A cloud dashboard synchronizes secure documents and data between two storage services.
Microsoft Purview's on-demand classification currently runs only when an administrator starts a scan. Microsoft has now put an item on its roadmap to change that. Roadmap ID 579194, titled "Microsoft Purview: recurring classifier refresh," would turn on-demand classification into recurring classification refreshes. The stated goal is to keep classification of existing content current as data and context change. The item is marked In development. Microsoft lists a preview for February 2027 and general availability for April 2027.

Scheduling a classification scan sounds minor. For anyone responsible for data loss prevention, sensitivity labels, retention or insider risk in Microsoft 365, it matters more than that. Those controls are only as accurate as the classification data beneath them, and that data goes stale between manual scans.

A cloud dashboard synchronizes secure documents and data between two storage services. What the roadmap item says​

The roadmap entry contains these details:

  • Roadmap ID: 579194
  • Product: Microsoft Purview
  • Status: In development
  • Platform: Web
  • Release rings: Preview and General Availability
  • Cloud instance: Worldwide (Standard Multi-Tenant) only
  • Preview: February 2027
  • General availability: April 2027

Microsoft's roadmap page warns that its release dates are estimates and that all information is subject to change. Microsoft's admin documentation for the roadmap adds that a feature keeps the "In development" status during preview because preview features are still being validated and may change before production. The roadmap item lists no GCC, GCC High or DoD availability.

The entry leaves several important questions unanswered. It does not say:

  • how often refreshes would run
  • whether administrators can set the schedule
  • which locations or classifier types are covered
  • whether recurring runs are on by default
  • how billing, notifications and policy actions would work

Anyone who says these details are settled is guessing.

In short: this is a planned change on an estimated timeline, available only in the worldwide commercial cloud for now. The direction is clear, but the controls have not been described.

Why "admin-initiated" is the key word​

To see what changes, start with how on-demand classification works today. Microsoft Learn says Microsoft Purview on-demand classification identifies and classifies sensitive content in historical data stored in SharePoint, OneDrive, and endpoints. It covers files that aren't classified or modified for a long time, files that are never classified, or files that need updated classification on previously classified files.

It does not run continuously. Microsoft's endpoint FAQ says on-demand classification runs only when an admin starts a scan. The endpoint getting-started guide describes the split: Endpoint DLP classifies files in real time when they're accessed, modified, or shared. On-demand classification extends that coverage to inactive and historical files at rest that Endpoint DLP hasn't processed.

Real-time classification handles files people are actively using. Files that nobody opens, such as an old contracts folder from 2019, stay as they were last classified, or unclassified, until an administrator runs a scan. Suppose you add a new sensitive information type, tighten the confidence levels on a custom classifier, or retrain a trainable classifier. Unless someone remembers to rescan, those dormant files are still judged by the old rules.

Practitioners have pointed this out for a while. Consultant Joanne Klein noted that at-rest content must be classified before service-side Purview controls can consume the classification details as a condition to take action. She advised that teams changing custom classifiers should use on-demand classification to reclassify historical content. A recurring refresh would make that a scheduled task instead of something an administrator has to remember.

In short: today, keeping dormant content correctly classified depends on an administrator starting a scan. Roadmap 579194 is meant to automate that.

What a refresh would feed into​

Classification itself does not enforce anything. Microsoft's Learn article says on-demand classification updates classification results, and matching policies decide what happens next. For SharePoint and OneDrive, updated results can trigger active policies, including policies that apply or remove sensitivity labels. Microsoft also says on-demand classification is not an auto-labeling simulation. Once a file is scanned, it is evaluated against DLP, Information Protection, Data Lifecycle Management and Insider Risk Management policies.

The endpoint documentation gives more detail. It says on-demand classification identifies and classifies sensitive content but doesn't apply sensitivity labels. To apply labels based on the results, configure auto-labeling policies. It also notes that classification results feed into policies, but enforcement happens through Endpoint DLP when files are accessed or shared.

Analysis: the effect of a recurring refresh will depend on which policies are active when it runs. A scheduled refresh in a tenant with aggressive auto-labeling or retention policies could change a lot more than one in a tenant that only reports. Microsoft also notes that correct classifications don't need refreshing every time an auto-labeling policy changes. A policy change and a classifier change are different events, and they call for different responses.

Today's limits​

Microsoft has not said whether the recurring version keeps these limits. They are still the best guide to what scheduled refreshes would involve. According to Microsoft Learn:

AreaCurrent behavior
Default scopeItems created or modified in the past year, all supported file extensions, all classifiers configured in the tenant
Classifier selectionUp to 50 classifiers when choosing specific ones
Scan limit (SharePoint/OneDrive)150,000 locations and 100 million files per scan, based on estimation
Overlapping scansA location in several active scans is processed by one and marked skipped in the others
Estimation gapClassification can start up to 30 days after estimation; a shorter gap gives more accurate counts and costs
Content Explorer lagUp to seven days to show newly classified content
Partial classifier scansOnly the selected classifiers are updated; others may stay unevaluated and may not trigger their policies

On Windows 10 and 11 endpoints, Microsoft requires at least four virtual processors and 4,096 MB of memory. Each device gets 2 GB of data discovery bandwidth per rolling 24 hours, and scanning pauses when it reaches that cap. On the device side, classification runs on the device. Only classification results and metadata are sent to the Microsoft Purview service.

The partial-classifier row deserves attention. If a recurring refresh only covers a subset of classifiers, files could still have out-of-date results for the classifiers that weren't selected. That would be a problem if anyone assumes the refresh covers everything.

Cost​

Microsoft's endpoint guidance says on-demand classification uses a pay-as-you-go billing model. Usage is reported in the Microsoft Purview Usage Center, and charges appear on your Azure subscription bill. The main Learn article says the feature can use pay-as-you-go billing, per-user licensing or both, depending on the Purview capabilities involved.

Cost is decided by files, not classifiers. Microsoft says scan cost depends on the total number of files scanned, but choosing more classifiers can bring more files into scope. Microsoft also warns that tenants using EDM classifiers should be aware that On-Demand Classification scans (including estimation and classification) rescan previously classified files, which can increase overall scan costs. Excluding the EDM classifier reduces that cost.

Analysis: if the schedule can't be controlled, a recurring refresh could mean a recurring bill. That is the first thing finance teams will want to know once Microsoft publishes the settings. Some help may come from result reuse. Microsoft's endpoint best-practices page says previously classified results apply across: on-demand → on-demand, on-demand → Endpoint DLP, and Endpoint DLP → on-demand. To reuse Endpoint DLP results, advanced classification must be enabled on the DLP policy. The same page says reuse also requires unchanged file content and classifier definitions, a classifier set that matches or is a subset of the earlier scan's, and up-to-date Windows cumulative updates on the device. Microsoft has not confirmed whether recurring refreshes will reuse results this way.

What to do before February 2027​

Recurring refresh isn't available yet. Until then, the documented manual process is still the way to do this:

  1. Run full cycles. Microsoft advises completing each scan from estimation to classification, and checking that a device's previous scan has reached Classification Complete, Classification Cancelled, or Estimation Cancelled state before adding it to a new one.
  2. Rescan after classifier changes, on a schedule. Microsoft's endpoint guidance recommends monthly or quarterly scans when classifiers or policies change, because unchanged files that were already classified aren't normally rescanned.
  3. Pilot first. For large endpoint deployments, Microsoft suggests starting with 10–20 users, then 50–100, and splitting large tenants into batches.
  4. Narrow your scope. Adaptive scopes can select SharePoint sites by URL, name or metadata. Administrative units can restrict scans and results to delegated boundaries. The consultancy Cyclotron has made the same point: leverage date and time ranges or targeting specific sites, to manage costs and efficiency.
  5. Check permissions. Running scans requires the Compliance Administrator role group. Viewing results requires Content Explorer Content Viewer or Content Explorer List Viewer.
  6. Follow roadmap ID 579194. Microsoft's roadmap supports RSS feeds for individual items, so you can see when the status or dates change.

Possible benefits and concerns​

The benefit is clear. Classification data goes stale over time, and that hurts any organization relying on DLP or on keeping AI tools away from mislabeled files. Microsoft itself lists reducing the risk of AI tools surfacing unlabeled information as a reason to run on-demand classification. Running refreshes on a schedule would also mean less depends on an administrator remembering.

There are also open questions. Automatic scanning that is billed per file needs strong spending controls. A scheduled refresh in a tenant with active labeling policies could also change labels without anyone watching. Regulated customers in government clouds aren't included yet.

Bottom line: Purview administrators should note the February 2027 preview, keep running scans manually for now, and wait for Microsoft to explain how the schedule, scope and billing will work before planning around this feature.

 

References

  1. Microsoft Purview: recurring classifier refresh Microsoft 365 Roadmap 2026-11-17T19:05:12.369378Z
  2. Understanding Microsoft Purview's On-Demand Classification - Cyclotron cyclotron.com
  3. Microsoft Copilot Roadmap | Microsoft 365 microsoft.com