Ask AI — What It Can Do and How to Govern It

Ask AI — What It Can Do and How to Govern It

1. What it is

Ask AI is the platform’s conversational assistant. Each question is routed to a specialist agent that is wired into the app it covers — scheduling for shift questions, leave management for PTO balances, service desk for ticket status — so answers come from live data, not just documents. When a question spans two areas, one agent hands the conversation to another mid-turn.

• Enablement: Ask AI is a licensed app that requires the Standard platform tier or above. An admin enables it per tenant at Apps → Ask AI → Settings; it is not on by default.
• What it is not: Ask AI is the assistant your employees talk to. It is not the place where you pause AI or manage approvals — those kill switches live in Admin → AI Operations.

2. Standing it up

1. Go to Apps → Ask AI → Settings (or Admin → App Marketplace → Ask AI → Configure). Enable the app and confirm that Enable Action Execution, Enable General Questions, and Enable Schedule Queries are on — all three default to on, but toggling any one off silently removes that class of answer.
2. Open Ask AI → Admin → Agent Guidelines → Context and write your Business Context — the who-we-are paragraph every agent prepends to its answers. Without it, agents answer accurately but generically. Save as a draft first; publish when ready.
3. On the AI Features tab under Agent Guidelines, confirm that Multi-Agent and Shared Knowledge are on. Then review the knowledge sources listed below the toggle — Knowledge Base, HR Policies, SOPs, Sites, Wikis, and Drive. Every source defaults to on; turn off any that should not appear in AI answers.
4. Verify the rollout by opening the Test Suite (available to admins and managers). Pick an agent, type a representative question, and confirm the answer reflects your published guidelines and knowledge sources.

Silent pairing to watch for: enabling Ask AI alone makes the chat surface available, but individual agents only answer when their underlying app is also enabled and licensed. An employee asking a scheduling question gets “I can’t help with that” if the Shifts app is not turned on for the tenant.

3. How it fits together

Agents and apps

Every marketplace app that has an agent contributes one (or sometimes two — a read-only help agent and an admin-operations agent) to the pool. The platform ships about 100 built-in agents; apps built with App Builder or the plugin system add their own automatically. An agent is available to a user only when the app it belongs to is enabled for the tenant and published to that user’s audience.

Guidelines and overrides

Agent Guidelines are the instructions every agent follows — your business context, tone, guardrails, and action rules. They follow a draft/publish workflow: saving a draft stages changes without affecting live agents; publishing promotes them immediately. Guidelines are organized into five tabs:

• Context — business context, custom instructions, universal guardrails, and glossary terms.
• Communication — tone, greeting, apology and escalation messages, and response format.
• Actions — action-specific guardrails, notification rules, and whether the agent shows its reasoning.
• Overrides — per-agent tweaks that layer on top of the global guidelines (extra instructions, tone, guardrails).
• AI Features — org-level toggles for multi-agent routing, shared knowledge, journeys, reasoning, and the planner.

Knowledge layers

Ask AI draws answers from three layers, each governed differently:

• App data — the live records inside each app (shifts, tickets, leave balances). Governed by the app’s own permissions; the agent sees only what the asking user can see.
• Shared knowledge — your Knowledge Base articles, HR Policies, SOPs, Sites, Wikis, and Drive files. Admins control which of these six sources are searchable on the AI Features tab.
• Verified Knowledge — corrections and curated Q&A pairs that an admin has approved. These override retrieval when a matching entry exists. Managed at Ask AI → Admin → Verified Knowledge.

Tool confirmation

When an agent is about to do something that changes data — delete a record, submit a form, send a notification — it pauses and asks the user to confirm before proceeding. The admin can disable all write actions globally by turning off Enable Action Execution in Settings.

4. Running it

Publishing or updating guidelines

1. Open Ask AI → Admin → Agent Guidelines and select the tab to edit.
2. Make changes and click Save Draft. The draft is staged but agents continue running the published version.
3. When ready, click Publish. Changes apply to all agents immediately.
4. To undo a draft before publishing, click Discard Draft — the tab reverts to the published values.

Reviewing verified knowledge

1. Open Ask AI → Admin → Verified Knowledge. New entries appear on the Pending tab.
2. Review each entry. Approve to make it live (Ask AI will use it in answers), or Reject to discard it.
3. Active entries can be edited in place or Retired when they go stale.

Running tests against your guidelines

1. Open Ask AI → Admin → Test Suite. The page shows whether guidelines are published or still on defaults, and flags any unpublished draft.
2. Select an agent, type a question, and run the test. The result shows the agent’s answer under the current published guidelines.

Managers can also access the Test Suite; all other admin surfaces require admin or above.

Handling a data request

1. Open Ask AI → Admin → Data Requests.
2. Select the employee. Choose Export for an Article 15/20 access request — this downloads a JSON file of all their Ask AI records for this tenant.
3. For an Article 17 erasure request, choose Erase (permanently deletes all records) or Redact (keeps the structural record but removes the content). Both are irreversible.

Data requests are scoped to the current tenant. An employee with memberships in multiple tenants requires a separate request from each.

Customizing the assistant’s identity in Chat

Go to Chat → Settings → Ask AI Identity. Admins can change the assistant’s name, description, and avatar. This controls how Ask AI appears inside the Chat app.

5. Settings

Settings page (Ask AI → Admin → Settings)

Setting	Default	What it changes
Enable Action Execution	On	When off, agents answer questions but never write data or take actions.
Enable General Questions	On	When off, agents do not answer “how do I…” platform questions.
Enable Schedule Queries	On	When off, the scheduling agent does not answer shift questions.
Include Platform Help Articles in Answers	Off	When on, agents also search the platform’s own help documentation — not just your company’s content.
Enable Voice Mode	On	Allows employees to speak to Ask AI instead of typing.
Voice Daily Limit	30 minutes per user	The maximum voice minutes per user per day. Set to 0 for unlimited.
Allow users to clear their conversation	On	When on, users can clear their visible conversation history. Clearing is cosmetic — the record stays available for governance until retention removes it.
Keep conversation history for	24 months	Conversations older than this are permanently deleted on the next nightly sweep. Shortening this removes history immediately and cannot be undone.

AI Features tab (Ask AI → Admin → Agent Guidelines → AI Features)

Setting	Default	What it changes
Multi-Agent	On	Allows one agent to hand a question to another when it spans two areas.
Shared Knowledge	On	Agents search your company’s knowledge content (KB articles, HR policies, SOPs, Sites, Wikis, Drive). The per-source toggles below this control which of the six sources are included.
App Help	On	Agents search in-app help articles when answering.
Journeys	On	Enables multi-step goal tracking for employees. When on, a My Journeys tab appears.
Session Mining	Off	When on, the system mines past conversations to surface patterns.
Reasoning Loop	On	Agents use a multi-step reasoning process for complex questions.
Planner	Off	Enables agent work plans for multi-step tasks.

Moderation (platform-wide, not per-tenant)

Ask AI enforces a rate limit of 120 messages per user per hour. Every message is checked for offensive content, prompt injection attempts, and spam. Admins are not exempt from these checks.

6. More help

• Ask AI FAQ — specific setup and operating questions, limits, and troubleshooting.
• AI Operations — pausing AI, approvals, and agent health (the kill switches).
• Ask AI — the assistant answers questions about itself from these articles.