Manufacturing: classification and lifecycle management as the foundation for Copilot

A vehicle engine manufacturer with tens of thousands of employees across more than a hundred countries had to retain some data for decades under local regulations, while most of that data was unstructured: Word and PowerPoint files, Teams chats, emails, videos. That kind of data is hard to classify, and an earlier approach of static labels and manually moving files into an archive made collaboration harder rather than easier.

The organization automated classification and sensitivity labeling with Microsoft Purview Information Protection, using hundreds of prebuilt rules per jurisdiction to recognize sensitive information. Labels are persistent and travel with a file even outside its original environment, so employees can keep working in their everyday apps instead of moving files into a separate archive. Data Lifecycle Management adds customized retention periods per file type, so outdated data gets cleaned up automatically. This classification also forms the foundation for a Microsoft 365 Copilot pilot: generated Copilot responses inherit the classification level of the underlying data, so a response touching confidential information gets flagged as such too.

The organization did not want to wait on Copilot until every piece of governance was perfect, but also did not want to start blind. By automating classification and lifecycle management first, it built a clean, controlled data foundation for the Copilot pilot to run on, without sensitive information slipping unnoticed into AI responses.

Third-party classification tools were considered, but proved less flexible at plugging into existing DLP and label policy within the same Microsoft 365 environment. Sticking with the existing static labels and a central archive did not solve the scale problem: too much data was kept too long, driving up both storage costs and the complexity of eDiscovery.

• Automatic classification covers the large majority of files, but still depends on the quality of the prebuilt rules per jurisdiction.
• Persistent labels let employees keep working in their own apps, which is more pleasant than a central archive, but requires training so labels get applied consistently.
• Copilot responses that inherit the source's classification level only work reliably once that underlying classification is already in place.

Microsoft Purview (Information Protection, Data Lifecycle Management), Microsoft 365 Copilot.

• Automate classification and lifecycle management before a Copilot rollout, not as a reaction to an incident afterward.
• Use persistent labels that travel with the file, so employees can keep working in their existing apps.
• Treat retention as an ongoing process: tie automatic cleanup to the type of data instead of one-off manual actions.

The biggest saving did not come from tighter security, but from the cleanup itself: automatic retention periods substantially reduced the number of outdated files, cutting both storage costs and the amount of data a Copilot query or an eDiscovery process had to search through. A smaller, cleaner data estate turned out to be more valuable than extra classification rules layered on top of a system full of stale data.