VASCOM TECHNOLOGY AUTOMATION CO., LTD
Home › News › Five Questions to Ask Before Buying an AI…

Five Questions to Ask Before Buying an AI Tool

Learn how to assess an AI tool’s data handling, accountability, scaling costs, and data export options so you can identify risks and retain control before signing a contract.

An impressive demo is not enough to decide on buying an AI tool. Once it enters daily work, the tool will touch internal data, access rights, and approval processes. Evaluation should therefore start from real usage situations and from the work the tool will actually support, rather than from the features shown in a polished presentation. The five questions below help a business clarify risk, cost, and room to stay in control before signing a contract.

Where is the data processed, and who can see it?

This question does not stop at where the server is located. Data can pass through several components: uploaded files, conversation content, records from management software, AI processing services, and backup systems. The supplier should describe the data flow in writing, covering each step, rather than replying only that data is kept secure.

  • Processing and storage location: which region the data sits in, and whether it passes through any third-party providers.
  • Access rights: which supplier staff can view the data, under what circumstances, and whether those accesses are logged.
  • Purpose of use: whether data is used to train or improve models, and whether opting out is done through a setting or a separate clause.
  • Retention period: how long conversation history, attachments, and backups are kept after a user deletes them.

During the trial, use an account with limited rights and check the sharing, permission, and deletion settings directly. For sensitive information, state clearly in the internal rules which data types must never be entered into the tool, and check that the setting actually blocks them.

When the tool produces a wrong result, who is responsible?

Do not ask only whether the AI is accurate. Present a concrete situation rather than a general one: the tool creates a wrong order, suggests a wrong stock level, or sends the wrong content to a customer, so what is the handling process and who is notified first? From there, determine whether the tool only offers suggestions or is allowed to perform actions on systems by itself.

  • Define which tasks must always have a person check before confirmation.
  • Require logging so that input data, the AI result, and the user’s final action can be reviewed.
  • Clarify the supplier’s responsibility when the fault comes from the system, a model change, or an integration with other software.

For actions that can create financial obligations or change important data, keep a manual approval step. Support terms, incident reporting, and limits of liability also need to be read together with the internal operating process.

How do costs change as the number of users grows?

The price on a quotation rarely reflects the full operating cost. Build a scenario close to reality: which departments will use it, how many people in each, how often they run a task, what file volumes are involved, and which systems must connect. Then ask the supplier to price that exact scenario.

  • Fees per user, per usage event, or per volume of data processed.
  • Setup, integration, training, and account administration costs.
  • Charges when limits are exceeded, storage is added, or advanced features are used.
  • Conditions for price changes, renewals, and minimum number of accounts.

Also check whether the tool provides a usage dashboard and limit alerts. If usage cannot be controlled from the start, expanding to more departments can make costs hard to forecast.

Can the data be exported and the service stopped without losing data?

Export capability needs to be tested with a real action, not just read in the documentation. An export file must retain enough context for the business to read, check, or move to another system. Being able to download only what appears on screen usually does not meet this need.

  • Check the export format, the data structure, and whether attachments can be downloaded.
  • Determine which data is not included in the export: edit history, logs, access rights, or workflow configuration.
  • Ask clearly how long data can be accessed and exported after the contract ends.
  • Clarify how data is deleted from the system, backups, and related accounts.
  • Prepare a plan for continuing the work if the tool stops operating, including the person responsible and where data will be received.

Before a wide rollout, practice the exit process once: export the data, check completeness, revoke access, and disconnect integrations. If this step cannot be completed, the business should reconsider how much data goes into the tool.

What to do now

Start with a small evaluation that produces documents and concrete test results.

  1. Choose a real process for the trial, with sample data and a person responsible for checking the results.
  2. Send the supplier the set of questions on data, errors, costs, export, and contract termination, and ask for written answers.
  3. Export all data to a separate file and check whether any fields are missing.

The five questions above do not need perfect answers from the start, but they do need written answers before a contract is signed. A suitable tool is one the business knows how to stop using when necessary.

Tips and how-to Talk to VASCOM
More news

Keep reading

Zalo ☎
VASCOM
Call 0822.577.579 Get a consultation