What is Enterprise AI? An assistant that talks to your company data

On Monday morning, before the management meeting, everyone asks the same questions: how much did we sell last week, where are we against target, which shipments are late? The answers sit on an ERP screen, in an Excel file and in the memory of a few people. Enterprise AI is an assistant to which you can put questions about the company’s own data, by voice or in writing, and which gives the answer as a short reply together with a screen built for that question.

In this article we explain the concept through YazSis’s Enterprise AI assistant, Fero: what it is, what it is not, how it works in the background and which data source it can work with today.

What is enterprise AI not?

• It is not a general-purpose chatbot. It deals not with the world’s general knowledge but with your company’s data; the questions are about business topics such as sales, budget, stock, orders and shipments.
• It is not a menu of ready-made reports. Instead of choosing from a fixed list of reports, you ask the question in everyday language; the screen is built around that question.
• It is not a tool that changes data. Fero works read-only. It connects to the database only with an account that has read permission, and data cannot be changed.
• It is not a model that makes up figures. Calculations such as percentages and differences are done by the system, not by the AI; the model cannot state a number that is not in the tool result.

What happens in the background when a question is asked?

1. You ask. By voice or in writing, in everyday language: “How are our sales today?”
2. Fero understands the question. It works out the intent, the period and whom or what you mean.
3. Safe business tools run. Data is reached only through 17 defined, read-only business tools. The AI does not write SQL, and no arbitrary query is run.
4. The data is calculated. Indicators, percentages and differences are calculated in the back end.
5. The answer and the dashboard arrive. A short reply and indicators specific to the question come to you together. The screen is built from 13 approved dashboard components; the AI does not generate code.

A good enterprise AI should be the one that gives the right number before it is the one that speaks fluently.

In a classic dashboard someone designs the screen in advance, and the manager tries to find what they are looking for among ready-made charts. If they cannot find it, a report request is opened and an answer is awaited. In enterprise AI the order is reversed: the question comes first and the screen is built around it. This difference shows up especially when you deepen a “why did it drop?” question step by step with “which sales rep, which customer, which period?”.

What can be asked?

• Sales: “Where are we against this month’s sales target?”, “Show the sales trend for the last six months.”
• Customers and sales reps: “Which customers do we sell most to?”, “Which sales reps are below target this week?”
• Budget: “Which customers are below budget?”
• Shipments and stock: “Which loadings are late?”, “Which products are at critical stock?”
• Executive summary: “Is there anything I should pay attention to today?”

Follow-up questions keep the context. When you say “show it by sales rep” or “look at the last three months”, the screen is not reset; it is updated on the same subject or drills into detail. In the sample flow on the site, you first ask which sales reps are below target, then drill into one rep’s performance and see that the drop comes from two customers. Fero can also point out what needs attention on its own: when an executive summary is requested in the sample data, it brings forward the late loadings, the critical stock and the customer below budget.

Which data does it work with? The honest position

The data source layer is designed with an adapter architecture: when a new source is added, the business tools and screens do not change. The source that is ready today is MySQL. Sources such as SQL Server, PostgreSQL, REST API, Logo, SAP, Odoo and Excel are handled within a project; each of them requires separate work and is not offered as a “ready-made connection”. The demo screens on the site are representative and prepared with sample data; the language and interface are adapted to the project.

Who is it meaningful for?

Enterprise AI is meaningful for roles whose decisions rest on numbers but who depend on someone else to reach those numbers: the general manager, sales and operations managers, finance and budget owners. The difference is most visible in companies whose data is gathered in a database but where nobody can spare time for reporting. If the data is scattered or of poor quality, the assistant does not fix it; faulty data means a faulty answer. That is why you need to start by looking at the state of the data.

Where to start?

The best place to start is the questions managers ask again and again every week. In which table do the answers to these questions sit, who prepares them, and how long does it take? Once you have drawn up the list, it also becomes clear which data source and which business tools are needed. Starting with a small set of questions and widening the scope makes it easier both to check accuracy and for the team to get used to it. In the first stage, check whether the answers match the figures you know exactly; this comparison is the foundation of the trust you will place in the assistant.

You can look at the assistant’s screens and way of working on the Enterprise AI page and request a live demo.