Building an AI HR Help Desk at Zapier

What they built

• An AI-powered HR help desk bot that automatically responds to employee questions posted in a Slack channel
• The bot reads the full message thread for context, searches existing HR documentation, and posts a relevant, conversational reply directly in the thread
• Includes a sensitive-topic escalation path: if a message flags as sensitive (e.g., harassment), the bot notifies the user it's escalating and tags the appropriate HR team (e.g., Employee Relations) rather than auto-responding
• Supports ongoing conversation: the bot re-reads the thread on each new message and continues to provide contextual follow-up answers
• Built in Zapier using just two core steps - a Slack trigger and an AI action - with no code required

Why they built it

• HR teams were fielding the same routine questions repeatedly via Slack and email (benefits, policies, onboarding, etc.), pulling focus away from higher-value people work
• Employees had to navigate too many documentation sources to find answers, creating friction and delays
• The problem was observed not just at Zapier but across nearly every company Emily partners with as an AI automation engineer
• Zapier's team specifically needed coverage during reduced-capacity periods (e.g., when staff are at conferences or OOO)

Tools used

• Zapier - automation platform used to build and connect the workflow
• Slack - trigger source (new message posted to an HR channel); also used for bot replies and Slack search to surface previously answered questions
• Google Docs (or Notion, Airtable, websites) - existing HR documentation linked as knowledge sources (e.g., internal mobility policy, laptop provisioning guide, international mobility guide)
• Zapier AI action (agent step) - reads thread context, searches knowledge sources, drafts and posts the reply
• Zapier Paths - routes messages down different branches depending on sensitivity or topic type

How you could build this yourself

1. Identify your trigger channel - create or designate a Slack channel (or Teams equivalent) where employees ask HR questions
2. Set up your Zapier trigger - use "New message posted to channel" in Slack as step one; for private/sensitive use cases, use a "pushed message" trigger from DMs instead
3. Add an AI agent action step - write plain-language instructions telling the AI to:
   • Read the full Slack thread (not just the triggering message) for context
   • Search your linked HR documents for relevant answers
   • Return a concise, relevant reply - not a data dump
   • Escalate and tag the right HR team if the message appears sensitive
4. Link your existing documentation - connect Google Docs, Notion pages, Airtable bases, or internal websites you already use; no need to rebuild a new knowledge base
5. Add Zapier Paths (optional but recommended) - create a "sensitive" path that logs to your HRIS, notifies Employee Relations, or adds to a tracking database; create a "standard" path for routine auto-replies
6. Enable Slack search as a tool - give the AI access to Slack search so it can surface answers your team has already given in other threads
7. Test with a live demo question before turning it on fully - verify tone, accuracy, and escalation logic
8. Turn it on and iterate - when documentation gaps appear (bot can't answer), update your docs; the knowledge loop improves over time
9. Start here before touching sensitive data - this is a low-risk first automation; avoid connecting comp, performance, or ER data until your team is comfortable with the tooling

Source

Built by Emily Mabie, Senior AI Automation Engineer, People Team at Zapier, and presented as part of an Open Org "Shipped by HR" session on building an AI HR Help Desk.