How to Enable AI Reply Suggestions in NeuraDesk
This guide walks you through enabling AI reply suggestions in NeuraDesk, covering prerequisites, setup steps, and troubleshooting tips so your team can respond faster.
On this page
Enabling AI reply suggestions lets your support team draft responses faster by showing context-aware drafts right inside each ticket. Before you start, you need an active account with admin access and at least a few dozen historical tickets so the model has data to learn from. The feature is included free on every plan, and setup takes only a few minutes.
Step 1: Sign In and Open the Admin Panel
- Go to the login page and sign in with your admin credentials.
- Click your avatar in the top-right corner and choose Settings from the dropdown.
- In the left sidebar, select AI & Automation.
Why this matters: only workspace admins can change AI settings. If you don’t see the AI & Automation section, confirm your role with the account owner.
Step 2: Locate the AI Reply Suggestions Toggle
- Inside AI & Automation, find the card labeled AI Reply Suggestions.
- The card shows a brief status summary: whether the feature is on, how many suggestions were generated last week, and the current learning mode.
Why this matters: this screen is the single control point for the feature. All configuration — enabling, disabling, and behavior options — lives here.
Step 3: Enable the Feature
- Flip the Enable AI Reply Suggestions switch to On.
- Choose a suggestion behavior:
- Auto-show — suggestions appear automatically when an agent opens a ticket.
- On demand — agents click a “Show suggestion” button to see a draft.
- Click Save changes at the bottom of the card.
Why this matters: the toggle activates the model for your workspace. The behavior setting lets you match the workflow your team prefers — some agents like suggestions upfront, others prefer to request them only when stuck.
Step 4: Review and Adjust Learning Preferences
- Still on the AI & Automation page, scroll to Learning preferences.
- Select one of three modes:
- Learn from all replies — the model updates after every agent reply.
- Learn from accepted suggestions only — the model only trains when an agent inserts a suggestion without major edits.
- Pause learning — the model stops adapting (useful during onboarding or policy changes).
- Save again.
Why this matters: learning mode controls how quickly suggestions improve. For most teams, Learn from accepted suggestions only balances relevance with safety. You can read more about how the model adapts in How AI Reply Suggestions Learn From Your Replies.
Step 5: Test with a Real Ticket
- Open any open ticket in your shared inbox.
- If you chose Auto-show, a suggestion banner should appear above the reply box within a second.
- If you chose On demand, click the Show suggestion button that now sits next to the Reply button.
- Inspect the draft: it should reference the customer’s last message and your internal knowledge (if any).
- Optionally insert the suggestion, edit, and send.
Why this matters: a quick test confirms the pipeline works end-to-end — ticket ingestion, model inference, and UI rendering. If nothing appears, see the troubleshooting section below.
Common Mistakes to Avoid
- Expecting suggestions on a brand-new workspace — the model needs historical tickets to generate relevant drafts. Import or forward at least 50–100 past conversations before enabling.
- Forgetting to click Save changes — the toggle and learning mode each require a separate save; the UI will warn you if you navigate away with unsaved changes.
- Choosing “Learn from all replies” too early — if agents frequently rewrite suggestions, the model can learn bad patterns. Start with Learn from accepted suggestions only.
- Assuming suggestions appear on every ticket — the system only shows a draft when confidence exceeds an internal threshold. Low-confidence tickets (e.g., very short or ambiguous messages) will show no suggestion.
- Overlooking the “Pause learning” mode — useful when you’re rolling out a new refund policy or product launch and don’t want outdated replies to influence the model.
Troubleshooting
No suggestion banner appears
- Verify the feature is enabled and saved.
- Confirm the ticket has enough context (at least two customer messages).
- Check that the ticket isn’t marked as spam or archived.
- If the issue persists, see Why AI Suggestions Are Not Appearing.
Suggestions are generic or off-topic
- Increase the volume of training data by ensuring more tickets are resolved in the platform.
- Switch learning mode to Learn from accepted suggestions only for a week.
- Review the model’s recent performance in the AI & Automation dashboard (shows acceptance rate and average edit distance).
Agents don’t see the “Show suggestion” button
- Ensure they’re using a supported browser (Chrome, Firefox, Safari, Edge latest versions).
- Clear browser cache or try an incognito window.
- Confirm the agent’s role includes “Reply to tickets” permission.
Next Steps
Once suggestions are live, treat the first two weeks as a calibration period. Encourage agents to accept or dismiss each suggestion — that feedback loop is what sharpens the model. For a structured rollout, refer to Best Practices for Testing AI Reply Suggestions Pre‑Launch.
Ready to speed up your support replies? try NeuraDesk free
Join teams already using NeuraDesk
Ready to get started?
Create your workspace and try NeuraDesk for free.
Sign up free