7 questions to ask when choosing AI software

Enterprise AI software is software that answers questions on a company’s own data, prepares reports or runs a process, with its output bound by rules of permission, accuracy and oversight. This definition matters, because a general-purpose chat tool and a system that touches your company data are not at the same level of risk.

Demos look alike: a fluent answer, a sleek screen. The difference is made on the part of the demo you do not see. The seven questions below are a framework you can take to an evaluation meeting, whichever vendor you work with.

Seven questions to ask the vendor

1. Which business question will it answer? “Let’s use AI” is not a goal. The goal is a question with a clear decision point, for example “where are we against this month’s sales target?” Choosing a product before the question is clear is premature, because which data is needed, and with what permission, follows from the question.
2. Which data sources does it connect to, and which is ready today? Assume that not every source on the vendor’s list is equally mature. Ask the vendor to separate what is ready from what will be set up within the project scope.
3. How does it access the data? Is it read-only, or can it change the data? Does the AI write free-form queries, or do only defined tools run?
4. Who calculates the number? A language model speaks fluently, but if it produces numbers it can be wrong. Look for a structure in which the system calculates percentages and totals and the model merely relays them. This distinction determines the level of assurance with which a number used in a management meeting reaches the table.
5. Who sees what, and where does the data go? Can permissions be defined per user? Ask in writing which services the data passes through and where it is processed; an answer of “it is secure” is not enough.
6. How is a wrong answer noticed? Find out who will check the output, at which step human approval comes in, and how quality will be measured over time. If no method is offered for verifying AI output, the risk stays entirely with you.
7. Is it stated clearly what is ready and what is a project? This question shows the vendor’s honesty more than any other. It is hard to work for long with a vendor who describes a feature that is not ready as if it were; an answer of “within the project scope” is not bad news, it is a clear answer.

Three signs to watch for during the demo

• Is the data sample or real? A demo with sample data shows the product’s interface, not the problems in your data. If possible, ask for a trial on a source similar to your own data structure.
• Is the answer consistent when the same question is asked in different words? Managers do not ask questions in the set phrases of a demo. “How are sales?” and “how much did we sell today?” should give the same result.
• Can you ask where the number comes from? If you cannot trace which data and which calculation a figure came from, it becomes hard to make decisions with that figure.

How does YazSis answer these questions?

Enterprise AI (Fero) is the assistant that lets you ask questions, by voice or in writing, about your sales, budget, shipment, order and stock data. The answers to the seven questions are as follows:

• Business question: manager questions such as the sales target, representatives below target, budget status, delayed shipments and critical stock.
• Source: the source that is ready today is MySQL. SQL Server, PostgreSQL, REST API, Logo, SAP, Odoo and Excel are handled within the project scope.
• Access: it connects to the application database with an account that has read-only permission, so data cannot be changed. The AI does not write SQL; only defined business tools run.
• Numbers: KPIs, percentages and differences are calculated in the back end; the model cannot state a number that is not in a tool result.
• Screens: the dashboard is built only from approved components; the AI does not generate code.

Fero is not a chatbot or a ready-made report screen: every question turns into an answer and a screen specific to that question. Follow-up questions keep the context; when you say “show the top five” or “show by representative”, the screen is not reset but updated on the same topic or drilled down into detail.

If your AI need is not yet clear

For some companies the real question is not choosing a product but deciding where AI will create value. Not every process needs AI; sometimes a regular report or a simple workflow is enough. If the use case is not yet clear, our approach has four steps:

• Limit the data: define the right data source and the access permission.
• Validate the use: match the opportunity to a real business problem and decision point.
• Connect it to the flow: tie the output to an existing workflow, approval or report.
• Monitor and improve: measure quality and keep human oversight and the feedback loop.

The use case is also chosen within this framework: an enterprise knowledge assistant, document analysis, controlled agent structures, smart reporting, anomaly detection or workflow automation. Which one will create value is determined by evaluating access, data quality, auditability and business outcome together.

A good AI demo shows how fluent the answer is; a good AI system shows why the number is right.

Take these seven questions to the evaluation meeting in writing and note the answer given to each. For details about Fero, see Enterprise AI; for use-case discovery, see YazSis AI.