Skip to content

Policy Q&A Assistant

Example prompt: "When someone messages our #ask-ops Slack channel with a question, search our Notion policy wiki for the answer and reply in-thread with the answer and a link to the source page."

The Problem

Operations, HR, and IT teams spend hours each week answering the same policy questions. "How many days of parental leave do I get?" "What is the expense limit for client dinners?" "Can I buy a new monitor?" The answers all live in a well-maintained wiki, but nobody reads the wiki. Questions get asked in Slack, DMs, or over a shoulder, and the same answer gets retyped repeatedly, sometimes differently each time. Worse, when the policy changes, stale answers keep circulating because the original Slack thread is never updated.

How GloriaMundo Solves It

We build a workflow triggered by new messages in a dedicated question channel like #ask-ops or #it-help. An LLM step first classifies whether the message is a policy question or small talk, so the bot only replies when useful. If it is a question, an integration step queries the Notion or Confluence wiki for relevant pages. A second LLM step reads the top matching pages and drafts an answer that is grounded in the source — it does not make up policy, and it includes a link to the exact page it cited. A conditional step adds a confidence check: if the match is weak, the bot posts a "not sure — escalating to the team" message and pings the ops on-call rota instead of guessing. All drafts can be previewed in Glass Box while the workflow is being tuned.

Example Workflow Steps

  1. Trigger (integration): Fires when a new message is posted in the #ask-ops Slack channel.
  2. Step 1 (LLM): Classify whether the message is a policy question, a reply, or off-topic.
  3. Step 2 (conditional): If it is a question, continue. Otherwise, stop.
  4. Step 3 (integration): Search the Notion policy wiki for pages matching the question.
  5. Step 4 (LLM): Read the top matches and draft an answer with a direct citation.
  6. Step 5 (conditional): If confidence is low, escalate to a human reviewer instead of replying.
  7. Step 6 (integration): Post the grounded answer as a threaded reply in Slack with the source link.

Integrations Used

  • Slack — receives questions and posts answers threaded to the original message
  • Notion — searchable policy wiki and source of truth for answers
  • Confluence — alternative wiki source if Notion is not used
  • PagerDuty — optional escalation when no confident answer is available

Who This Is For

Operations, HR, and IT teams at companies with 30-500 employees where the same policy questions arrive daily and maintaining a wiki only helps if people actually get directed to it.

Time & Cost Saved

Answering policy questions in Slack typically consumes 30-60 minutes a day across an ops or HR team. More importantly, it pulls focus from work that actually needs human judgement. This workflow deflects the bulk of repeat questions while making the source document the canonical answer — so when the policy changes, everyone starts getting the updated answer automatically. For a 100-person company, that reclaims 3-5 hours per week for the ops team. The workflow uses LLM and integration steps and costs a small number of credits per question answered.