The ticket nobody needed to file.
“What's the parental leave policy?” “What's my cost centre?” Questions like these are the background radiation of every HR and IT function — high volume, low value, endlessly repeated. The Employee Self-Service agent is Microsoft's answer, and unusually, Microsoft publishes the actual per-query economics. This lesson is built around those numbers, because they decide whether the thing is worth deploying.
01 What it is, structurally
The Employee Self-Service agent handles employee queries and requests on key HR and IT topics inside Microsoft Copilot and Copilot Chat. Architecturally it is built on Copilot Studio and uses agent flows — which is why its prerequisites read like a Power Platform project rather than a Copilot toggle. It leans on Microsoft Entra for authentication, and Microsoft recommends configuring single sign-on in Entra because the agent trusts Entra.
The agent includes the Workday, SAP SuccessFactors and ServiceNow premium connectors at no additional cost. That is a real saving. You still need the appropriate licences for those third-party systems themselves — the connector is free, the system it connects to is not.
02 The economics nobody reads
Here is the part that makes this lesson worth its tier. Users with a Microsoft Copilot licence are covered. Users without one are billed pay-as-you-go, and Microsoft publishes a worked cost breakdown for roughly a thousand users:
| Query type | Example | Share of traffic | Cost per query, unlicensed user |
|---|---|---|---|
| Knowledge search, no cached user context | “What is the parental leave policy?” | 81% | 16 cents |
| Knowledge search, cached user context | Same question, context cached | 2% | 12 cents |
| Topics | “What is my cost center #?” | 14% | 5 cents |
| Verbatims | Sensitive statements routed to a fixed response | 3% | 1 cent |
Microsoft's own summary: the average price per user without a Microsoft Copilot licence is 15 cents per query. The reason knowledge search dominates the cost is visible in the step breakdown — the tenant call for content is 10 cents and the orchestrator summarising that content is 2 more, while determining intent is free. Non-LLM answers cost a cent.
03 How much volume should you actually expect?
Microsoft publishes two real deployment samples, both at 40,000 enabled users, and they are a useful reality check against the business case someone is about to write:
| Metric | Sample 1 (28 days) | Sample 2 (two months) |
|---|---|---|
| Enabled users | 40,000 | 40,000 |
| Monthly active users | 1,100 — 3% of activated | 2,075 — 5% of activated |
| Total conversations | 5,129 | 11,000 |
| Total interactions | 6,917 | 16,000 |
| Interactions per conversation | 1.34 | 1.45 |
| Conversations per user | ~5 | ~5 |
Read that honestly. Three to five percent monthly active of an enabled population, and conversations that are barely more than a single exchange. This is a genuinely useful tool with modest adoption — which is fine, as long as nobody promised the board that 40,000 people would stop emailing HR.
04 The deployment is a programme, not a switch
Microsoft's own role table is the clearest signal of scale. Deploying this properly needs a Global admin to assign roles, a Power Platform administrator to create environments and install the agent, a Power Platform maker to configure it (Microsoft recommends the agent owner does this), external system administrators for the third-party integrations, an information security team to allowlist inbound requests and manage SSO, and a change control board to approve the architecture, the data governance and the responsible-AI policy.
Three more roles do no configuration but decide whether it works at all: HR, IT and legal/privacy representatives. Their job is to identify knowledge sources, supply the frequent queries, and flag sensitive topics. That last one is why the “verbatims” category exists in the cost table — some questions must never get a generated answer.
Microsoft's guidance is to use the least privileged role for each activity and use just-in-time access for elevated ones. Standing Global Admin for an HR chatbot project is the kind of thing that shows up in an audit two years later.
05 Before you start
Readiness checklist
- Count your unlicensed users. At roughly 15 cents a query, they are the entire variable cost. Multiply by 5 conversations a year per active user, at 3–5% monthly active.
- Set up Copilot Studio capacity — the agent uses agent flows, so this is not optional. Prepaid message capacity is the alternative to pay-as-you-go.
- Configure pay-as-you-go in the Power Platform admin center: create a billing plan, link an Azure subscription and resource group, select the environment, and choose Copilot Studio as the product. Microsoft explicitly warns this last step is critical and isn't covered in the admin center itself.
- Federate identity to Entra and configure SSO before you integrate Workday, SuccessFactors or ServiceNow.
- Get HR, IT and legal to write the sensitive-topics list before go-live, not after the first incident.
- Run FlightCheck — Microsoft's readiness check that validates your tenant, environment and external-system configuration against the agent's requirements.
If that list looks heavier than you expected, that's the honest answer: this is the most enterprise of Microsoft's built-in agents. For a lighter grounded-answers approach over policy documents, SharePoint agents get you most of the way for a fraction of the effort. And whichever you choose, it still has to clear the Agent Registry approval gate.