What You're Building

By the end of this, you'll have a small chat icon in the corner of your website. Customers click it, type their question, and your AI classifies it and either responds automatically or sends it to you on Slack.

No help desk. No ticketing system. No month of configuration. Just a widget, some rules, and a Slack channel.

Step 1: Get the Widget Script

Sign up and grab your widget embed code. It's a single script tag that looks like this:

<script src="https://widget.supp.support/widget.js" data-key="YOUR_API_KEY"></script>

Paste this before the closing </body> tag on your site. If you're on a platform like Webflow, Squarespace, or WordPress, there's a "custom code" section in your site settings where you can drop it in.

If you're running a React or Next.js app, add it to your root layout or _document file.

That's it for installation. The widget appears immediately.

Step 2: Connect Slack

In your Supp dashboard, go to Integrations and connect Slack. Pick a channel (or create one called #support). Every message that needs a human will show up there.

This takes about 2 minutes. OAuth flow — click authorize, pick your channel, done.

Step 3: Create Your First 5 Routing Rules

Go to the routing section. You're going to create rules for your most common question types.

The system already knows 315 support intents — billing, passwords, refunds, order tracking, bug reports, and so on. You just pick the ones that matter for your product and tell the system what to do when they come in.

Here are 5 rules most teams start with:

Rule 1: Password reset

• Intent: password_reset
• Action: Auto-respond with "Here's how to reset your password: [link]"
• Confidence threshold: 80%

Rule 2: Pricing question

• Intent: pricing_inquiry
• Action: Auto-respond with "Check our pricing here: [link]. Plans start at $X/month."
• Confidence threshold: 85%

Rule 3: Bug report

• Intent: bug_report
• Action: Create a GitHub issue + notify Slack channel
• Confidence threshold: 75%

Rule 4: Cancellation request

• Intent: subscription_cancellation
• Action: Auto-respond with cancellation instructions + notify Slack (so you can follow up personally if you want)
• Confidence threshold: 80%

Rule 5: Everything else

• Default action: Notify Slack channel

Five rules. Maybe 10 minutes to set up. These alone will handle 40 to 50% of your incoming messages automatically.

Step 4: Set Confidence Thresholds

Each rule has a confidence threshold. If the classifier is 90% sure a message is a password reset question, the auto-response fires. If it's only 60% sure, the message goes to Slack for a human to look at.

Start conservative. Set thresholds at 80 to 85% for auto-responses. You can lower them later once you see the system performing well.

For actions that have consequences (creating GitHub issues, sending external notifications), keep the threshold higher — 85 to 90%.

Step 5: Test It

Open your website. Click the widget. Type a test message like "how do I reset my password" and watch what happens. You should get an auto-response within a few seconds. Then type something weird like "my cat sat on my keyboard" and check that it shows up in your Slack channel as an unrouted message.

Run 5 to 10 test messages covering different scenarios. Adjust your rules if anything feels off.

What Happens Next

After a week of real traffic, check your dashboard. You'll see:

• How many messages came in
• What percentage was auto-resolved
• Which intents are most common
• Your average response time

Use this data to add more rules. If you see "feature request" messages coming in frequently, create a rule that tags them and sends them to a specific Slack channel or Linear project. If "order status" questions are common, create a rule that responds with your tracking page link.

Even with just 5 rules, you'll cut your support workload in half. Add more as you learn which questions keep coming in.

Common Questions

Does this work with single-page apps? Yes.

Can I customize the widget's appearance? Colors, position, welcome message, and response delay are all configurable in the dashboard. You can match it to your brand so it doesn't look like a generic third-party widget.

What if a customer sends multiple messages? The widget tracks conversations. Each message gets classified independently, but they're grouped into a single conversation thread.

Do I need a help center or knowledge base? No. Classification-based support doesn't reference a knowledge base. It classifies intent and routes accordingly. If you have docs, link to them in your auto-responses. But they're not required — the classifier works on its own training data, not your content.