What should you ask before choosing a multi-tenant Defender and Sentinel platform?

Six questions for choosing between Microsoft's own multi-tenant stack and a Defender and Sentinel platform, each answered from sources.

Reviewed by ContraForce team · Last verified

Before choosing a multi-tenant Microsoft Defender XDR and Sentinel platform, ask six questions of it and of Microsoft's own multi-tenant tools: how each reaches customer tenants, whether response waits for a per-tenant approver, how far agents act alone, whether verdicts reach the native incident, which ticketing systems are native, and who pays for licensing, the platform and your team's time.

Last verified: 2026-11-22. Each answer for Microsoft's own stack links its main Microsoft sources, every Microsoft source is listed at the foot of the page, and each ContraForce answer in the table links the documentation page that states it.

A managed service provider has two ways to run Microsoft Defender XDR and Sentinel for its customers. The first is Microsoft's own multi-tenant stack: Defender multitenant management, cross-tenant delegated access, Entra B2B, Azure Lighthouse, Microsoft Sentinel, the Integrated Security Operations Center (ISOC) in Defender, and Microsoft Perception and Security Copilot agents. Each customer licenses those from Microsoft. The second is a platform on top of the same Microsoft products, which you pay for.

Licensed by the customer does not mean free to you. On Microsoft's own stack, your tenant still has licence requirements, and your team sets up content distribution, builds Sentinel playbooks and handles per-tenant setup. On a platform, you pay its price and should ask what it still leaves your team to build. The six questions below apply to both paths, and a section after them answers each one for Microsoft's own stack.

Ask for each answer in writing, with the documentation page that states it, before a trial starts. A vendor that leaves one of these unanswered has usually answered it: the missing answer is a no.

How does the platform reach each customer tenant, and what does the customer consent to?

Microsoft offers several ways into a customer tenant, and they do not reach the same data. Cross-tenant delegated administration in Microsoft Defender became generally available on 2026-10-20: Microsoft builds it on Entra governance relationships with granular delegated admin privileges (GDAP), and it now reaches Microsoft Sentinel resources as well as Defender, through Defender unified RBAC or Azure RBAC roles. That makes GDAP-based delegated access part of Microsoft's own multi-tenant path, beside Entra B2B guest access and Azure Lighthouse. Some Learn pages still say GDAP does not reach Sentinel data, or that it is in preview, so check the date on any Microsoft page quoted to you.

A platform can also connect through an enterprise application registered in the customer's Microsoft Entra tenant, which the customer's administrator consents to once. Microsoft's admin consent documentation names the role this needs: a Privileged Role Administrator, "for granting consent for apps requesting any permission, for any API."

A complete answer names:

Can automated response wait for an approver, tenant by tenant?

Automated response in a Microsoft Defender XDR and Sentinel estate acts on a customer's identities, mailboxes and devices, so the gate in front of it matters as much as the action. The useful question is where that gate lives, what its default is, and who sets it.

A managed service needs the gate per customer, because customers differ. One customer allows the provider to isolate a device without asking; another wants their own administrator to approve every password reset. A setting that applies to every customer at once cannot express that.

A complete answer covers:

How far does the agent act on its own, and what sets the threshold?

AI investigation agents for Microsoft Defender XDR and Sentinel incidents usually separate two things: investigating an incident and acting on it. Ask for both to be configurable separately, and for the threshold that moves an agent from the first to the second to be visible rather than implied.

Ask for:

Does the verdict land back in the customer's Sentinel and Defender incident?

Microsoft Sentinel stores a closed incident's verdict in fixed fields. The automation rule schema accepts exactly four classifications, BenignPositive, FalsePositive, TruePositive and Undetermined, and four closing reasons beside them, so any tool that writes a verdict back has to map into those values.

This matters for a provider because the customer's own security team, auditor or incoming provider will read the native incident, not a platform's copy of it. A verdict that stays in the platform leaves the customer's Sentinel workspace showing an incident closed with no reason.

Ask, per detection source:

Which ticketing systems are native, and which connections does your team build?

For a provider running Microsoft Defender XDR and Sentinel, ticketing is where multi-tenant delivery either reaches the customer or stops at the SOC. The distinction to draw is between a native integration, where a ticket can be created, linked and searched from inside the incident, and a webhook, API or email route, where your team builds and maintains the connection.

Both are legitimate. The risk is counting every ticketing system that can receive an email as an integration. Ask for:

Who pays for Microsoft licensing, for the platform, and for your team's time?

On either path, each customer keeps paying Microsoft for Defender XDR and Sentinel, you pay for any platform you choose, and you carry your team's time plus any licences or consumption billed to your own tenant. Microsoft lists the subscriptions that give Defender XDR access, including Microsoft 365 E5 or A5 and Microsoft 365 E3 with the Microsoft Defender Suite add-on. Sentinel is separate: Microsoft's onboarding guide says that to use Sentinel in the Defender portal "You don't need Microsoft Defender XDR or an E5 license", and Sentinel is billed to the Azure subscription that runs it.

On Microsoft's own stack, your costs can include licence requirements in your own tenant, consumption charges, and the engineering and analyst time to build and run what the tools leave to you. A platform adds its price and still leaves some work to your team, so count your team's time on either path.

Ask, for each path:

How does Microsoft's own stack answer the six questions?

Microsoft's own stack answers all six with tools each customer licenses. It does several things well: verdicts are native, delegated access now reaches Sentinel, and automatic attack disruption contains high-confidence attacks without waiting for approval.

QuestionMicrosoft's own stackSource
Tenant access and consentNative: delegated access, B2B or Lighthouse. Cross-tenant delegated administration, generally available since 2026-10-20 and open to MSSPs that aren't Cloud Solution Provider partners. Once governance invitations are turned on (they are off by default), the customer invites you and approves your access request; Microsoft Entra names at least a Tenant Governance Administrator to accept it. Your security groups then appear in the customer tenant as remote tenant groups, without B2B guest accounts, and the customer assigns them Defender unified RBAC or Azure RBAC roles, Sentinel included. Entra B2B guest access and Partner Center GDAP, which the customer's Global Administrator approves, remain options. Advanced hunting in multitenant management queries each tenant's primary Sentinel workspace without Azure Lighthouse. Microsoft's multitenant management pages, last updated before that general availability, say Lighthouse, which the customer deploys per subscription, is required to query a secondary workspace in another tenant and may be needed for some Sentinel automation rule and playbook scenarios. Defender multitenant management is limited to 100 target tenants and doesn't support Defender for Business tenants.Delegated access GA, Governance relationships, Multitenant management, MSSP implementation guide
Per-tenant approvalPer tenant, and per product. Defender for Endpoint sets an automation level per device group; tenants created on or after 2020-08-16 default to full automation, which Microsoft recommends, and since 2026-09-01 Microsoft says Defender for Endpoint's automated investigation and response no longer runs as a separate investigation experience, while Defender for Office 365's remains available. Defender for Office 365 automated investigation queues email remediation for approval in the tenant's Action center, whose History tab is the audit log, except for malicious similarity clusters, including URL and file clusters, which it approves itself. Automatic attack disruption acts without approval even at Semi automation, and its actions can be undone; per-tenant exclusions are the brake, and a device group can be set to no automated response, which Microsoft advises only for a limited number of devices. Sentinel automation rules are defined per workspace and change the incident or run a Sentinel playbook, so an approval step is one you build into the playbook, as in Microsoft's pattern that emails Block and Ignore buttons and waits for the answer. Content distribution pushes custom detections, endpoint security policies and Sentinel content to many tenants, but can't distribute automation rules that trigger a playbook, and device-group automation levels and attack disruption exclusions are not among the content types it lists.Automation levels, Action center, AIR in Defender for Office 365, Configure attack disruption, Sentinel automation rules, Content distribution
Agent autonomyAgents ask before significant actions; attack disruption doesn't. Microsoft announced Microsoft Perception, previously Project Perception, as generally available at Ignite 2026; some of its Learn pages still say Limited Public Preview. Its agents pause and ask for approval before significant actions, such as disabling a user account or isolating a device, and stay paused until someone responds; Microsoft says these gates are configurable. The Triage Agent resolves alerts it classifies as false positives, leaves true positives open for an analyst, and runs automatically on the supported alert types you select. The Attack Investigation Agent recommends response actions for your team to review. Microsoft says agent sessions stay within one tenant's boundary. Separately, automatic attack disruption contains compromised assets without approval; for containment, Microsoft maintains a confidence level of 99% or higher.Ignite 2026 Book of News, Approve or stop agentic work, Triage Agent, Attack Investigation Agent, Perception application card, Automatic attack disruption
Verdict write-backNative, synced both ways. From Defender multitenant management you can classify, tag, assign or change the status of a customer's incident, or open it in that customer's tenant. Classification and status can change across tenants at once, but bulk assignment works within one tenant, and what you can change depends on the permissions each tenant has granted. Microsoft maps Defender's classification into Sentinel's classification and reason, and syncs status, owner and closing reason both ways; several changes to one incident within 5 to 10 minutes reach Sentinel as one update carrying the latest. The Perception Triage Agent writes its classification and status into the Defender incident, with an explanation of its verdict.Multitenant incidents, Defender and Sentinel sync, Move Sentinel to Defender, Triage Agent
TicketingNative case management; external tickets through Graph or playbooks. Defender has native case management, and its multitenant portal shows cases from many tenants in one view; Microsoft presents it as a way to end reliance on external ticketing systems. Incident cases, the recommended way to manage incidents across tenants, are in preview. For an external ticketing system, Microsoft recommends the Microsoft Graph REST API, whose permissions are granted per tenant and per application. In Sentinel, Microsoft's pattern is an automation rule that runs a Logic Apps playbook to create or update the ticket, and Microsoft ships ready-made playbook templates that sync Sentinel incidents with ticketing systems in both directions. Those playbooks live in your tenant or the customer's, and some cross-tenant setups need an extra Azure Lighthouse delegation. The Defender portal can also generate code-based playbooks from natural language, which call third-party APIs through integration profiles.Case management, Multitenant cases, MSSP implementation guide, Graph security authorization, Sentinel playbook use cases, Generate playbooks with AI
Who pays: the customerMicrosoft licences and Azure consumption, per customer. Each customer pays Microsoft. Defender XDR comes with subscriptions such as Microsoft 365 E5, or E3 with the Microsoft Defender Suite add-on, and Sentinel bills to the Azure subscription that runs it. Where you manage customer workspaces through Azure Lighthouse, Microsoft says related costs are charged to each managed tenant, whose data stays in its own tenant and under its ownership, which supports requirements to store data within geographical boundaries. Customers with Microsoft Defender Suite, Microsoft 365 E5 or E7 are eligible for ISOC, which brings SIEM capabilities into the Defender portal; an ISOC workspace needs an Azure subscription, and ingesting more data can add charges. Microsoft announced ISOC's core capabilities as generally available at Ignite 2026, with the full breadth of SIEM capabilities and the included Defender data retention still in preview, and its Learn pages still describe ISOC as a preview.Defender XDR prerequisites, Sentinel billing, Sentinel at scale with Lighthouse, ISOC overview, Ignite 2026 Book of News
Who pays: consumptionPer use: Logic Apps playbooks and Perception agents. Sentinel's Logic Apps playbooks can add charges, and Microsoft says they can be deployed in your tenant or the customer's. Microsoft Perception uses consumption-based pricing, priced separately from Security Copilot, with a cost estimator that works per agent outcome and budgets that notify rather than cap; ask Microsoft which tenant is billed when its agents work a customer's incidents.Run Sentinel playbooks, Sentinel at scale with Lighthouse, Microsoft Perception, Manage Perception costs
Who pays: youYour tenant's licences and capacity, and your team's time. Your own tenant has licence requirements: for governance relationships, your tenant and the customer's each need at least one Microsoft Entra ID P1 licence and either one Microsoft 365 E5 licence or Sentinel enabled in Defender. Azure Lighthouse has no extra cost. Security Copilot uses the customer's capacity when you work through B2B or, as a Cloud Solution Provider partner, through GDAP, which Microsoft allows only for those partners, and yours when you prompt against a customer's Sentinel workspace through Lighthouse; Microsoft 365 E5 and E7 customers get a monthly allowance. Your team's time goes into delegation for each customer tenant, which Microsoft's Tenant Governance APIs in Microsoft Graph can automate, and, for content beyond native distribution's "create once, deploy everywhere", custom CI/CD pipelines that Microsoft says need custom scripts and configuration files.Governance relationships, Azure Lighthouse, Security Copilot for MSSPs, MSSP implementation guide

How does one platform answer all six?

ContraForce answers all six from its published documentation: scoped Microsoft Entra enterprise applications rather than GDAP, a per-workspace approval gate that holds Gamebooks by default, per-severity agent autonomy behind a confidence threshold, verdict write-back to Microsoft Sentinel and Defender XDR, native create and link with three ticketing systems, and a price paid by the provider, with hosting in the United States, the United Kingdom or the European Union.

QuestionAnswerSource
Tenant access and consentPurpose-built Microsoft Entra enterprise applications, two core and up to five more consented per module, mostly with application (app-only) permissions; a few flows, such as portal sign-in, Sentinel log search and, by default, password reset, use delegated permissions. A Global Administrator or Privileged Role Administrator grants the one-time consent, and the role is not retained. Access runs through these applications rather than GDAP.Enterprise application permissions
Per-tenant approvalSet per workspace. While a workspace's Allow service providers to run gamebooks setting is off, which is the default, every Gamebook an agent submits waits for approval. Turning it on requires the Workspace Owner role, and run policies can still send an action to an approver. A Workspace Profile sets the approvers and agent policy once and applies them to every attached workspace.Configuring agents, Workspace Profiles
Agent autonomyEach run is investigation only (analysis only, no response actions) or investigation and response. Per severity, the agent runs Manual or On Queue, and Agent Shifts limit working hours. Allow agent to run gamebooks, off for a new agent, decides per severity whether the agent's Gamebooks can run or wait for approval; with it on, a 0 to 100 Confidence to respond threshold decides whether an On Queue run builds a Gamebook, and below it the response is left to an analyst.Configuring agents
Verdict write-backMicrosoft Sentinel receives the classification one to one, with reasons mapped to its fixed reason list. Microsoft Defender XDR receives the classification and determination, and most verdicts round-trip exactly. The mapping page also covers two third-party XDR modules: one receives the analyst verdict, the other only the closure, because its API has no classification field. Closures made in a vendor's own portal are reconciled back.Incident classification mapping
TicketingNative in-product create and link with three ticketing systems, listed in the API endpoint reference.API endpoint reference, OpenAPI spec
Ticketing by webhookOther ticketing systems are reached through signed webhooks or email-to-ticket, built by the provider.Webhook event reference, Email-to-ticket
Who paysYou pay for the platform, and your customers don't: a monthly plan, with plan prices published on the pricing page, plus a flat rate per incident a Security Delivery Agent processes, which is quoted directly rather than published. Microsoft keeps billing each customer for Sentinel and Defender as it does today. On the self-hosted option the agents run in your own cloud subscription, which is billed for their model usage, and the platform doesn't bill per run. For each new customer, your team arranges the one-time module consent with the customer's administrator, then attaches the workspace to a Workspace Profile, whose one apply sets access, modules, notifications, Gamebook approvers and agent policy; a webhook with no workspace list already sends events from every workspace you manage, including new ones.Pricing, Cost Management Center, Workspace Profiles, Configure webhooks
HostingHosted by default in the United States, the United Kingdom or the European Union.Hosting regions

Plans and workspace allowances are on the pricing page. For the wider evaluation, including evidence standards, model governance and proof-of-value scenarios that apply to any AI SOC platform, use the AI SOC platform RFP checklist.

Sources

Microsoft licensing and access models change often, and several Microsoft Learn pages above still lag announcements Microsoft has made since they were last updated. Confirm each Microsoft statement against the linked source, and each vendor answer against the vendor's current documentation, before you choose.

Continue the evaluation

Sources and review method

Product capabilities were reviewed against the page-specific primary sources below on 2026-11-22. Performance claims require the population and limitations stated in the linked methodology.

Related guides on evaluating security platforms

Ready to Transform Your Security Operations?

See how ContraForce runs the full delivery loop for service providers. Connect a supported tenant and validate the workflow in your own environment.

Start for free Read the docs

Connect one workspace: scoped Entra ID application consent, no endpoint agents, no log forwarding.