What makes an engineering archive useful for AI training
Records that follow a piece of work through its steps and show how it ended.
Typical candidates are email archives, order and project files, quotes and costings, scanned folders, complaint and credit note records, inspection and maintenance records, ticket and CRM histories, technical drawings, CAD models, change requests and commissioning records.
What usually stays with the company
Current designs, current prices and anything a competitor could use stay where they are. Records that belong to customers or suppliers, such as their drawings, are considered only if the company holds the rights to them and no confidentiality agreement stands in the way.
Who has to approve
Someone who can grant the rights to the records, which usually means company management, informed by legal counsel and the data protection officer. Records that contain personal data of employees or customers raise further questions, and in some countries employee representatives have a say. The details differ by country and by company, and this page is not legal advice.
How an initial review works
After a confidentiality and data-processing agreement, a sample is reviewed together with the company’s rights to it and its suitability for AI customers. The company then sets out scope, purposes of use, recipients and payment in writing. Records are prepared, anonymized and checked, and payment follows the milestones agreed. No file is transferred before a signed agreement is in place.
Data partnerships →What control the company keeps
The company decides what goes out, for which purposes and to whom. A company that would rather keep its records has another route: a private model trained on them, for its own use.
A private model trained on your own records →What anonymization leaves open
Replacing names, addresses and identifiers makes records safer to use, and it is one step in a review. A drawing can be confidential without naming anyone, and a customer’s design does not become the company’s to license because the customer’s name was removed. Those questions are settled in the review and in the contract.
What has to be clear before compensation is discussed
Four things come first: which records are in scope, whether the company can grant the rights to them, whether the sample is useful to AI customers, and the terms under which they may be used and by whom. Payment can then be a one-off price or a share of what is earned from licensing, and it follows milestones agreed in writing.
Starting without sharing files
A first conversation needs a description and no files: the formats, the rough volume, the years covered and the subject matter. Who holds the rights? Does the material include customer or supplier records? Does it contain personal data? Which records still give the company an edge?