An employee portal is judged by what it shows: the request form, the leave balance, the ticket status. The cost lives in who it cannot identify.

Microsoft publishes the documentation for its Employee Self-Service agent, and one line in it is more useful than any feature list. It is worth reading as a scoping document rather than a product page.

What the Workday integration page actually says

The agent, per Microsoft, is "built on Copilot and uses AI to provide relevant information to employees and take actions on their HR data." Then the page states the boundary:

"Workday integration is currently configured to get only Employee details and doesn't work for External staff or Non-Employees."

That is a documented scope decision, not a defect. It is also the population most likely to call the service desk: contractors, agency staff, seasonal hires, and people who have accepted an offer but have not started.

The same page documents two integration paths. The legacy path "uses two integration system users (ISUs) in Workday and a Reports as a Service (RaaS) report for user-context lookup." A simplified path "reduces required Power Platform connections from four to two and removes both the ISU dependency and the user-context RaaS report." New deployments are directed away from the legacy page; existing ones can upgrade incrementally and decommission the older pieces on their own timeline.

The grant list is longer than the feature list

Most portal business cases estimate effort from screens. The documentation estimates it from entitlements.

The integration page's permission tables name two integration security groups, and between them the domains listed cover Workday accounts, work email, current staffing information, worker ID, tenant reporting setup, custom report creation, job information, compensation package setup and integration build.

None of that is decoration. A portal that shows a manager the team's job titles needs the job information domain. One that shows work email needs the work email domain. The page then adds a caution that governs scope: it describes those configurations as basic requirements and notes that other permissions may be needed depending on the topics a team implements.

Microsoft's separate Workday administrator reference adds the detail that matters for governance. Permissions are granted at the integration level rather than the report or task level, and which domain governs a given data element depends on how the tenant is configured.

Then there is the handoff. The client ID and the endpoints generated after the API client is created are to be passed to the Microsoft Entra administrators through a secure channel. Where the integration accounts also exist in Entra, the Workday usernames have to line up with the Entra accounts for OAuth to work.

Network is a third gate. The agent reaches HR data through a RaaS endpoint over SOAP, and the administrator reference directs teams whose organizations filter outbound traffic to hand the REST and SOAP host names to the network administrator. In practice three groups — HR operations, identity and network security — each hold a piece of a launch a steering committee approved as a single line item.

Why the population gap is a capacity problem

This section is analysis, not a Microsoft claim.

A portal earns its keep by removing requests from a queue. If it covers employees and the queue still holds contractors, seasonal staff and pre-hires, the queue does not shrink. It splits into one track the portal handles and another a person still handles, and the second track is usually invisible in the business case.

The split has a second cost that rarely appears in a project plan. Two intake paths mean two sets of expectations, two response times, and a recurring argument about which population qualifies. Staff capacity that looked recovered shows up as triage instead.

None of that argues against building the portal. It argues for enumerating the population before the acceptance criteria are written, because a documented limit discovered at go-live becomes a change request.

Extending the tool people already use

Microsoft's November 12, 2026 post on its own customer engagements offers a placement argument that travels better than its numbers.

At Northern Trust, engineers worked alongside the bank's experts to "build an AI-powered onboarding agent within the Excel and SharePoint tools analysts already use." The stated result: "data preparation time fell 83%, from two and a half weeks to three days, with 96.5% first pass quality score in testing," while "analysts remain in control."

Those figures are Microsoft describing a Microsoft engagement. They were not independently audited for this edition, and the workflow tested is one onboarding process, not a companywide capability. The transferable claim is the placement decision — inside the surface the analyst already opens — not the percentage.

The same post is careful in a different place, and the difference is instructive. For 3M's order-management workflow, Microsoft writes that the solution "is expected to significantly reduce manual effort" — a forward-looking statement, labeled as such. For Publicis, the account notes a prior rollout of Microsoft 365 Copilot to "all 114,000 of its employees," a deployment fact rather than an outcome.

Read together, the three examples model the right grammar for an internal-tool case: what was built, where it sits, what was measured, and what is merely expected.

A portal does not fix data quality. It publishes it.

Internal tools rarely create bad data. They surface it, at scale, to the people who trust it most.

A stale manager field becomes a request routed to someone who left. A duplicate worker record becomes a leave balance that shows twice. A cost center that was never updated becomes an approval that stalls. Each of those is a back-office error until a portal puts it on a screen with a submit button.

Microsoft's IDC-commissioned research, published the same day, supplies the measurement context: "Only 61% of organizations are highly confident they can measure AI's business value and ROI, versus 93% of Frontier Firms." The same study reports that 66% of leaders say AI has delivered substantial or transformative results, while only 48% are scaling that value across business units — and it states plainly that the returns it reports "are leader-reported, and the research shows association rather than cause."

The lesson for a portal business case is narrow and useful. If the benefit is described as hours saved, nobody will be able to check it. If the benefit is described as queue volume, mix and time to resolve, measured before and after, someone can.

Three ways to scope the build

This section is analysis. The options are delivery models, not vendor recommendations.

Adopt the packaged extension. Fastest path, and it inherits the vendor's population model. Every population the package does not cover becomes your process, and every domain in its permission table becomes your grant to own. The cost is configuration and coordination across three teams.

Build a thin portal over supported interfaces. A custom front end over the HR system's documented APIs, with your own identity mapping for the populations the vendor model excludes. The cost is upstream engineering and a maintenance obligation. The benefit is that the definition of "worker" belongs to you.

Normalize worker identity first. Build the mapping layer — employee, contractor, seasonal, pre-hire, rehire — before building any screen. This is the least satisfying option to demo and the one that decides whether the other two work. It also makes a later swap of front end or vendor a routing change rather than a rebuild.

Most organizations start with the first and discover they needed the third. The gap is usually a data model, not a product.

Four questions for the approval packet

  • Which populations does this cover, and who serves the rest? Name contractors, seasonal staff, pre-hires and alumni in the requirement. Write the answer into acceptance criteria, not a footnote.
  • Who owns each grant? Every security domain in the permission table needs a named owner and a review date, not a group mailbox.
  • What happens when a worker record changes? Manager moves, cost center changes and rehires are the cases where a portal shows its age fastest. Ask for the handling rule before go-live.
  • What is the measured baseline? Queue volume, request mix and time to resolve, recorded before launch. Without it, the benefit is an opinion.

A 30-day check

  • Days 1 to 5. Pull the worker population by type from the HR system. Count employees separately from contractors, seasonal staff and pre-hires.
  • Days 6 to 10. Map the domains the portal needs to the teams that can grant them. The gaps are the schedule.
  • Days 11 to 15. Run ten real requests through today's process and time each step. That is the baseline the portal will be judged against.
  • Days 16 to 22. Pick five fields the portal will display and reconcile them to the source record. Count the mismatches; that number is your data-cleanup scope.
  • Days 23 to 30. Write and test the fallback path for one uncovered population. If that path has no owner, the business case is incomplete.

What remains unestablished

  • Whether the simplified Workday integration path changes the employee-only population limit. The integration page states the limit without distinguishing between paths.
  • What the packaged agent costs to operate across HR, identity and network teams — review capacity, tooling and staff time are not quantified.
  • Whether Microsoft's Northern Trust figures hold beyond the onboarding workflow and test period described.
  • Whether any organization measured queue volume, request mix or resolution time after deploying a portal of this kind.
  • How the domain permission table changes for portals that add scheduling, expenses or learning, each of which may require different APIs.
  • Market data. None was reviewed for this edition, so no stock, index or valuation figures are reported.

Evidence notes and limits

Microsoft Learn's "Integrate Workday with your Employee Self-Service deployment" is the source for the quoted description of the agent as built on Copilot and taking actions on HR data; the quoted statement that the Workday integration is currently configured to get only Employee details and does not work for External staff or Non-Employees; the quoted descriptions of the legacy path using two integration system users plus a RaaS report and of the simplified path reducing Power Platform connections from four to two; and the eight Workday administrator activities listed in the deployment-roles table. The permission tables, credential handoff and network items are reported as paraphrase attributed to the page generally rather than as quotation: the two integration security groups and the domains described above, the note that those configurations are basic requirements with additional permissions possible depending on implemented topics, the direction to share the client ID and endpoints with Microsoft Entra administrators, the requirement that Workday usernames line up with Entra accounts for OAuth, and the coordination with information security over permitted hosts. The page was retrieved under a localized path that served the English body; the same document appears under other locale paths and is a single source, cited once.

Microsoft Learn's Workday administrator reference for the Employee Self-Service declarative agent is the source, in paraphrase, for granting access at the integration-permission level rather than the report or task level, for the observation that the domain governing a given data element depends on tenant configuration, for handing API client values and endpoints to the maker through an approved channel, and for directing organizations that filter outbound traffic to give the REST and SOAP host names to the network administrator. It was retrieved under a localized path serving a non-English body and is cited only for those configuration details, which were read from the localized text and restated in English as paraphrase.

Microsoft's "From AI ambition to business impact," published November 12, 2026, is the source for the quoted Northern Trust passage including the 83% data preparation reduction, the move from about two and a half weeks to three days, the 96.5% first-pass quality score in testing, and the statement that analysts remain in control; for the described 3M order-management workflow and its "expected to" framing; and for the Publicis deployment to 114,000 employees. It is Microsoft describing its own customer engagements; the figures are company-stated and were not independently audited.

Microsoft's "Scaling AI value without losing control: Lessons from Frontier Firms," published November 12, 2026, is the source for the quoted 61% versus 93% measurement-confidence finding, the 66% versus 48% scaling finding, and the quoted disclosure that returns are leader-reported and show association rather than cause. It is vendor-commissioned, leader-reported research reviewed at the level of its published findings.

No audit, contract, pricing page or third-party measurement was reviewed. The population-gap argument, the data-quality argument, the three scoping options, the four questions and the 30-day check are this publication's analysis. Nothing here establishes that any organization reduced cost, recovered staff capacity or improved service levels.

The advertisement above is separate from this analysis and is not a source for any claim in it.