This glossary explains 30 terms that come up when businesses automate work, adopt AI or connect their systems. Each definition is written in plain English, starts with what the term means, and adds why it matters in practice. Where a term links to a guide or service, that page goes into more depth.
Agentic AI
Agentic AI is an approach where an AI system plans and carries out multi-step tasks toward a goal, choosing which tools to use and what to do next, rather than answering a single prompt. It can save real effort on open-ended work, but it needs clear limits, logging and human approval for actions that move money or reach customers.
AI agent
An AI agent is a piece of software that uses a language model to read a situation, decide on an action, and call tools such as email, a CRM or a database to carry it out. Agents work best on narrow, well-defined jobs with checks in place. Our guide to AI agents for small business covers practical uses.
API
An API (application programming interface) is a documented way for one piece of software to request data or actions from another. When two systems both offer APIs, they can exchange information directly without anyone retyping it or a bot clicking through screens. A good API is usually the sturdiest and cheapest basis for an integration or automation.
BPMN
BPMN (Business Process Model and Notation) is a standard set of symbols for drawing business processes, covering events, tasks, gateways for decisions, and flows between participants. It lets analysts, managers and developers read the same diagram the same way. Small teams rarely need the full standard, but its core shapes make process maps clearer and easier to hand over.
Consent Mode
Consent Mode is a Google feature that tells Google tags whether a visitor has agreed to analytics and advertising cookies, and adjusts what the tags collect accordingly. When consent is refused, tags send limited signals instead of full data. It matters for privacy compliance and for keeping campaign measurement as accurate as the rules allow. See our advertising service.
CRM
A CRM (customer relationship management) system is software that stores customer and prospect records, conversations, deals and follow-up tasks in one place. It becomes far more useful when connected to accounting, email and the website, so records stay current without manual updates. Our guide on how to integrate QuickBooks with a CRM shows a common example.
ERP
An ERP (enterprise resource planning) system is software that runs core business operations such as finance, purchasing, inventory, production and sometimes payroll from one shared database. ERPs are powerful but often hard to change, so they are a frequent integration point. Older ERPs without good APIs are one place where RPA or middleware still earns its keep.
ETL
ETL (extract, transform, load) is a pattern for moving data: extract it from source systems, transform it into a consistent format, and load it into a destination such as a reporting database or data warehouse. ETL jobs usually run on a schedule. They suit reporting and analytics well, while real-time needs are better served by APIs and webhooks.
Hallucination
A hallucination is an output from an AI model that sounds confident and plausible but is factually wrong or invented, such as a made-up figure, policy or citation. It is a known property of language models, not a rare bug. The practical answers are grounding the model in your own data, checking outputs, and keeping a person in the loop for important decisions.
Human-in-the-loop
Human-in-the-loop is a design where a person reviews, approves or corrects an automated or AI-generated result before it takes effect. It is the standard pattern for AI that drafts replies, extracts invoice data or recommends decisions. As measured accuracy improves, the share of items needing review can shrink, but high-risk actions usually keep a human approval step.
IAM (identity and access management)
IAM (identity and access management) is the set of policies and systems that decide who can sign in to which applications and what each person or service is allowed to do. It covers accounts, roles, permissions, password and MFA rules, and removing access when people leave. Automation and AI agents need IAM too, with their own limited, auditable access.
Idempotency
Idempotency is the property of an operation that produces the same result whether it runs once or several times. In integrations it prevents duplicate invoices, orders or emails when a message is retried after a timeout. Well-built automations use idempotency keys or checks for existing records, because networks fail and retries are a normal part of reliable systems.
Integration middleware
Integration middleware is software that sits between business systems and handles the movement, transformation and routing of data between them. Instead of every system connecting directly to every other, each connects once to the middleware. It makes integrations easier to monitor and change. Our system integration service covers when middleware is worth it and when direct connections suffice.
iPaaS
An iPaaS (integration platform as a service) is a cloud service for building and running integrations and workflows between applications, usually with ready-made connectors and a visual editor. Tools such as Zapier, Make and n8n fall in or near this category. Our n8n vs Zapier vs Make guide compares them on cost, control and limits.
LLM (large language model)
An LLM (large language model) is an AI model trained on very large amounts of text to predict and generate language, which lets it summarize, classify, extract, translate and draft. LLMs power most modern AI assistants and agents. They are flexible and fast, but probabilistic, so business use needs grounding in your data, testing on real examples, and sensible review.
Low-code / no-code
Low-code and no-code are approaches to building apps and workflows through visual editors and configuration rather than traditional programming. No-code aims at business users; low-code allows some scripting for flexibility. They speed up simple projects, but costs, vendor limits and maintainability matter as complexity grows, which is when custom software development can become the better choice.
MCP (Model Context Protocol)
MCP (Model Context Protocol) is an open protocol that standardizes how AI applications connect to external tools and data sources. Anthropic introduced it on November 25, 2024 (source checked October 8, 2026). A system exposes capabilities as an MCP server for compatible clients; access control and logging still need careful design.
OAuth
OAuth is an open standard that lets one application access another on a user's behalf without seeing that user's password. The user approves specific permissions, and the application receives a limited, revocable token. Most modern integrations with tools like QuickBooks, Google Workspace or Microsoft 365 use OAuth, so token storage and renewal are part of building them properly.
OCR / intelligent document processing
OCR (optical character recognition) converts images of text, such as scanned pages or PDFs, into machine-readable text. Intelligent document processing goes further, using AI to understand the document, find fields like totals and dates, and validate them. It is the foundation of invoice and form automation, covered in our guide to automating invoice processing.
Process mapping
Process mapping is the practice of documenting how a piece of work actually happens, step by step, including owners, systems, handoffs, rules, exceptions, volumes and time. It is the first step before automating anything, because it shows what to remove, simplify or automate. Our business process mapping template gives you a format to copy.
Prompt engineering
Prompt engineering is the practice of writing and testing the instructions, examples and context given to an AI model so it produces reliable, useful output for a specific task. In business systems it includes setting the output format, listing rules, and supplying reference data. Good prompts are versioned and tested on real examples, just like any other part of the software.
RAG (retrieval-augmented generation)
RAG (retrieval-augmented generation) is a technique where a system first searches your own documents or data for relevant passages, then gives them to a language model to use when answering. It lets AI respond from current, company-specific information rather than only its training data, and reduces hallucination. Answer quality depends heavily on how well the source content is organized.
RPA (robotic process automation)
RPA (robotic process automation) is software that operates existing applications the way a person would, clicking through screens, copying fields and filling forms according to fixed rules. It suits stable, high-volume tasks, especially in systems without APIs. It can break when screens change. Our guide on RPA vs AI automation explains when each fits.
Single source of truth
A single source of truth is the one agreed system or dataset that holds the authoritative version of a given piece of information, such as customer details or product prices. Other systems read from it or sync to it rather than keeping their own copies. Agreeing on it early prevents conflicting records and makes integrations far simpler to build.
SSO (single sign-on)
SSO (single sign-on) is an authentication setup that lets a person sign in once with one identity and then access several applications without logging in to each separately. It improves convenience and security, since access can be granted or removed centrally. SSO usually relies on standards such as SAML or OpenID Connect and is a core part of IAM.
Structured data / JSON-LD
Structured data is information on a web page marked up in a standard vocabulary, usually Schema.org, so search engines and AI systems can understand what the page describes. JSON-LD is the format Google recommends for adding it, as a script block separate from the visible content. It supports rich results and clearer interpretation. See our website development service.
Swimlane diagram
A swimlane diagram is a process map drawn in parallel lanes, with one lane for each role, team or system, and each step placed in the lane of its owner. Every arrow that crosses a lane is a handoff, which is where delays and errors often appear. That makes swimlanes one of the quickest ways to spot what needs redesigning.
Total cost of ownership (TCO)
Total cost of ownership (TCO) is the full cost of a system over its working life, not just the purchase or build price. It includes licences, hosting, AI usage, maintenance, support, training, upgrades and eventual replacement. Comparing options on TCO often changes the answer. Our AI automation ROI calculator guide shows how to include these costs.
Webhook
A webhook is an automatic message one system sends to another, over HTTP, the moment a specific event happens, such as a new order, a paid invoice or a form submission. It lets integrations react in near real time instead of checking for changes on a schedule. Reliable webhook handling needs verification, retries and idempotency to avoid missed or duplicate actions.
Workflow automation
Workflow automation is the use of software to run a sequence of business steps automatically based on triggers, rules and data, such as routing an approval, creating a record, or sending a notification. It ranges from simple tool-to-tool workflows to multi-system processes with AI steps. Our process automation service covers design, build and ongoing support.