AI & Technology Definitions

A set of rules and protocols that allows different software applications to communicate with each other. In AI contexts, APIs enable your systems to send data to AI services and receive predictions or responses.

AI systems that can autonomously make decisions and take actions to achieve specified goals, often using tools and external data sources. These systems go beyond simple Q&A to perform complex multi-step workflows.

Technologies and strategies for analyzing business data to inform decision-making. Traditional BI focuses on historical reporting, while AI-enhanced BI adds predictive capabilities.

A state statute intended to enhance privacy rights and consumer protection for California residents. Critical when evaluating AI vendors that process customer data.

The amount of text (measured in tokens) that an AI model can process at once. Larger context windows allow the AI to "remember" more information during a conversation or analysis. For example, Claude 3.5 has a 200K token context window.

Software systems designed to manage a company's interactions with current and potential customers. Common CRM platforms include Salesforce, HubSpot, and Microsoft Dynamics. Often integrated with AI for predictive analytics and automation.

The physical location where data is stored. Many regulations (GDPR, HIPAA) require that certain types of data remain within specific geographic boundaries.

A legal contract between a data controller and data processor that outlines responsibilities for protecting personal data. Required for GDPR compliance when using third-party AI vendors.

Mathematical representations of text, images, or other data as vectors (arrays of numbers). Embeddings allow AI to understand semantic similarity - for example, "dog" and "puppy" would have similar embeddings even though the words are different.

Integrated software platform that manages core business processes (finance, HR, supply chain, manufacturing). Common ERP systems include SAP, Oracle, Microsoft Dynamics, and Epicor.

The process of taking a pre-trained AI model and further training it on your specific dataset to improve performance for your use case. Generally requires ML expertise and significant data volume.

Large-scale AI models trained on broad data that can be adapted to a wide range of tasks. Examples include GPT-4, Claude, and Gemini. These models serve as the "foundation" for many AI applications.

European Union regulation governing data protection and privacy. Applies to any organization processing data of EU residents, regardless of where the organization is located. Violations can result in fines up to 4% of global revenue.

AI systems that can create new content (text, images, code, music, etc.) based on patterns learned from training data. Distinct from traditional AI that focuses on classification or prediction.

When an AI model generates information that sounds plausible but is factually incorrect or fabricated. A critical concern for business applications where accuracy is essential.

U.S. legislation that provides data privacy and security provisions for safeguarding medical information. Required when AI systems process health-related data.

The process of using a trained AI model to make predictions or generate outputs on new data. This is distinct from training. Inference costs are often priced per API call or per token.

AI models trained on vast amounts of text data that can understand and generate human-like text. Examples include GPT-4, Claude, Gemini. The "large" refers to billions of parameters that enable sophisticated language understanding.

A subset of AI focused on systems that learn from data to improve their performance over time without being explicitly programmed. Includes supervised learning, unsupervised learning, and reinforcement learning.

A standardized way for AI applications to connect to data sources and tools. MCP servers act as intermediaries that allow AI agents to interact with your enterprise systems safely. Introduced by Anthropic in late 2024.

Software that acts as a bridge between different applications or systems, enabling them to communicate. Often used to connect AI solutions to ERP/CRM systems without direct API integration.

A computational model inspired by biological neural networks in the human brain. Consists of interconnected nodes (neurons) organized in layers that process and transform data. The foundation of modern deep learning.

An authentication protocol that allows users to log in to applications using existing credentials (like Google or Microsoft accounts) without sharing passwords. Critical for secure SSO implementation.

An identity layer built on top of OAuth 2.0 that allows clients to verify user identity and obtain basic profile information. Widely used for enterprise single sign-on implementations.

Security standards for organizations that handle credit card information. Critical when AI systems process payment data or connect to systems containing payment information.

The practice of crafting effective instructions (prompts) to get desired outputs from AI models. A critical skill for Tier 1-2 implementations that don't require custom model training.

A type of attack where malicious instructions are embedded in user input or data to manipulate AI behavior. For example, tricking an AI assistant into revealing sensitive data or bypassing safety restrictions.

A technique where an AI model retrieves relevant information from a knowledge base before generating a response. This grounds the AI's answers in your actual data, reducing hallucinations. Common for Tier 3 implementations.

An architectural style for building web APIs that uses HTTP requests to access and manipulate data. The most common type of API used for integrating AI services with enterprise systems.

A financial metric measuring the profitability of an investment, calculated as (Gain - Cost) / Cost. For AI projects, includes both direct cost savings and value from improved outcomes.

An XML-based protocol for exchanging authentication and authorization data between parties. Commonly used for enterprise SSO implementations.

Search technology that understands the meaning and context of queries rather than just matching keywords. Uses embeddings to find conceptually similar content even when exact words don't match.

A contract defining expected service performance metrics (uptime, response time, etc.) and remedies if not met. Critical for production AI deployments where downtime impacts operations.

Authentication scheme allowing users to log in once and access multiple applications without re-entering credentials. Essential for enterprise AI tool adoption and security.

Comprehensive assessment of all costs associated with a technology investment over its lifetime, including licensing, implementation, training, maintenance, and hidden costs.

The basic units that AI models process. Roughly, 1 token ≈ 4 characters or ¾ of a word. AI pricing is typically per-token for both input (prompt) and output (completion).

A neural network architecture introduced in 2017 that revolutionized natural language processing. Uses attention mechanisms to process entire sequences simultaneously. The foundation of modern LLMs like GPT and Claude.

The technique of starting with a pre-trained model and adapting it for a specific task, rather than training from scratch. The foundation of modern AI - you don't need to train your own LLM from scratch.

A specialized database optimized for storing and querying embeddings (vector representations of data). Essential for RAG systems and semantic search. Examples: Pinecone, Weaviate, Chroma.

The ability of an AI model to perform tasks it wasn't explicitly trained on, using only instructions in the prompt. Modern LLMs excel at zero-shot tasks without needing examples.

Software robots that automate repetitive, rules-based computer tasks by mimicking human actions. RPA bots follow predefined workflows to perform tasks like data entry, form filling, and system navigation. Unlike AI, RPA doesn't learn—it executes programmed instructions.

Technology that converts images of text (from scanned documents, photos, or PDFs) into machine-readable text data. Modern OCR often uses deep learning (Tier 3) for accuracy, but basic OCR is Tier 1-2 technology.

Branch of AI focused on enabling computers to understand, interpret, and generate human language. Includes tasks like sentiment analysis, translation, text summarization, and question answering. Modern NLP uses transformer models (Tier 3-4).

A type of deep learning neural network architecture specifically designed for processing grid-like data such as images. CNNs automatically learn spatial hierarchies of features, making them ideal for computer vision tasks like object detection and image classification.

Neural network architecture designed for sequential data like time series, text, or speech. RNNs have loops that allow information to persist, making them suitable for tasks where context and order matter. Modern variants include LSTMs and GRUs.

A specialized type of RNN designed to learn long-term dependencies in sequential data. LSTMs solve the "vanishing gradient" problem of traditional RNNs, making them better at remembering information over long sequences. Commonly used for speech recognition and language modeling.

Specialized processors originally designed for graphics rendering, now essential for training deep learning models. GPUs can perform thousands of mathematical operations in parallel, making them 10-100x faster than CPUs for AI training tasks.

Google's custom-designed processors specifically optimized for machine learning workloads. TPUs are even more specialized than GPUs for AI tasks, offering higher performance for training and inference of neural networks.

Process of extracting data from source systems, transforming it into a usable format, and loading it into a target database or data warehouse. Essential for preparing data for AI/ML models. Modern "ELT" variants load first, then transform.

Network of physical devices embedded with sensors, software, and connectivity that collect and exchange data. IoT generates massive amounts of real-time data that can feed AI systems for predictive maintenance, quality control, and operational optimization.

Unique identifier for each distinct product and service that can be purchased. Used in inventory management to track stock levels, sales, and fulfillment. Distributors may manage thousands to tens of thousands of SKUs.

Software system that manages transportation operations including route planning, carrier selection, freight audit, and delivery tracking. Modern TMS platforms incorporate Tier 2 ML for dynamic route optimization and cost reduction.

Software that controls warehouse operations including inventory tracking, picking, packing, shipping, and labor management. Advanced WMS platforms use Tier 2 ML for slotting optimization, labor scheduling, and operational efficiency.

Small-scale implementation designed to verify that a concept or technology works for a specific use case before full-scale deployment. POCs typically last 4-12 weeks and test core functionality with real data.

Version of a product with just enough features to satisfy early customers and provide feedback for future development. In AI projects, an MVP tests the core AI capability without full-scale integration or optimization.

Measure of staffing that converts part-time work into the equivalent of full-time positions. For example, two half-time employees equal 1.0 FTE. Used to quantify team size requirements: Tier 0 needs 0.25-1 FTE, Tier 3 needs 5-12 FTEs.

Quantifiable measures used to evaluate success in meeting business objectives. Common AI project KPIs include accuracy rates, cost savings, processing time reduction, and error rates. Essential for measuring ROI and project success.

Standardized electronic exchange of business documents (purchase orders, invoices, shipping notices) between organizations. Common in distribution and manufacturing for automated order processing and supplier integration.

Warehouse facility that receives, stores, and ships products to customers or retail locations. Multi-DC networks require sophisticated inventory optimization (Tier 2 ML) to balance stock levels, minimize costs, and maintain service levels across locations.

Practices and tools for deploying, monitoring, and maintaining machine learning models in production. MLOps combines ML, DevOps, and data engineering to ensure models remain accurate, scalable, and reliable over time. Critical for Tier 2+ implementations.

Practices combining software development and IT operations to shorten development cycles and deliver high-quality software continuously. Includes CI/CD pipelines, infrastructure automation, and monitoring. Essential for production AI deployments.

Automated practices for frequently merging code changes (CI) and automatically deploying them to production (CD). Essential for AI projects to rapidly iterate on models, test changes, and deploy updates without manual intervention.

Encrypted connection between networks over the internet that allows secure data transfer. Common for hybrid AI architectures where on-premise ERP systems securely replicate data to cloud analytics platforms.

Software licensing model where applications are centrally hosted in the cloud and accessed via web browser or API. Customers pay subscription fees rather than buying perpetual licenses. Common for modern AI platforms—pricing typically per-user or per-API-call.

Auditing standard for service providers storing customer data in the cloud. SOC 2 reports verify that vendors have appropriate security controls for confidentiality, availability, processing integrity, and privacy. Required for enterprise AI vendor evaluation.

International standard for information security management systems. ISO 27001 certification demonstrates that an organization has systematic processes for managing sensitive data and security risks. Often required for enterprise AI vendors.

Security process requiring users to provide two or more verification factors to access a system (e.g., password + phone code). Essential for protecting AI systems and data. Also known as 2FA (Two-Factor Authentication) when exactly two factors are used.

Contract between a HIPAA-covered entity and a third-party service provider (business associate) that accesses protected health information (PHI). Required when AI vendors process healthcare data. Defines data protection responsibilities and liability.

Any health information that can be linked to a specific individual, protected under HIPAA regulations. Includes medical records, test results, insurance information, and billing data. AI systems processing PHI require BAAs and strict data controls.

Centralized repository that consolidates data from multiple sources (ERP, CRM, etc.) into a single database optimized for analytics and reporting. Not inherently "AI"—this is 1990s technology. Essential infrastructure for Tier 2+ AI but expensive (cloud data warehouse costs $2K-$20K/month).

Storage repository that holds vast amounts of raw data in its native format (structured, semi-structured, and unstructured) until needed. Unlike data warehouses, data lakes store data before defining its use. Required for Level 3+ data readiness and Tier 3+ AI implementations.

Process of using domain knowledge to create new variables (features) from raw data that make machine learning models more accurate. For example, creating "day of week" and "is_holiday" features from date fields for demand forecasting. Critical for Tier 2 ML success.

Process of feeding data to a machine learning algorithm to learn patterns and make predictions. Training requires historical data, computing resources (often GPUs), and time (hours to weeks). Distinct from inference (using the trained model). Tier 2+ requires ongoing retraining.

Process of integrating a trained AI model into production systems where it can make predictions on new data. Includes infrastructure setup, API integration, monitoring, and fallback mechanisms. Tier 2+ deployments require MLOps expertise.

Machine learning approach where models learn from labeled training data (input-output pairs). For example, learning to predict customer churn by training on historical data where you know which customers left. Most common ML type for business applications (Tier 1-2).

Machine learning that finds patterns in data without labeled examples. Common applications include customer segmentation (clustering), anomaly detection, and dimensionality reduction. Used when you have data but no predefined categories or outcomes.

Machine learning where an agent learns by taking actions in an environment and receiving rewards or penalties. Used for complex optimization problems like route planning, game playing, and robotic control. Tier 5 technology requiring specialized expertise.

Software deployed and running on servers physically located at the organization's facilities, as opposed to cloud hosting. Organizations maintain full control but bear infrastructure and maintenance costs. Hybrid architectures (on-premise ERP + cloud AI) are common.

Applications designed specifically to run in cloud environments, using microservices, containers, and dynamic orchestration. Cloud-native apps scale automatically, update without downtime, and are more resilient than traditional architectures. Common for modern AI/ML platforms.

IT architecture running some systems on-premise (like ERP) while using cloud services for specific capabilities (analytics, AI, backups). Extremely common—70%+ of enterprises use hybrid models. Allows gradual cloud migration without disrupting critical systems.

Automated coordination of multiple AI models, tools, or services to accomplish complex tasks. In AI context, orchestration manages workflow between LLMs, data sources, and business systems. MCP (Model Context Protocol) is a standardized orchestration approach.

Powerful machine learning technique that builds models sequentially, with each new model correcting errors from previous ones. Popular for structured business data (Tier 2). Common implementations include XGBoost and LightGBM. Often outperforms deep learning for tabular data.