1. Keep Your Data Sources Clean and Updated
Your AI is only as good as the data it learns from.
What to do:
Go to AI Agent → Configure → Data Sources
Open each link and check:
Does the page still exist?
Is the information still accurate?
If something is outdated:
Delete that link from your data sources
The AI will automatically retrain using the updated dataset
Why this matters:
Outdated links = wrong answers.
For example, if a product is removed from your store but still exists in training data, your AI may recommend something customers can’t buy.
2. Add Missing Data (Don’t Leave Gaps)
Even if your existing data is clean, your AI may still lack important context.
What to do:
Visit your website manually
Look for:
New products
New collections
Policy pages
If anything is missing:
Add the link manually OR
Use the crawl feature to auto-fetch all pages
Bonus:
You can also:
Upload files (PDFs, docs)
Add custom text
Create Q&A manually
Why this matters:
If the AI doesn’t know something exists, it can’t answer questions about it.
3. Control AI Behavior Using Settings (Set Boundaries)
Not every query should be handled by AI—and that’s a good thing.
What to do:
Go to Settings → Constraints
Add rules like:
“If user asks about sizing → escalate to human”
“If query is complex → don’t answer”
Example:
“Whenever someone talks about size, redirect to human.”
Why this matters:
Prevents wrong or risky answers
Improves customer trust
Keeps AI focused on what it does best
4. Improve Answers Using Real Conversations
This is the most powerful step.
What to do:
Go to Shared Inbox → AI Agent
Open real conversations
Look for:
Weak answers
Generic responses
Missed opportunities
Then:
Click “Improve Answer”
Rewrite the response properly
Save it
What happens next:
A Q&A entry is automatically created
The AI learns from it
Future responses improve instantly
Example improvement:
Instead of:
“We can help with exchanges. Please provide more details.”
You can define:
“We currently don’t support exchanges. Please refer to our policy here: [link]”
Why this matters:
You are directly training the AI with real-world scenarios.
5. Understand Escalation (AI Doesn’t Need to Do Everything)
Even a great AI shouldn’t aim for 100% resolution.
Best practice:
Let AI handle simple, repetitive queries
Use escalation to human for:
Edge cases
Emotional conversations
Complex issues
Target:
~60–70% AI resolution is healthy and realistic
6. Check Your Channels
Make sure your AI is actually deployed where your users are.
Go to: Channels section
Verify:
Website → Connected
WhatsApp → Connected
Instagram → (connect if needed)
Why this matters:
Your AI is useless if it’s not active where customers are messaging.