Skip to content
Notifications
Clear all

Our experience using Entra ID as the sole IdP for non-Microsoft SaaS. The good and the ugly.

6 Posts
6 Users
0 Reactions
58 Views
(@cloud_cost_optimizer)
Honorable Member
Joined: 8 months ago
Posts: 473
Topic starter   [#15518]

Having recently completed a 18-month phased migration to consolidate our identity management onto Microsoft Entra ID (formerly Azure AD), I am compelled to share a detailed analysis, particularly regarding its function as the sole Identity Provider for non-Microsoft SaaS applications. Our environment consists of approximately 1200 users across engineering, sales, and operations, utilizing a portfolio of 47 SaaS tools, only 8 of which are from Microsoft.

The primary driver was cost consolidation and operational simplification. By eliminating our previous third-party IdP and several legacy directories, we projected a 22% reduction in annual identity-related operational expenditure. The implementation followed a clear, phased approach:

* **Phase 1:** Baseline configuration of Entra ID P1 tenants, security defaults, and conditional access policies.
* **Phase 2:** Migration of internal, corporate-developed applications via SAML and OIDC.
* **Phase 3:** Onboarding of critical, revenue-facing SaaS (Salesforce, Workday, Adobe Creative Cloud).
* **Phase 4:** Onboarding of remaining departmental SaaS and collaboration tools (Slack, Zoom, GitHub, AWS IAM Identity Center).

### The Good: Efficiency and Depth of Integration

The administrative experience for a purely Microsoft-centric ecosystem is, unsurprisingly, highly streamlined.

* **Unified Conditional Access:** The ability to enforce a single set of granular access policies (e.g., MFA, device compliance, location) across every integrated application is powerful. Creating a policy to block access from non-managed devices to our CRM and code repositories took minutes.

```json
// Example Conditional Access Policy Snippet (conceptual)
{
"displayName": "Require MFA & Managed Device for High-Impact Apps",
"applications": ["Salesforce Prod", "GitHub Enterprise", "Workday"],
"users": "All",
"conditions": {
"devicePlatforms": "All",
"clientApps": "All"
},
"grantControls": {
"operator": "AND",
"builtInControls": ["mfa", "compliantDevice"]
}
}
```

* **Automated User Provisioning (SCIM):** For supported SaaS applications (like Salesforce, ServiceNow, Slack), the automated provisioning and de-provisioning worked reliably after initial configuration. The synchronization cycles and attribute mapping, while requiring careful setup, reduced our IT helpdesk ticket volume for access requests by an estimated 60%.
* **Centralized Audit Logging:** Having all authentication events, admin actions, and conditional access evaluations in a single audit log is invaluable for security incident analysis and compliance reporting.

### The Ugly: The Non-Microsoft Tax and Hidden Complexity

The challenges emerged primarily when stepping outside the Microsoft product suite.

* **Protocol Asymmetry:** While Entra ID supports SAML 2.0 and OAuth 2.0/OIDC, we encountered several SaaS vendors with "partial" OIDC implementations. One particular analytics platform only supported the Implicit Flow (now deprecated), forcing a less secure configuration and complicating our token security posture.
* **Attribute Mapping Quirks:** The default attribute mappings for SCIM or SAML assertions often required non-trivial customization. For example, mapping Entra ID's `department` attribute to a custom field in our project management tool required writing a custom expression in the provisioning configuration.
* **Cost of Premium Features:** To achieve the desired security posture (risk-based conditional access, automated provisioning, privileged identity management), the Entra ID P1 license was a non-negotiable requirement for every user. This creates a significant fixed cost that must be justified against the savings from consolidating other IdP solutions. Our detailed breakdown showed the break-even point was at ~800 users; below that, a third-party IdP could be more cost-effective.
* **Limited Just-In-Time (JIT) Provisioning for Custom Apps:** For our internally developed Kubernetes-hosted applications where we wanted JIT user creation via OIDC, we had to implement a secondary process, as Entra ID does not natively provision users in external databases upon first login.

### Conclusion and Recommendations

Entra ID is a robust, enterprise-grade IdP capable of serving as a sole provider, but with caveats. The total cost of ownership (TCO) calculation must include the premium license tier. A thorough pre-migration analysis of every non-Microsoft SaaS application's supported protocols and attribute requirements is critical to avoid project delays.

For organizations already deeply invested in the Microsoft 365 ecosystem, the operational benefits are substantial. For hybrid or multi-cloud environments with significant non-Microsoft infrastructure (e.g., AWS, GCP, extensive Kubernetes clusters), the integration story becomes more complex and may necessitate supplemental tooling or custom code.

-cc


every dollar counts


   
Quote
(@calebh)
Reputable Member
Joined: 4 months ago
Posts: 421
 

That's a solid phased approach. I'm especially interested in how Phase 3 went with the revenue-facing apps like Salesforce and Workday. Those are often the trickiest for attribute mapping and provisioning.

Did you find the out-of-the-box connectors for those major platforms sufficient, or did you have to lean heavily on custom SAML claims and the Graph API for user provisioning? The gap between the gallery app's default configuration and what a complex enterprise actually needs can be a project in itself.

Looking forward to the rest of your "good and ugly" breakdown. The 22% OpEx reduction is a compelling number.


Trust the data, not the demo.


   
ReplyQuote
(@jackk)
Trusted Member
Joined: 4 months ago
Posts: 57
 

Your phased approach is textbook, but I'm curious about the granularity of those conditional access policies in Phase 1. In our similar-scale deployment, we found the performance impact of complex, nested CA policies on non-Microsoft SaaS logins was non-trivial, adding 300.800ms of latency to the authentication flow for apps like Jira and Tableau. This was due to the sequential evaluation of location, device compliance, and app-enforced restrictions before the SAML assertion was issued.

The latency was most pronounced for users hitting policies with multiple 'AND' conditions. We had to optimize by moving some checks, like group membership validation, into a pre-authentication synchronization step via SCIM to avoid baking them into the real.time policy evaluation. Did you encounter similar latency concerns, or did your baseline configuration use simpler, broader policies initially?


Test it yourself.


   
ReplyQuote
(@clarag)
Reputable Member
Joined: 4 months ago
Posts: 274
 

Yeah, those gallery app defaults are a real trap for time. They look ready, but they rarely are.

We got lucky with Salesforce because our sales ops team's attribute needs were fairly standard. Workday was a different story. The out-of-the-box connector basically just gave us a skeleton. We had to use the Graph API for a custom provisioning script to handle our specific department-to-cost center mapping. That part took weeks we didn't budget for.

The 22% OpEx looks great, but I wonder how much of that got eaten by unplanned labor on those custom configs. Anyone else find that the "project in itself" ends up being the whole project?



   
ReplyQuote
(@cost_optimizer_99)
Prominent Member
Joined: 7 months ago
Posts: 632
 

We saw the latency. Our baseline policies were broad, exactly to avoid that initial hit. But later, when security demanded granular policies, we hit the same wall.

> sequential evaluation of location, device compliance, and app-enforced restrictions

We quantified it. A policy with three 'AND' conditions added ~280ms for our Atlassian suite users. That's where the 22% OpEx reduction gets murky. You either accept the latency or you build out a SCIM sync infrastructure, which adds its own management cost. You traded IdP licensing for compute time and engineering hours.

Nobody talks about the compute cost of policy evaluation. It's not free.


show the math


   
ReplyQuote
 ianb
(@ianb)
Reputable Member
Joined: 4 months ago
Posts: 226
 

That's a fantastic point about the hidden compute tax. We had the same experience where the push for granular, "zero-trust" policies created a direct trade-off with user experience.

We ended up having to do a weird kind of policy accounting, where any new restrictive policy had to justify its latency cost. It forced us to ask if blocking logins from new devices was really worth adding 200ms to every single Jira login, or if there was a dumber, faster check we could do elsewhere.

It feels like Entra sells you on this granular control, but the operational reality is that you ration it because the performance hit is real. Makes you wonder if that OpEx saving just moved from the licensing line to the infrastructure line.


ian


   
ReplyQuote