AI glossary for businesses

An AI agent is an AI system that works towards a goal by planning steps, using tools and acting on its own, rather than only answering one question at a time. It can read emails, look up data, update a spreadsheet or draft a reply, then check the result and choose the next step, within limits you set.

An AI copilot, or AI assistant, is an AI feature built into software you already use, such as email, documents, spreadsheets or a CRM, that suggests, drafts and summarises while you stay in control. Examples include Microsoft 365 Copilot, Gemini in Google Workspace and Notion AI. The person reviews and decides; the assistant speeds up the work.

AI governance is the set of roles, rules and routines a company uses to decide which AI it uses, how, and who is responsible. It covers approving tools, protecting data, checking quality, meeting legal duties such as the EU AI Act and GDPR, and reviewing results over time. For an SME it can be light, but it should be clear.

An AI hallucination is an answer from an AI model that sounds confident and plausible but is false, unsupported or invented. Examples include made-up facts, wrong figures, fake quotes or sources that do not exist. It happens because language models generate likely-sounding text rather than checking facts against a trusted source.

AI literacy is the knowledge and skills people need to use AI tools sensibly: what they do well, where they fail and what risks they bring. Under Article 4 of the EU AI Act, organisations that provide or use AI systems must take measures to ensure sufficient AI literacy among their staff, a duty in force since 2 February 2025.

An AI pilot is a small, time-boxed test of one AI use case with real users, real data and success measures agreed in advance. Its job is to show whether the idea works in your business before you spend more. A good pilot ends with a clear decision: scale it, change it or stop it.

An AI roadmap is a prioritised plan that shows which AI use cases a business will tackle, in what order, who owns each one and how success will be measured. It connects business goals to concrete pilots. For an SME, a useful roadmap is short, detailed for the next 90 days, and revised as you learn.

An AI usage policy is a short internal document that tells staff which AI tools they may use, for which tasks, with which data, and how to check the results. It turns general principles into clear do's and don'ts. It is one of the simplest ways to reduce data leaks, errors and legal risk when a team starts using AI.

An AI use case is a specific, repeatable task where AI helps a defined group of people reach a measurable result. "Use AI in marketing" is not a use case. "Draft first versions of product descriptions for new arrivals, checked by the e-shop manager" is. Clear use cases are the building blocks of any AI roadmap.

An API (application programming interface) is a defined way for one piece of software to request data or actions from another. Instead of a person clicking through screens, programs exchange structured messages. AI vendors such as OpenAI, Anthropic and Google offer APIs, so businesses can add AI models to their own systems and automations.

Artificial intelligence (AI) is software that performs tasks that normally need human judgement, such as understanding text, recognising images, spotting patterns or making predictions. It learns these skills from large amounts of data instead of following only hand-written rules. Most small businesses use AI through ready-made tools rather than building their own.

A chatbot is software that holds a conversation with people through text or voice, usually on a website, in a messaging app or inside a company tool. Simple chatbots follow fixed menus and rules. AI chatbots use language models to understand free-form questions and write their own replies, which makes them more flexible but also less predictable.

Computer vision is the area of AI that lets software interpret images and video: recognising objects, reading text, counting items or spotting defects. It sits behind face recognition on phones, document scanning and automatic photo tagging. For businesses, it turns pictures, scans and camera feeds into data you can search and use.

A context window is the maximum amount of text an AI language model can take into account at one time, measured in tokens. It includes your instructions, any documents you add, the earlier conversation and the model's own reply. Anything beyond that limit is cut off or ignored, so the model effectively cannot see it.

A data protection impact assessment (DPIA) is a structured risk assessment required by Article 35 of the GDPR before processing that is likely to result in a high risk to people's rights and freedoms. It describes the processing, checks that it is necessary and proportionate, assesses the risks and sets out measures to reduce them.

Embeddings are lists of numbers that represent the meaning of a piece of text, an image or another item, so that a computer can compare items by meaning. Items with similar meaning get similar numbers, even if they use different words or languages. They power semantic search, recommendations and the retrieval step in RAG.

The EU AI Act (Regulation (EU) 2024/1689) is the European Union's law on artificial intelligence. It sorts AI systems by risk, bans a small set of harmful practices, sets strict duties for high-risk systems and adds transparency rules for tools like chatbots. It applies to companies that build AI and to those that use it at work.

Fine-tuning means taking an AI model already trained on general data and training it further on a smaller set of your own examples. The goal is to make it follow a specific style, format or task more consistently. It changes how the model behaves, but it is not the best way to teach it facts that change often.

Generative AI is a type of artificial intelligence that creates new content, such as text, images, audio or code, in response to a request written in everyday language. It learns patterns from very large collections of existing content and uses them to produce new, original-looking output. ChatGPT, Claude and Google Gemini are well-known examples.

A high-risk AI system is an AI system that the EU AI Act considers able to seriously affect people's health, safety or fundamental rights. It includes AI used as a safety component of regulated products and AI used in areas listed in Annex III, such as recruitment, credit scoring and education. These systems must meet strict requirements.

A large language model (LLM) is an AI model trained on very large amounts of text to understand and produce human language. It works by predicting the most likely next piece of text, which lets it answer questions, summarise, translate and draft. LLMs power tools such as ChatGPT, Claude, Google Gemini and Microsoft 365 Copilot.

Machine learning is a branch of artificial intelligence in which computers learn patterns from past data and use them to make predictions or decisions, without being programmed with explicit rules for each case. It powers sales forecasts, spam filters, product recommendations and fraud detection, and it is also the foundation on which today's generative AI is built.

Multimodal AI is AI that can work with more than one type of content, such as text, images, audio and video, within the same model. You can show it a photo and ask about it, or speak and get a written answer. Recent versions of ChatGPT, Google Gemini and Claude all accept images as well as text.

Natural language processing (NLP) is the field of AI that lets computers read, interpret and produce human language, written or spoken. It powers spam filters, translation, sentiment analysis, chatbots and today's large language models. For a business, NLP turns emails, reviews and documents into something software can sort and act on.

Open-source AI models are AI models whose weights, and sometimes code and training details, are published so anyone can download, run and adapt them. Examples include Meta's Llama, Mistral's open models, Google's Gemma and Alibaba's Qwen. Many are more accurately called "open-weight", because their licences or training data are not fully open.

A prompt is the instruction, question or material you give an AI tool to tell it what you want. It can be one short sentence or a full brief with context, examples and the format you expect. The quality of the prompt strongly affects the quality of the answer, so clear prompts save editing time.

Prompt engineering is the practice of designing, testing and improving the instructions you give an AI model so that it produces reliable, useful results. It combines clear wording with techniques such as giving context, showing examples, asking for a fixed format and breaking a task into steps. For most businesses it is a working skill, not a technical job.

Retrieval-augmented generation (RAG) is a method in which an AI system first searches a chosen set of documents for relevant information and then uses that information to write its answer. It lets a language model answer from your own policies, catalogues or manuals instead of relying only on what it learned in training, and it can show its sources.

The ROI (return on investment) of AI compares the value an AI project creates with its full cost. Value can be time saved, extra revenue, fewer errors or faster service. Costs include licences, setup, training and the time people spend checking output. ROI is only credible when you measure it against a baseline, ideally in a pilot.

A token is the small unit of text that an AI language model reads and produces. A token can be a whole word, part of a word, a number or a punctuation mark. Models count everything in tokens: the length of your input, the size of the context window and, when you use an API, the price you pay.

A vector database is a database built to store embeddings, the lists of numbers that represent meaning, and to find the items closest to a query very quickly. Instead of matching exact words, it returns content that is similar in meaning. It is a common building block behind AI search, recommendations and RAG assistants.

Workflow automation means using software to carry out the repeatable steps of a business process, such as copying data, sending notifications or creating tasks, without manual work. Tools like Zapier and Make connect apps through simple "when this happens, do that" rules. Adding AI lets those workflows also read, sort and draft text.