ChatGPT
ChatGPT is OpenAI’s artificial intelligence assistant. In companies it is used to query internal documents and procedures, write and review texts, analyse long material and add automated steps to processes through APIs. It comes in personal and business editions, which differ mainly in how data is handled and in administration tools.
ChatGPT at a glance
Answers on internal documents
You upload manuals, price lists, procedures, contracts or specifications and ask in plain language what you need to know. The assistant answers based on that material and not only on its general knowledge. It is the fastest way to make searchable an archive that today is browsed by hand.
Assistants configured for a task
You can create assistants dedicated to a recurring activity, with fixed instructions and attached reference documents: standard sales replies, first-level internal support, formal document checks. Users no longer have to explain the context every time and answers stay consistent across colleagues.
Writing, review, analysis
First drafts of quotes and communications, rewriting technical texts in plain language, summaries of meetings and documentation, comparison between versions. The result is working material to review, not a text ready to send: the signature and responsibility remain with whoever sends it.
API and automation
The same models can be called from software. This is how copy-and-paste work stops: the incoming email is classified and routed, data from a PDF ends up in the ERP, the draft reply reaches the operator ready to approve. This part requires development and maintenance, not just a licence.
Where data ends up
Business editions of ChatGPT contractually provide that customer content is not used to train models, together with admin controls, company access management and retention policies. Free and personal editions follow different terms: using them with company data is a choice to be made consciously, not out of habit.
What it does not do
It is not an archive or a source of truth: it can produce wrong statements phrased convincingly, especially if it does not have the right documents in front of it. It should be placed where an error is visible and correctable by a person, and kept out of steps where a wrong answer reaches the client or moves money unchecked.
ChatGPT features
In companies ChatGPT comes in three forms. The application, which people work with every day on texts and files. Shared spaces and assistants, where instructions and reference documents are fixed so the department works with the same context. And the API, through which models enter existing applications and flows. Business editions add the admin side to all this: access tied to company identity, centralised user management, visibility on usage.
Before features, the data perimeter must be decided: which information can leave your environment, under which contractual terms, and whether residency constraints or alternative delivery models are needed, such as OpenAI models made available within Microsoft’s cloud platform. These choices change the architecture and must be made at the start; the topic is covered on the page Digital sovereignty.
- Analysis of uploaded files: text documents, PDFs, presentations and spreadsheets become the basis for answers, with an explicit request to stick to that material.
- Custom assistants: permanent instructions and attached documents for a recurring task, shareable with colleagues in the department.
- Connection to company sources: in business editions the assistant can be enabled to search content in authorised company systems, according to the permissions of whoever asks.
- Data processing and calculations on files, useful for checks, preliminary reconciliations and preparing lists that today are tidied up by hand.
- Support for writing code, queries and small automations, with human review before putting them into production.
- API for application integration, choosing the model based on the balance between answer quality, speed and cost per call.
- Centralised administration in business editions: login with company credentials, user creation and revocation, control over content retention.
- Data terms of business plans: company content does not feed model training, unlike free personal use.
How we put it into operation
The starting point is not the licence but the activity to lighten: who does it, on which documents, how much time it takes and what happens when the result is wrong. In practice ChatGPT has often already entered the company through the back door, with personal accounts and company data pasted in. The first useful step, then, is not adding a tool but setting a rule: which edition is used, for which information, under which contractual terms, and in which cases it is not used at all.
Then the setup is sized and implemented: a plan suited to the number of users and the type of data, access tied to company identity, assistants configured on the right documents, possible API integration with ERP, document management or CRM, and training for people, because a tool used badly creates more work, not less. From then on we stay in operation: monitoring real usage, revising instructions when answers are not convincing, adapting when models and terms change, ongoing support with one contractual point of contact.
ChatGPT is part of Artificial intelligence. See also the Aesir method and our success stories.
Want to know whether ChatGPT makes sense for your company?
We start from the process, not the product. A first meeting with no commitment.