Cloud & Email

DLP policy not behaving as expected? Check the scope, rule and actual action.

Detection, notifications and blocking are different outcomes. Define the intended information-handling rule, confirm its real coverage and investigate with authorised harmless examples while keeping existing protections in place.

16 October 2026  •  8 min read

Two adults discussing plain colour cards beside a closed tablet folio in a ceramics showroom

For business data owners and authorised administrators

Use the actual organisation, workload and supported policy capabilities. Privacy/legal applicability, evidence access and changes need the responsible authority; a regulatory template is not a compliance certificate.

Step 1 of 9

In this guide
  1. 01Intent, ownership and actual coverage
  2. 02Conditions, effective actions and evidence
  3. 03Pilot, verification and help

Describe the data, action and expected result

“DLP is not working” needs a specific policy outcome to investigate.

Write a non-secret description of the intended rule: what information should be protected, who is acting, which service or device is involved, what activity should trigger a response and what should happen. For example, an approved test item being shared to an approved test recipient is different from merely opening a file.

Record the observed result, time/time zone, relevant app/service and whether you expected detection, a notification, an alert, restriction or an allowed override. These are different outcomes. A missing policy tip does not by itself prove the item was not evaluated, and a recorded match does not prove the intended activity was blocked.

If real information may already have been disclosed, follow the responsible team’s incident process promptly. Do not recreate the incident by sending customer records, removing labels or using a personal account. Preserve relevant evidence privately and use only authorised harmless test material for a controlled investigation.

Define the exact information/activity/scope and intended action. Separate detection, notification and enforcement; possible real exposure needs incident response rather than another live leak test.

Next: Agree requirements, ownership and the policy boundary

Agree requirements, ownership and the policy boundary

A template name does not determine the business’s legal obligations.

Identify the business/data owner, authorised policy administrator and appropriate privacy/legal advisers. Agree which information and legitimate workflows need protection, which behaviour is restricted and who handles exceptions and alerts. Microsoft’s DLP planning guidance puts stakeholders, data categories and business processes before broad deployment.

Distinguish data loss prevention from sign-in controls, file permissions, sensitivity labels, retention, legal holds and backups. These can interact but serve different purposes. A DLP rule does not automatically grant the right people access, retain every deleted record or recover a missing file.

For a security/compliance configuration review, obtain the actual requirement and its scope instead of apply every template. Use the security and standards review guide to keep evidence and assurance claims bounded. Do not describe a named regulatory template, subscription or completed checklist as legal compliance or certification.

Agree the real requirement, data owner, legitimate workflows and exception process. Keep DLP, access, labels, retention and recovery distinct; templates do not establish compliance.

Next: Confirm the actual location, users and prerequisites

Confirm the actual location, users and prerequisites

A rule only affects the supported scope and capabilities actually deployed.

Have the authorised administrator identify the correct organisation and exact policy, service/location, included and excluded users/groups/sites/devices and any administrative-unit restrictions. Check the object actually involved in the failing scenario. A policy covering one SharePoint site does not automatically cover every device, Teams conversation or personal storage service.

Check current entitlement and workload/client prerequisites against Microsoft’s deployment documentation and service descriptions. Device controls can have separate onboarding, platform/application and policy-delivery requirements. Do not infer coverage solely because an administrator can see a setting or one user has a licence.

Use appropriate authorised read-only evidence first. If scope, assignment or processing is uncertain, record that gap and its owner rather than broaden all users/sites or grant global administrator access as a test. Use the licence guide for exact entitlement questions; do not purchase a different plan before identifying the requirement.

Verify the actual tenant, workload, inclusion/exclusion and prerequisites. Visibility of a setting or one assigned licence does not prove every user, device or location is covered.

Next: Check the rule conditions and content detection
More checks6 detailed checks

Check the rule conditions and content detection

An item can contain relevant information without meeting every configured condition.

Review the exact conditions, exceptions and activity that the rule requires. Check whether the intended information type or label is recognised, the match count/confidence criteria and any sharing/recipient/location condition. Read combined conditions carefully; a missing part can explain why the complete rule did not match.

Consider supported content formats, extraction or encryption constraints and workload-specific detection behaviour using the current documentation. A number that looks similar to a sensitive-information example may not meet its validation/context requirements. Do not use a real tax file number, customer record or secret simply to make a test match.

Keep an approved synthetic matching sample and an appropriate nonmatching sample in the authorised test plan, with the expected outcome written first. Protect any private diagnostic evidence. No match can reflect scope, content, activity, processing or a limitation; it is not automatically proof the platform is broken or that the information is safe to share.

Check the complete rule and documented detection limits with approved synthetic examples. Do not substitute real sensitive data or interpret one missing match as general safety.

Next: Review mode, priority and the effective action

Review mode, priority and the effective action

Saved rules and effective behaviour can differ by location and policy state.

Before changing anything, record the current policy/rule state, scope and intended result and agree preservation, impact and a rollback with the owner. Review whether the policy is off, in a simulation state or enforcing, and inspect the exact notification/override/action settings for the workload. Do not turn off an existing production protection to see whether the symptom goes away.

Microsoft’s policy reference describes different rule evaluation and combined-action behaviour for hosted services and devices. Overlapping policies can affect the result; do not assume the first rule, last edited policy or one visible notification is the only effective control.

Simulation is a diagnostic state with workload-specific behaviour. In particular, device actions can differ when simulation policy tips are enabled, and optional automatic transition to enforcement needs explicit ownership. Check the current documented behaviour for the selected settings; do not promise every simulation is consequence-free, immediately complete or incapable of restricting an activity.

Confirm effective workload-specific state, overlaps and actions without disabling production safeguards. Simulation/tips/automatic transitions require an owned impact review; saved configuration is not verified behaviour.

Next: Separate policy tips, alerts and the actual activity

Separate policy tips, alerts and the actual activity

A user message, match record and completed action are different evidence.

Check the actual client/service support and rule notification settings before expecting a policy tip. Microsoft’s notification guidance explains that overlapping matches do not produce a tip for every rule. The displayed tip is not a complete list of everything evaluated.

Have the authorised owner correlate the event time, relevant activity, policy/rule and actual action with appropriately scoped diagnostic records. Check permitted override configuration and whether an override was actually used. Do not treat a match event, alert email or an allowed action as interchangeable evidence of a block.

Account for relevant policy delivery and event/report processing using the current workload guidance rather than a guessed universal wait. A blank report may reflect its time range, filters, permissions or evaluation scope. Protect evidence access and avoid exporting broad customer records or expanding collection just to obtain a screenshot.

Correlate the intended activity with effective action and scoped records. Missing tips, delayed reports and a match alone do not establish enforcement or the absence of exposure.

Next: Test an approved pilot without exposing real data

Test an approved pilot without exposing real data

Tune the intended scenario while preserving legitimate work and existing controls.

Agree the authorised pilot boundary, harmless synthetic samples, permitted test destinations, expected matching/nonmatching outcomes and rollback before a change. Keep real personal information, customer data and secrets out of the test. Do not use a personal cloud account or uncontrolled external recipient to demonstrate a leak.

Use an appropriate controlled configuration and current simulation guidance where it suits the workload and intended checks. Existing-content evaluation differs across locations; a simulation result is not automatically every historical item. Retain effective production protections and explicitly review any notifications, device effects or automatic enforcement transition.

Review false positives, legitimate business impact and gaps with the data owner. Make the smallest justified supported change and repeat the same matching/nonmatching checks. Do not delete the whole policy, remove classification from real files, exempt the whole organisation or weaken retention/holds as a workaround.

Three blank colour cards beside a closed burgundy tablet folio on a white showroom table

Run a scoped authorised pilot with harmless samples and owned mode/transition behaviour. Preserve existing controls and legitimate workflows; no broad exemptions or real-data leakage test.

Next: Verify coverage, actions and the remaining limits

Verify coverage, actions and the remaining limits

Close the intended policy finding with evidence, not a successful save.

Verify the approved matching and nonmatching scenarios in the supported intended contexts after the authorised deployment. Record actual scope, effective policy state, activity, action and appropriate notification/evidence. Also confirm the necessary ordinary workflow still functions under the required safeguards.

Keep the outcome specific to the tested location, users/devices, content and date. A pilot does not prove deployment to every location or detection of every sensitive item. Record exceptions, unsupported cases, pending processing and any further rollout or incident follow-up with an owner.

Return diagnostic settings and evidence handling to the approved state and arrange ongoing review, alert handling and staff guidance. Stop unrelated changes once the intended result is verified. A technical DLP outcome supports a defined protection requirement; it does not certify the whole organisation or replace privacy/legal and business responsibilities.

Verify actual covered actions and legitimate work with dated scoped evidence. Keep unsupported, untested and unfinished areas open; a policy save or pilot is not full compliance.

Back to the symptom checker

Get help with the defined policy outcome

Send a safe reproduction and the actual requirement to the responsible team.

Provide the non-secret scenario, affected service/client, relevant time, intended action and observed evidence to the authorised policy owner. Keep customer content, secrets and unrestricted reports in an appropriate private channel. State whether there may have been a real exposure so incident response is not confused with routine tuning.

Our Microsoft 365 security review and Microsoft 365 support pages can start a scoped discussion. Agree actual entitlement, responsibilities, preservation and verification. Do not assume every subscription or support plan includes every workload, legal determination or ongoing alert response.

If the intended requirement cannot be met by the current supported scope, record the gap and appropriate owner decision. Use an approved alternative workflow or protection plan rather than bypass the rule with a personal account, unauthorised exemption or altered real document.

Use the actual owner and a scoped harmless reproduction. Keep incident handling and unresolved requirement gaps visible without bypassing policy or promising certification.

Frequently asked questions

Does a DLP policy template make my business compliant?

No. Confirm the requirements that actually apply, the intended information-handling outcome and the effective covered controls. Templates support configuration; they do not determine legal applicability, replace business processes or certify the whole organisation.

Does a missing policy tip mean DLP did not evaluate the item?

Not necessarily. Client support, notification settings and overlapping rule behaviour can affect the tip. Correlate the intended activity with scoped policy and action evidence instead of treating the visible message as a complete evaluation record.

Why did a rule record a match without blocking the activity?

Detection and enforcement are different. Review the actual policy state, configured actions, workload, exceptions, permitted overrides and overlapping rules. A saved Block setting or match record alone does not establish the effective action in every context.

Is simulation mode guaranteed to have no effect on users?

Do not assume that. Workload-specific settings, notifications, device actions and optional automatic transition to enforcement need review. Use current documentation and an approved scoped plan while preserving existing protections; simulation is not equivalent to an isolated test environment.

Can I test the policy with a real customer file sent to my personal account?

No. Use authorised synthetic samples and approved controlled destinations with expected matching and nonmatching outcomes. Real exposure needs incident response, not repetition. Keep private evidence controlled and do not use personal accounts or broad exemptions to bypass safeguards.

More practical IT guides

View all guides

Share this guide