Glossary of AI in business
Fifteen terms you will meet in our pages and in discussions about AI. First in plain words, then the technical meaning.
- On-premise AI (local AI)
In plain wordsAn AI running on a machine at your premises, not on the Internet.
Technical meaningThe language model and its data run on a server in your premises or on your network. No request goes through an external provider.
AI installed in your company- Language model (LLM)
In plain wordsThe program that reads and writes text like a human.
Technical meaningA neural network trained on large volumes of text to predict the continuation of a sequence. It is then adapted to a use case through instructions and the context provided to it.
- AI copilot
In plain wordsAn AI that prepares the work; a human validates.
Technical meaningAn assistant embedded in an existing process: it suggests a reply, a document or an action, and the user reviews, corrects and triggers the send.
Sales and customer service- AI agent
In plain wordsAn AI that chains several actions to reach a goal.
Technical meaningA language model equipped with tools (search, database, API) that it calls in a loop until a task is done. Scope and action rights are defined in advance.
- RAG
In plain wordsThe AI looks up the answer in your documents before writing.
Technical meaningRetrieval-Augmented Generation: before generating, the system retrieves the relevant passages from your data and places them in the model's context. Answers rely on your sources, not on the model's memory.
Client case- Vector database
In plain wordsA database that finds similar texts by meaning, not by the exact word.
Technical meaningEach text is converted into a numerical vector (embedding). Search compares vectors to find the passages semantically close to a question.
- Hallucination
In plain wordsWhen the AI invents plausible but false information.
Technical meaningThe model generates a statement not grounded in its sources. It is limited by providing context (RAG), citing sources and keeping human validation.
- Context
In plain wordsWhat the AI has in front of it when answering.
Technical meaningAll the text given to the model for a request: instructions, history, retrieved documents. Its size is limited; what is not in it, the model does not see.
- Prompt
In plain wordsThe instruction given to the AI.
Technical meaningThe instructions, examples and constraints that frame the model's answer. In a copilot it is written once, tested and versioned like code.
- On-premise
In plain wordsInstalled in your premises, on your hardware.
Technical meaningHosting on infrastructure the company owns and administers, as opposed to the cloud. You control physical access, the network and backups.
AI installed in your company- Data sovereignty
In plain wordsYou decide where your data is and who accesses it.
Technical meaningData stays under the company's jurisdiction and control. No foreign subcontractor has access to it, neither for processing nor for training models.
Data and Swiss FADP- Swiss FADP (nLPD)
In plain wordsThe Swiss data protection act, in force since 2023.
Technical meaningThe revised Federal Act on Data Protection (1 September 2023). It requires, among other things, documenting processing, informing data subjects and securing data. Keeping data on site simplifies this documentation.
Data and Swiss FADP- ERP
In plain wordsThe management software: items, customers, orders, stock.
Technical meaningEnterprise Resource Planning. The reference source for business data; a copilot connects to it through exports or an interface (API) to read items, customers and history.
- Match rate
In plain wordsThe share of conversations that can be linked to a known customer.
Technical meaningA data quality indicator: the proportion of exchanges (chat, e-mail) reliably associated with an ERP customer account. It determines how relevant the displayed history is.
Client case- Availability
In plain wordsThe share of time the service responds.
Technical meaningPercentage of time during which the application is reachable and working. Measured by monitoring that only exposes technical indicators, not the data.
AI installed in your company