LLM (large language model)
An LLM is a model trained on very large amounts of text that predicts the next words, and can therefore read, summarise, classify and write. It is not a knowledge base about your company, and on its own it does not check whether what it wrote is true.
Inside a business system an LLM handles text, not decisions. It reads an order email and pulls out the line items, assigns a category to a complaint and routes it to the right team, drafts a reply, turns an error message into plain language. The result reaches a person as a proposal to approve, not as a finished record in the database.
The biggest limitation is hallucination: the model always answers, including when it has nothing to go on, and it does so in a confident tone. That is why a working solution grounds answers in the company's own documents (RAG), allows an explicit "I don't know", and puts the output through ordinary rules before anything is saved: the tax ID has to exist in the customer file, line items have to add up to the document total, the item code has to come from the catalogue.
Cost is counted on the amount of text going into the model and coming out of it, so it grows with every request and every attached document. Dumping the whole database into a single request is both expensive and ineffective. When planning a budget, know from the start how many requests a month a given process generates and where to set a cap.
Data is a separate decision. Before customer emails, contracts or HR records go to a model, the company should know where that model runs, whether a data processing agreement is in place, and whether the data leaves the European Economic Area. For sensitive data the alternative is a model running on your own or rented infrastructure: slower and more expensive to maintain, but nothing is sent outside.