Hi,
If you have a company corpus of data and you feed that into the model as fine tuning, it will learn the style of the interactions, it will not learn the interactions themselves. For example, if you were to show it thousands of books an author has written, it will learn “how” the author wrote, not what the author wrote.
With this in mind, possibly you are using the wrong tool for the task, the alternative is embeddings and in those you can feed it your company documentation in sections and then perform a semantic similarity search on the data to retrieve relevant sections which can then be included as context in a prompt containing the users query, see OpenAI Platform