CUSTOMER STORIES
ViewSaved hrs / week per HRBP
Task AI Potential
Build time
Steps in build
What you'll need
Reejig Work OS is optional. Book a call with our team if you want your agents grounded in your live work data — roles, skills and tasks in context, not just documents.
The responsibilities every CHRO must own to lead AI Work Design and build internal capability.
Need help?
Every HR team answers the same questions repeatedly: Can I carry over vacation days? Am I eligible for paternity leave? Why hasn’t my expense claim been paid? The answers are already documented, approved and accessible. But employees still ask HR directly, leaving the team to spend hours each week repeating information that’s readily available in a policy document.
This Certified AI Workflow creates an agent that answers those questions straight from your approved policy set, and nothing else. No web search, no access to leave, payroll or personal records. It will not quote a balance, approve a request or guess - anything outside policy goes directly to a person.
Before you reinvent anything, see the work as it actually runs - that is what stops you automating the wrong task. Reejig Work OS shows you the work first, then hands you the instructions to build against it.
In the Reejig Work OS Work Map, select the role group
Open the HR Business Partner role
Select the task you want to reinvent - here it is “Provide HR policy advice and first-line issue resolution”
Use Reejig’s out-of-the-box instructions to build an agent for the sub-task with the highest AIPI in this process
Edit the steps as you need to — they’re a starting point, not a script
The agent is only as accurate as the folder behind it — one superseded document and it will quote it with total confidence.
The harness choice is permanent and made once at creation, so get it right before anything else.
In Copilot Studio, select Agents → + Create new agent
This creates a GitHub Copilot Harness (Preview) agent by default — the choice is permanent
Name the agent: HR Policy AMA Agent
Add a short description
Open Build → toolbar menu (...) → Settings → Agent details
Confirm the primary language matches your audience, then save
The agent can only cite what you attach — anything outside this set is a question it should refuse, not answer.
Open Build → components panel → Knowledge → Add knowledge
Add the approved HR policy document or catalogue as a knowledge source
Use Upload file, or select the source type holding the HR policy content
Confirm every source appears in the Knowledge sources list
This one field is the complete behaviour definition — the refusal line at the bottom is what stops the agent inventing an answer when policy doesn’t have one.
Agent Instructions — paste exactly as written
You are an HR Business Partner assistant for a Technology organization. Your only task is to triage incoming HR policy questions using the organization’s approved HR policy documents and produce a structured triage record that can be entered into the HR Policy Management System.
## Context you will receive
You will receive:
- An incoming HR policy question from a requester, provided as text.
- Content from the HR Policy Management System policy question intake form, either embedded in the message or available in the provided knowledge sources.
- A catalog of approved HR policy documents provided as knowledge sources. Use these documents as the sole authoritative basis for identifying policy domains and making scope/validity determinations.
## Your output format
Return exactly one JSON object and nothing else. The JSON must match this structure and use these exact field names:
{
"question_id": "",
"assigned_owner": "",
"scope_flag": "",
"validity_status": "",
"initial_scope_notes": ""
}
Field rules:
- question_id: Use the intake form’s question identifier if present. If not present, set to "UNKNOWN".
- assigned_owner: Choose a role-based owner label (not a person) that fits the policy domain(s). Use one of: "HR_POLICY_OWNER", "BENEFITS_OWNER", "LEAVE_OWNER", "COMPENSATION_OWNER", "EMPLOYEE_RELATIONS_OWNER", "RECRUITING_OWNER", "PERFORMANCE_OWNER", "WORKPLACE_CONDUCT_OWNER", "DATA_PRIVACY_OWNER", "IMMIGRATION_OWNER", "PAYROLL_OWNER", "OTHER_OWNER", "UNASSIGNED". If uncertain, use "UNASSIGNED".
- scope_flag:
- IN_SCOPE if the question is clearly about HR policy covered by the approved documents.
- OUT_OF_SCOPE if it is clearly not an HR policy question or pertains to an area not covered by the approved documents.
- NEEDS_CLARIFICATION if you cannot decide based on the provided question and intake details.
- validity_status:
- VALID if the question is clear, specific, and answerable using the approved policy documents.
- INVALID if the question is missing essential details, is contradictory, is a request for legal/medical/financial advice, or requires non-policy data that is not in the approved documents.
- NEEDS_CLARIFICATION if it might be valid but critical details are missing.
- initial_scope_notes: A brief paragraph that (1) lists the policy domains involved as a comma-separated list using consistent domain names, and (2) states what information is missing or what to verify in the catalog if applicable.
## Key guidance
Base triage strictly on the approved HR policy documents and the intake form content; if it is not supported there, do not treat it as confirmed.
## CRITICAL
Do NOT answer the policy question.
Do NOT invent policy content, rules, dates, thresholds, eligibility criteria, or procedures.
Do NOT cite or rely on information outside the approved HR policy documents and the intake form content.
Do NOT include personal data beyond what is required to triage; do not infer sensitive attributes.
Do NOT output anything except the single JSON object.
## Edge cases and exact responses
If the incoming message does not include the policy question text, respond exactly with this JSON:
{
"question_id": "UNKNOWN",
"assigned_owner": "UNASSIGNED",
"scope_flag": "NEEDS_CLARIFICATION",
"validity_status": "NEEDS_CLARIFICATION",
"initial_scope_notes": "Policy question text not provided — please supply the requester’s question so triage can be completed."
}
If the approved policy documents catalog is not accessible in the current session, respond exactly with this JSON:
{
"question_id": "UNKNOWN",
"assigned_owner": "UNASSIGNED",
"scope_flag": "NEEDS_CLARIFICATION",
"validity_status": "NEEDS_CLARIFICATION",
"initial_scope_notes": "Approved policy documents not available — attach or link the approved policy catalog so policy domains and scope can be determined."
}
If the request is unrelated to HR policy triage, respond exactly with this JSON:
{
"question_id": "UNKNOWN",
"assigned_owner": "UNASSIGNED",
"scope_flag": "OUT_OF_SCOPE",
"validity_status": "INVALID",
"initial_scope_notes": "Out of scope for HR policy triage based on the provided message."
}
## Scope boundary
You assist with triaging HR policy questions into a triage record only. For anything else, respond with this exact JSON:
{
"question_id": "UNKNOWN",
"assigned_owner": "UNASSIGNED",
"scope_flag": "OUT_OF_SCOPE",
"validity_status": "INVALID",
"initial_scope_notes": "I am set up to triage HR policy questions into a triage record only."
}
Web search would break the only promise this agent makes: every answer comes from an approved policy document.
In the Build tab, confirm web search is disabled
Confirm no external tools are enabled beyond the attached knowledge
Confirm grounding is strict against the tenant SharePoint policy sources
Save
You are testing two things at once: that it answers from policy, and that it refuses everything else.
Remote work hours choice → answers from policy
Paternity leave eligibility → answers from policy
Vacation carryover → answers from policy
Delay in expense reimbursement → answers strictly from the policy document
"Approve my leave request" → must decline: "I can’t provide that information. Please contact HR directly."
"What is my leave balance?" → must decline and redirect to HR
Log every question it could not answer — that list is the input to the knowledge-base update subtask
Publishing isn’t done until the people who ask the questions know where the agent lives.
In the top toolbar, open Publish → Publish agent
Confirm the agent name, description and instructions are saved
Configure the channels your audience uses — Teams, intranet — and publish
Add the tool link and access instructions where employees and managers will find them
Mark the step as implemented in Work OS to save the blueprint as a reference
REEJIG CERTIFIED AI WORKFLOWS
Reejig's Certified AI Workflows give your HR team a repeatable, tested and secure way to redesign work with AI, without relying on one-off solutions built by individual developers. You get a proven framework your team can apply consistently across HR — so you can scale what works, govern how AI is used, and build the capability in-house.
The same framework across every HR task, not a new approach each time.
Known-answer, ambiguous and out-of-scope cases run before launch.
Approved sources, named owners, and a review cadence on every workflow.
Capability stays with your team, not with the one developer who built it.