Management Cortex
Every HR Question Answered — From the Handbook, by HR, or Routed Cleanly.
The Cortex answers what the handbook covers. Routes the rest to HR. Builds a validated FAQ from every answer HR gives.
What it solves
HR teams in small and mid-sized service companies spend a substantial share of their week answering the same questions they answered last week. PTO policy. Benefits enrollment. Expense submission. Parental leave. Performance review timing.
The handbook contains the answer to most of them. Nobody reads the handbook. Employees Slack HR instead — easier, faster, more conversational.
The shortcut everyone reaches for — a chatbot on the handbook — produces confidently-wrong answers when the policy is nuanced. The cost of a wrong answer on something like leave eligibility or expense reimbursement is high.
How It Works
The Cortex has access to the company's HR documents in extracted, queryable form — handbook, policies, benefits summaries, recent updates.
When an employee asks a question, the Cortex runs two-stage retrieval. First it checks the validated FAQ for a confirmed answer. If found, it returns the cached answer with citation.
If not in the FAQ, it loads the relevant policy documents in full and composes an answer — only if it can cite confidently. If the docs don't address the case or address it ambiguously, the Cortex responds: "I'll route this to HR" and creates a delegation record.
HR reviews delegations on their own schedule, answers them in their own voice, and the answer flows back to the employee.
Every HR answer is automatically added to the validated FAQ — citation-grade, because the SME wrote it. The next time the same question comes in, the Cortex answers from the FAQ.
The cost curve flattens: month one ~100% of questions reach HR; month six ~30% reach HR, the rest answer from FAQ.
The model is biased toward over-delegation early. The cost of a confident wrong answer on HR is higher than the cost of HR receiving a delegation that turned out to be answerable. Trust calibrates upward as the FAQ accumulates.
What You Get
- HR document extraction and ingestion (handbook, policies, benefits summaries)
- Two-stage retrieval (FAQ first, full-document fallback) with citation-grade answers
- Delegation workflow — pending queue, full-context routing to HR
- Automatic FAQ promotion from every HR-answered delegation
- Full-document loading for confident composition (no RAG chunking)
- Conversational interface in the channel employees already use
- Usage analytics — what's being asked, what's being delegated, where the FAQ has gaps
In practice
A 60-person service company deploys the HR Assistant in Microsoft Teams. Month one: the Cortex handles 60% of questions (the easy ones — PTO accrual, open enrollment dates), delegates 40% to HR. The HR team answers delegations once per day in batches; each answer flows back to the employee and is added to the FAQ. Month six: the Cortex handles 85% of questions; HR's delegation queue is 4–6 items per week instead of 40–60. The HR team's time on repeat questions drops 75%. The newly-freed time goes to strategic work — onboarding redesign, performance review system improvements, retention analysis.