AI Act Readiness Starts With a Plain Inventory of Tools Touching Customers

Inventory before policy

Many companies write AI principles before they know where AI sits in the business. That order creates a nice document and a weak control. A seller using AI in customer support, listing translation, fraud review or pricing needs a plain inventory first.

The inventory should not be technical theatre. It should answer who uses the tool, what data it sees, what output it creates and whether a human checks the result before customers feel it. A spreadsheet can do the job if it has owners and review dates.

Case pattern: the support shortcut

A seller adds an AI writing assistant to speed customer replies. Agents use it for returns, warranty questions and product-use advice. The tool saves time until one reply gives advice that does not match the manual. The issue is not that AI wrote the sentence. The issue is that nobody mapped which topics require human review.

An inventory would have shown that support messages touch warranty, safety and product claims. Those topics need approved answer banks or escalation rules. AI can assist, but it should not invent product instructions from a weak file.

Controls that fit small teams

Small teams should focus on customer impact. If a tool only summarizes internal notes, the risk differs from a tool that sends messages or changes listing claims. The inventory should separate those uses and set review rules for each.

Review the inventory after adding a tool, changing data access or moving AI output closer to customers. The owner should sign off the workflow, not just the software purchase.

• List AI tools used in support, listings, pricing and fraud review.
• Record what data each tool can see.
• Name a human owner for each workflow.
• Require review for safety, warranty and legal claims.
• Update the inventory when tools or workflows change.

Field review

A practical review starts with one live product, one active order and one current customer-facing page. Put those records beside the article topic and ask whether they still describe the same business reality. If the public page, the supplier file and the internal decision record point to different answers, the team has found the gap that will matter during a platform review, customs question or customer dispute.

The review should produce a small decision note. It should name the file owner, the missing evidence, the business action and the date for the next check. That note matters because cross-border teams change quickly. A future reviewer should be able to see why the business accepted, corrected, paused or escalated the issue without searching private messages.

Use the same test after the next supplier change, route change, campaign launch, listing edit or complaint pattern. The point is not to create a larger archive. The point is to keep the commercial record current while the business keeps moving. A file that was true last quarter can become misleading after one product edit or fulfilment change.

A good checkpoint is whether a new employee could open the folder and answer the main question in ten minutes. If the answer depends on one veteran employee, a chat thread or a supplier promise that nobody saved, the record is too fragile for a fast-moving marketplace or border process.

That simple test keeps the article grounded in operations, not theory.

The handoff should also say what the team will not claim until evidence improves. Clear limits protect the business as much as strong proof does. When a record is partial, say which market, product version, route or customer promise it can support, and which one it cannot support yet.

That boundary should be visible to sales, support and finance.

If those teams cannot see the boundary, the next public promise will drift again.

For recurring risks, sample one file each month and record whether the boundary still holds. A small monthly sample often catches drift faster than a large annual review because it follows the way sellers actually change products, routes and campaigns.

Keep that sample note with the live file.

Closing note

AI Act readiness can feel abstract. The first useful step is concrete: know which tools touch customers and decisions.

A simple inventory gives teams a map. Without that map, policies stay detached from the workflows that create risk.

Reference links

• European Commission AI Act