HR playbook · AI Employee: Harry

HR Policy & Benefits Q&A

80% of HR questions self-served

The problem

Employees have simple HR questions all day — how much PTO do I have, what's the parental leave policy, does our health plan cover physio, can I expense my home internet. Every question routed through a ticket costs HR a touch, and every unanswered question costs an employee 20 minutes of hunting through Confluence. HR shouldn't be an interpretation service for their own handbook.

At a glance
Trigger
Chat
Approvals
None
What it does
Read-only
Systems
HR handbook KB · Benefits KB
How it feels in production

An hour-by-hour walkthrough.

9:47am. Alex DMs Harry: "does our dental plan cover orthodontics for adults?" Harry pulls the benefits guide from Confluence, finds the exact paragraph ("orthodontic care is covered for dependents under 19; adults may enrol in the supplemental orthodontic plan for $22/month"), and answers Alex with the paragraph shown alongside — one click opens the source Confluence page. 9:48am. Alex sees the answer and the source. Doesn't file a ticket. Doesn't ping HR. Moves on. Late afternoon. HR gets the daily digest: 47 questions answered, 3 escalated (ambiguous policy → HRBP), 2 flagged as knowledge gaps ("employees keep asking about X and the handbook doesn't cover it"). The gap list becomes the docs backlog for next week. When Harry can't ground the answer — the handbook doesn't cover it, or two policy docs disagree — he says so explicitly and escalates to HRBP with the question context. Never guesses about benefits or leave, because guessing wrong is a real cost.
How it works

Step by step.

  1. 01

    Answer from HR handbook + benefits KB

    Employee asks in Slack / Teams / portal. Harry retrieves from the indexed handbook + benefits guide + policy PDFs, drafts an answer, shows the source paragraph alongside.

    Slack · Teams · Knowledge Studio · Confluence · SharePoint
  2. 02

    Cite the source paragraph

    Every answer shows the exact paragraph from the source document with a link to open it. Reader can verify in one click; auditors can trace what was said and when.

    Knowledge Studio
  3. 03

    Escalate on ambiguity

    Verdict step abstains when confidence is low or two sources disagree. Escalation to HRBP with the question, the retrieved passages, and a note explaining why Harry declined to answer.

    Slack · HRBP queue
Systems and wiring

What you connect to make this run.

Knowledge Studio

read

Handbook, benefits guide, policy PDFs, HRIS-generated policies indexed. Reference-mode indexing preserves SharePoint / Confluence permissions.

Confluence · SharePoint · Google Drive · PDF upload

read

Native connectors sync policy docs on change. Retrieval is grounded, source paragraph shown, verdict abstains on low confidence.

Slack · Teams

trigger

Bot answers DMs and #ask-hr channel questions. Employees never wait; HRBPs never triage repeat questions.

What changes

Before and after, honestly.

HR ticket volume on policy questions
Before
30-50% of HR tickets are policy Q&A
After
Under 5% (only genuinely ambiguous or non-handbook cases)
Time from question to answer
Before
1-3 business days waiting on HR triage
After
Seconds in Slack
Employee 'gave up looking' rate
Before
Unmeasured; likely high (people don't file tickets for questions they gave up on)
After
Zero — Harry always responds, even if to say "escalating"
Handbook coverage gaps discovered
Before
Discovered when someone asks HR three times in a month
After
Weekly gap report surfaces gaps automatically
Frequently asked

Answers about this playbook.

What if our benefits guide is out of date?

Knowledge Health surfaces stale content. Documents that haven't been updated in a defined window (default 12 months) flag automatically, and the doc owner gets a review nudge. Also: if two policy docs disagree, conflict analysis flags it before an employee reads a wrong answer.

Can Harry answer questions about specific employees' PTO balance?

Yes — Harry reads the requester's own balance from HRIS. He never surfaces another employee's balance without HR-permission scope on the requester.

What about jurisdictions with different policies?

Region-tagged policy indexing. Harry reads the requester's employee location from HRIS and grounds the answer in the applicable region's policy. Explicitly mentions the region in the answer so there's no ambiguity.

How does it handle sensitive personal questions (mental health, harassment)?

Configurable escalation. Certain question patterns (mental-health resources, harassment reporting, whistleblowing) auto-route to an HRBP with the source-links but without the auto-answer. Preserves the human touch on sensitive matters.

Does it work in Teams and Portal, not just Slack?

Yes — same Q&A works in Teams, on the employee portal, in the widget, and via API. One indexed source of truth answers everywhere.

See it run on your data.

Free plan, no credit card. Connect the systems this playbook needs and run it against a past event first.