Data classification is the decision every other protection depends on. Encryption, access control, retention, and sharing rules all need to know how sensitive a piece of data is before they can act. Keep the scheme small (four tiers), attach a fixed set of handling rules to each tier, and make labeling the default rather than an optional extra.
Why classification comes first
Without a classification, every control is applied evenly or guessed. Evenly applied controls are expensive and slow people down. Guessed controls leave sensitive data under-protected. A clear scheme lets you spend effort where the risk is. ISO/IEC 27001:2022 treats this as a baseline requirement, with controls for classifying information (5.12) and labeling it (5.13), and NIST FIPS 199 provides the underlying idea of rating impact as low, moderate, or high.
A four-tier model that works
| Tier | Definition | Typical examples | Impact if disclosed |
|---|---|---|---|
| Public | Approved for anyone | Published reports, marketing material | None |
| Internal | For staff, low harm if leaked | Org charts, internal procedures | Low |
| Confidential | Business-sensitive or personal data | Customer records, contracts, financials | Moderate to high |
| Restricted | Severe harm or legal exposure if leaked | Credentials, regulated personal data, trade secrets | Severe |
Implementation in seven steps
- Scope and inventory. Start with the systems and repositories that hold the most data, such as file shares, collaboration tools, databases, and cloud storage. You cannot classify what you have not found.
- Define tiers and handling rules. Write one page: the four tiers, a plain-language definition, three examples each, and the minimum controls. If it does not fit on a page, it will not be used.
- Assign data owners. A named business owner decides the tier for each data set. Security advises; the business decides.
- Label. Apply labels automatically where patterns are reliable (for example national ID numbers or payment card numbers), and ask authors to confirm or raise the label manually. Default new content to Internal rather than leaving it unlabeled.
- Bind controls to labels. Link each label to encryption, access policy, external sharing rules, retention, and monitoring. A label that changes nothing is decoration.
- Train and communicate. A 20-minute walkthrough with real examples from each team does more than a policy document.
- Review. Reclassify when data changes purpose, ages, or is combined with other data. Review the scheme itself once a year.
A worked example
A regional retailer finds that a shared folder holds supplier contracts, customer complaint exports, and marketing images together. The contracts are rated Confidential, the complaint exports contain names and phone numbers and are rated Confidential, and the marketing images are rated Public after review. The folder is split by tier, the Confidential folders are limited to named groups and encrypted, and the sharing link feature is disabled for them. One restructure produces three different protection levels from the same storage.
Common pitfalls
- Too many tiers. Beyond four or five, people guess, and consistency falls.
- Classifying once and never revisiting. Data changes in sensitivity over time.
- Relying on users alone. Pair manual labels with automatic detection.
- Labels without enforcement. Every tier needs at least one control that actually changes.
Implementation checklist
- One-page scheme approved by the data governance forum
- Owners named for the ten most important data sets
- Default label set for new content
- Automatic detection rules tested on sample data
- Encryption, access, and sharing rules linked to each label
- Training delivered and a feedback route published
- Coverage and mislabeling measured monthly
How to measure progress
Track the share of in-scope repositories that have been scanned, the share of items carrying a label, the number of Restricted items found in locations not approved for them, and the time taken to fix each such finding. Coverage and fix time tell you more than a total count of labeled files.
Classification schemes often stall when they are owned only by the security team. Pair a security lead with a business data owner for each domain from the start, so that tier decisions have both context and authority.
Interactive: tier explorer
Select a tier to see the minimum controls it carries. Higher tiers add to the ones below.
- Access
- Need-to-know groups
- Encryption
- At rest and in transit
- Sharing
- Approved recipients only
- Retention
- Defined period, yearly review
- Monitoring
- Access logs reviewed
References and further reading
- FIPS 199: Standards for Security Categorization of Federal Information and Information Systems (NIST)Accessed November 7, 2026
- SP 800-60 Vol. 1 Rev. 1: Guide for Mapping Types of Information and Information Systems to Security Categories (NIST)Accessed November 7, 2026
- ISO/IEC 27001:2022 Information security management systems (ISO)Accessed November 7, 2026
- Sensitivity labels in Microsoft Purview (Microsoft Learn)Accessed November 7, 2026