AI Glossary for Business: The Terms That Matter
Why a glossary
Conversations about AI for business fill up fast with jargon, and some vendors lean on that. This glossary explains the key terms as they are actually used in projects, in plain language, without hype. Each term links to the page where we show what it looks like in practice.
Artificial intelligence (AI)
The broad umbrella: systems performing tasks that until recently required human judgment, from pattern recognition to drafting text. In a business context the real test is not “how smart is it” but which business decision it improves. That distinction is the backbone of AI for business: what actually works.
Machine learning
The main way AI is built today: models that learn patterns from historical data rather than being programmed with fixed rules. The practical implication: system quality depends on your data quality, which is why characterization always starts from the data, not the algorithm.
AI agent
Software connected to your organization’s knowledge and systems, operating within defined boundaries and performing tasks, as opposed to a chatbot that only talks. The full distinction, including the permissions and security questions, is explained in AI agents for business.
RAG (retrieval-augmented generation)
The language model answers only after relevant content has been retrieved from your own repository, so answers are source-grounded and checkable. It is the technique behind organizational knowledge agents, and behind the AI Mode on this very site.
Large language model (LLM)
A model trained on vast amounts of text that can understand and produce language. In business it becomes useful when connected to your knowledge (see RAG) and given boundaries; on its own it is impressive but disconnected from your reality. Such models can also run entirely inside a closed network.
Demand forecasting
Predicting expected demand for each product, across several horizons, for purchasing, production, and planning. Our working principle: a forecast that is not graded monthly against actuals is a polite guess. What it looks like in practice: demand and sales forecasting.
Dynamic safety stock
The reserve quantity held per item, recomputed as demand and supply shift, instead of a fixed percentage set once and forgotten. It is the foundation of per-unit inventory intelligence, with a success KPI fixed up front on purchasing error.
Computer vision
Models that identify, count, and measure in images and video, from visual quality control on a production line to measuring aerial imagery. Our approach always includes human oversight, as described in computer vision and inspection.
Air-gapped network
An environment fully disconnected from the internet. Contrary to common belief, complete AI systems can run inside one, including language models, document retrieval, and vision, without a single datum leaving. The specialty is detailed in secure and air-gapped AI.
Staged rollout
Bringing a system into use in measured steps instead of a hard cutover: first alongside the existing process, then at limited scope, finally as the default, with each step judged by numbers. That is how our engagements are built, honest exit points included.
Success KPIs
The numbers fixed before building that will define whether the system worked: for example, reducing purchasing error by at least 10% from the current state. Without them, success is a matter of impression. Ours are set during characterization, before a line of code is written.
Missing a term? Ask the AI Mode on this site, or write to us and we will add it.
What's the difference between AI and machine learning?
Artificial intelligence is the broad umbrella: systems performing tasks that used to require human judgment. Machine learning is today's main way of achieving it: models that learn patterns from historical data instead of being programmed with rules. Most business AI systems we build are machine learning over the organization's own data.
What is an AI agent?
Software connected to your organization's knowledge and systems that makes decisions within defined boundaries and takes actions, as opposed to a chatbot that only answers. A good agent also knows what it must never do alone and when to hand off to a human.
What is RAG?
Retrieval-Augmented Generation: a language model answers only after relevant content is retrieved from your knowledge base. Answers are grounded in your sources rather than the model's general memory, and every answer can point to where it came from.