What to ask an AI vendor about customer drawings

Can a customer drawing go into an AI tool?

A drawing you receive from a customer remains theirs. It carries dimensions, tolerances and sometimes a design a competitor would like to have. That is why it often arrives only after an NDA, and the NDA decides what you may do with it, not the AI tool.

Read what the agreement says about sharing with third parties, about processing outside your own business and about what happens to the files when the job ends. A software vendor counts as a third party here. If the agreement is silent, ask your customer before you upload.

What AI does for the cost engineer and the manufacturing engineer is covered in what AI means for a precision machine shop. This page is about the files: the STEP model and the PDF drawing that make up the request.

Are other firms wary of AI and data too?

Yes, and over the same points. Among EU enterprises with ten or more people employed that had considered AI but were not using it in 2025, the reason given most often was a lack of relevant expertise (71 per cent). Unclear legal consequences (53 per cent) and concerns about data protection and privacy (49 per cent) came next, according to Eurostat. Those figures describe firms that held back, not all firms.

The Dutch picture is similar. In provisional 2025 figures from CBS, Statistics Netherlands, covering firms with two or more people employed, those that had considered AI but did not use it cited lack of experience (73 per cent), privacy (49 per cent) and legal consequences (42 per cent). In Dutch industry, 16 per cent of firms used AI that year.

Two more points apply to small manufacturers. The Productiviteitsagenda 2026-2028 of Smart Industry, the Dutch manufacturing programme, calls AI that sends large volumes of data to the cloud often too risky for a small firm, citing data security, privacy and trade secrets. The OECD wrote in 2025 that small firms, especially in manufacturing, struggle to find AI vendors suited to their needs, and that a lack of good data and digital readiness holds them back.

What do you ask about your files and data?

Put these questions to every vendor, us included, and get the answers in the contract, not just on a website. A good answer gives a place, a period or a name.

It is not only about drawings. Estimating software also holds your hourly rate for each machine, your own purchase price for each material and your cost engineers' corrections. A market reference from Eurostat or the ECB is public; your own purchase price and your rates are not. And ask whether that data is used for your business alone.

• Where are the files kept: with which company, in which country, and which other parties process them?
• Who can open them: which people at the vendor, and how do you control who on your own team can open which customer's files?
• What are drawings and outputs used for: only your estimate, or anything else, such as training a model?
• How long are they kept, and how do you delete one drawing, one customer or everything?
• What do you take with you if you leave: which export, in which format, and what happens to the vendor's copy afterwards?
• Who changed what: is there a correction history with names and dates, and can you still open it later?

How do you keep drawings out of personal accounts?

A big risk lies in software nobody chose: a general chat assistant on a personal account, where someone on a busy afternoon pastes in a drawing to look up a material or a tolerance. The terms of that account apply there, not your agreement with the customer.

So agree one rule for the business: customer files go only into tools the business has chosen, under a business account. Write down which tools those are and who to ask when in doubt, and go through it with everyone who joins.

A rule like that only holds when the reason is clear to everyone: the drawing belongs to the customer, and a quick answer from a personal account can cost you an order.

What does the AI Act ask of a machine shop?

According to the European Commission, the EU AI Act entered into force on 1 August 2024 and has applied since 2 August 2026, with exceptions. The rules on AI literacy have applied since 2 February 2025.

Those rules sit in Article 4. Providers of AI systems, and businesses that deploy them, take measures for the AI literacy of their staff and of others who operate those systems on their behalf. Their knowledge and experience count, and so does the context of use.

A fixed rule on customer files, and on who confirms an output, fits that duty. Which other rules apply depends on what you use AI for. For that, read the text itself or ask the European Commission's AI Act Single Information Platform. This is not legal advice.

Who keeps control of the output?

Control is also about who decides. A sound way of working keeps a human-in-the-loop: software proposes a material, a tolerance or a time, and someone on your team confirms or corrects those proposals before they count in the price.

Two common ways of working fail at this point: the AI pilot that returns a price with no provenance, and the spreadsheet whose corrections only one cost engineer can follow. When a customer asks why a repeat part costs more this time, you point to the line that made the difference: a tighter tolerance or a higher purchase price. If you cannot, you either give a discount to keep the order or pad the price to cover your own uncertainty. The first costs margin, the second costs orders.

So ask: can you trace the provenance of every figure, back to the drawing, your catalogue or a colleague? Does a correction call for a reason, and does that reason stay once the cost engineer who gave it has left? More in holding on to a senior cost engineer's know-how.

How far to trust an estimate made with AI help is the subject of how accurate an AI estimate is. The full checklist for comparing software is in choosing estimating software for machining.