoning model is optimized to work through multi-step problems, evaluate possibilities, and produce answers that require planning or logical analysis.

Regression is a machine-learning method that predicts a continuous numeric value, such as revenue, delivery time, or demand.

Reinforcement learning trains an agent through rewards and penalties as it takes actions in an environment.

Responsible AI is the practice of designing and operating AI systems with attention to fairness, privacy, safety, transparency, accountability, and human impact.

Retrieval-augmented generation, or RAG, retrieves relevant information from approved sources and supplies it to a generative model before the model answers.

Robotic process automation, or RPA, uses software bots to imitate rule-based actions a person performs in digital interfaces.

Rule-based automation follows explicit conditions and actions, such as sending an alert when an invoice becomes overdue.

Semantic search uses meaning and context, often through embeddings, to find relevant results beyond exact keyword matches.

Sentiment analysis estimates the emotional tone or opinion expressed in text, such as positive, negative, or neutral.

Speech recognition converts spoken language into text that can be searched, analyzed, or used to trigger automated actions.

Supervised learning trains a model on labeled examples so it can learn to predict the correct label or value for new inputs.

Synthetic data is artificially generated information designed to resemble real data for training, testing, privacy protection, or simulation.

A system prompt is a high-priority instruction that defines an AI assistant's role, rules, boundaries, and expected behavior.

Task automation uses software to complete a specific repeatable activity, such as renaming files, sending reminders, or updating records.

A token is a unit of text or other input that an AI model processes; words may be split into one or more tokens.

Tokenization divides text, code, images, or other inputs into units that an AI model can process numerically.

Tool calling allows an AI model to request approved external functions, such as searching a database, sending an email, or updating a CRM.

Training data is the collection of examples used to teach an AI model patterns, relationships, and desired behaviors.

A transformer is a neural-network architecture that uses attention mechanisms to process relationships within sequences and powers many modern language models.

A trigger is the event, schedule, or condition that starts an automated workflow.

User-interface automation controls on-screen elements such as buttons, forms, and menus to complete tasks in applications that lack suitable APIs.

Unstructured data is information without a fixed table-like format, including documents, emails, images, recordings, and free-form text.

Unsupervised learning finds patterns, structures, or groups in data that does not have predefined labels.

Uptime monitoring automatically checks whether a website, API, or service is available and alerts the responsible team when it fails.

User intent is the underlying goal a person wants to achieve through a search, message, command, or interaction.

Validation checks whether data, an AI output, or a workflow result follows required rules before it is accepted or used.

A vector database stores and searches embeddings, making it useful for semantic search, recommendations, and retrieval-augmented generation.

A vector embedding is an array of numbers that represents the meaning or features of content so similar items can be found mathematically.

Versioning records changes to prompts, models, datasets, workflows, or code so teams can compare, reproduce, and safely roll back systems.

A virtual assistant is software that helps users complete tasks or find information through text, voice, or automated actions.

A voice agent combines speech recognition, language understanding, reasoning, and speech synthesis to hold spoken conversations and perform actions.

Web scraping automatically extracts information from web pages; it should be implemented in line with permissions, terms, privacy rules, and applicable law.

A webhook is an HTTP message sent automatically when an event occurs, allowing one application to notify or trigger another in real time.

A workflow is an ordered series of steps, decisions, people, and systems used to complete a business process or task.

Workflow automation uses software to move data and tasks through predefined steps, reducing repetitive work and improving consistency.

Workflow orchestration coordinates multiple automated processes, applications, dependencies, and error-handling rules from end to end.

XAI is an abbreviation for explainable artificial intelligence: methods that make an AI system's outputs and decision factors easier for people to understand.

XGBoost is an efficient gradient-boosted decision-tree algorithm widely used for classification, regression, ranking, and structured-data prediction.

Extensible Markup Language, or XML, is a structured text format used to store and exchange data between systems.

XML automation reads, transforms, validates, or routes XML data between applications without manual copying.

XPath is a query language for selecting elements and values within XML or HTML documents, often used in extraction and automation.

YAML is a human-readable data format commonly used to configure software, infrastructure, AI tools, and automated workflows.

Yield optimization uses data, rules, or predictive models to improve the return from limited resources such as inventory, ad spend, capacity, or pricing.

A YOLO model is a real-time object-detection model that identifies and locates multiple objects in an image or video in a single pass.

YouTube automation uses connected workflows to assist with research, scripts, asset production, publishing, metadata, moderation, and performance reporting.

Zapier is a cloud automation platform that connects applications through trigger-and-action workflows commonly called Zaps.

Zero-code automation is another term for no-code automation: building workflows through visual configuration without writing software code.

Zero-shot learning is the ability of a model to perform a task or recognize a category without receiving task-specific labeled examples.

Zero-shot prompting asks an AI model to complete a task using instructions alone, without providing worked examples in the prompt.

Zero-touch automation completes a defined process without routine human intervention while still using monitoring, controls, and exception handling.