Security projects: how do you get a control to everyone, and into IT projects before go-live?

Before your next security rollout starts, write down its scope, a coverage target such as 98% of in-scope accounts enforced, and an end date for every exception, then run it in report-only or audit mode, a pilot and two or three rings. In the company's other IT projects, ask the security questions at kickoff and sign off before go-live, and scale the effort by protection need so only high-need projects wait for a full review.

Reviewed
Next review
Version
1.0
Written for
Heads of security, IT leads who own security, and project managers in small and mid-size companies

If you only do five things

  1. Define done in numbers before you start.

    Count the scope from the inventory and set a target, such as 98% of in-scope devices in block mode by 31 March, with an exception cap. Count enforced, not deployed.

  2. Watch before you enforce.

    Keep a new Conditional Access policy in report-only mode for at least a week1; it logs but blocks no one2. Audit ASR rules first, except the standard protection rules3.

  3. Roll out in rings sized for your help desk.

    Pilot with IT and a champion per business unit, then a representative ring, then everyone, in waves within support capacity5.

  4. Give every exception an end date.

    Exclusion lists start short and grow11. Give each member of a policy's exclusion group a reason, an owner and an end date.

  5. Ask security at kickoff, not at go-live.

    The BSI requires involving the ISB early in larger projects and new IT systems13; Grundschutz++ requires a rule for when14.

Roll out a security change in this order

  1. Who and what is in scope?

    Count from the inventory, not the new tool; Defender's device discovery (Plan 2) finds devices not yet onboarded8. The security program sheet decides what goes first.

  2. What would break?

    Run report-only or audit mode for one to four weeks and review the results with app owners. For segmentation, allow and log first, as the network segmentation sheet describes.

  3. Who pilots it?

    IT and a champion per business unit3, with a channel for problems. What they hit becomes the user message and the help desk script.

  4. When does the next ring start?

    After a week with no open blocking tickets. Reuse your Windows update rings if you have them.

  5. When is it done?

    When the tail is closed by name, the old path is gone, such as SMS codes, and an owner gets a monthly coverage report.

Users, help desk and rollback

  • Tell users what changes, when, and how to get help1, a week before their ring and on the day.
  • Brief the help desk before each ring: the script, known errors and what to collect, including the correlation ID from More details1.
  • Write the rollback first: the trigger and the switch. Disable the CA policy or exclude a group1. Exclude the app from ASR or return the rule to audit3. Pausing Intune deployments stops only later rings; revert the policy6.
  • Exclude Mac, iOS and Android from report-only compliance policies, or their users may get certificate prompts2. Test them in an enforced pilot.
  • In Germany, ask the works council early about EDR, DLP and logging; it co-determines systems designed to monitor behavior or performance12.

Exceptions that end

  • Exclude groups, not people. One assigned Entra group per policy shows who is excluded11; break-glass accounts go in every one1.
  • Record why, what instead and until when: a reason, a compensating control, an owner and an end date per member. Approval levels are in the risk register sheet.
  • Let reviews remove people. Entra access reviews can ask members weekly and remove those who don't answer (Entra ID P2 or ID Governance)11.
  • Cap them. Our rule: at most 2% of the scope at go-live; any still open at the done date is renewed one level up, as in the risk register sheet. Report the count and the oldest monthly.

Measure adoption against the inventory

  • Coverage is enforced over in scope, counted weekly from the inventory by ring and business unit, with the tail named.
  • MFA and passkeys. Entra's Authentication methods > Activity lists who is MFA or passwordless capable and counts single-factor sign-ins9.
  • Conditional Access. The gap analyzer workbook lists sign-ins no policy applied to (P1 and Log Analytics)10.
  • EDR and ASR. Compare onboarded with discovered devices8; the ASR report needs Plan 2 or Defender for Business, else forward event 11223.

Mistakes that cost the most

  • Stopping at 80%. The tail holds the old devices, service accounts and executives. Give it its own phase and owners.
  • Exceptions that become permanent. Exclusions without an end date outlive their reason. Let them expire unless someone renews them.
  • No owner after go-live. Conditional Access policies have no owner field, so Microsoft suggests putting it in the name and keeping a register1.
  • Piloting only in IT. IT and lab devices usually lack a representative mix of apps and drivers4.

Security gates in IT and business projects

  • At kickoff, ask six questions: what data, who signs in and how, internet-facing or not, which supplier, who gets admin rights, when it goes live.
  • Rate the processes the system serves: normal or high. Grundschutz++ sets these two; on its catalog, high need adds a risk analysis14.
  • Write requirements into the plan. Normal: SSO with MFA, exportable logs, named admins, backups, patch deadlines. High: add a design review. Software you build: see the secure development sheet.
  • Pentest at go-live if the system faces the internet, has high need, is custom-built or changes sign-in; ENISA advises testing at set-up15. Otherwise scan and check settings. See the penetration testing sheet.
  • Sign off. Security confirms each requirement or lists open items with owners and dates; management approves the risk of unmet ones, which Grundschutz++ rules out for MUSS requirements14.

Security in tenders and contracts

  • Use the six NIS2 purchasing points: security requirements, updates for the whole lifetime or replacement after support ends, component information, security functions and secure configuration, assurance, and how you validate it15. They bind digital providers; in Germany, BSIG § 30(2) no. 5 requires security in acquisition of all NIS2 entities17.
  • Ask when support ends. From 11 December 2027, the CRA requires makers of new products to state the month and year at purchase16.
  • Require free, prompt fixes, tested and delivered as soon as an issue is known, as ENISA advises15.
  • Tie acceptance to evidence. Make acceptance and the final payment depend on SSO and MFA working, logs flowing and critical pentest findings fixed. Add the third-party risk sheet's clauses.

Don't become the bottleneck

  • Publish when security joins. Grundschutz++ doesn't need the ISB active in every security-relevant project14. Join the high-need ones.
  • Promise a turnaround, such as the operating model sheet's 5 working days. If you miss it, normal-need projects go ahead and log open items.
  • Pre-approve patterns: a SaaS requirement set, a contract annex, a reference design.
  • Put a champion in the PMO to ask the kickoff questions and call you in for high-need projects.

What you can skip

  • A risk analysis for every project. With the Grundschutz++ catalog, normal need requires none unless a requirement stays unmet14.
  • A pentest at every go-live. A standard SaaS app behind SSO needs a check of your tenant's configuration.
  • Auditing the standard ASR rules. Microsoft says they can typically go to block or warn without testing3.
  • Months in report-only. Our limit: a week to a month of real sign-ins. Enforce, then fix the tail.

What changed in 2026

  • Intune stages policy rollouts. Deployment plans, in preview since September 20267, push Windows policies and apps ring by ring and can be paused6.
  • Grundschutz++ wants a project rule. BSI-Standard 300-1 (20 October 2026) requires a security role in relevant projects and a rule for when14.
  • CRA support periods got clearer. The Commission's draft CRA guidance of 27 July 2026 says the five-year minimum isn't a default16.

What this looks like at your size

Small team, no security staff

  • The IT lead or provider runs rollouts; a managing director approves exceptions.
  • Pilot with two or three users, then everyone.
  • Check IT purchases against the six NIS2 points.

Growing, with IT or security staff

  • Write one page per rollout: scope, target, rings, rollback, exceptions, owner.
  • Report coverage at the program's monthly meeting.
  • Ask the kickoff questions in every IT project.

Enterprise or regulated

  • Ring by region or site, and hold mission-critical devices until last4.
  • Make security sign-off a gate in the project method.
  • Put a security annex in every IT tender template.