Set up AI replies

What AI adds to your pipeline

Without AI (Core plan), Signals routes messages based on channel, source, and business hours. Your rules fire, templates send, and tasks get created, all without AI.

With AI (Pro and Enterprise plans), Signals also:

• Classifies intent: booking request, complaint, general inquiry, etc.
• Detects urgency: low, medium, high, critical
• Reads sentiment: positive, neutral, negative
• Scores leads: ranks quality 0–100 with hot/warm/cool/cold tiers
• Generates replies: writes contextual responses using your brand voice and templates as guidance
• Runs Signal Guard guardrails: checks every AI reply for compliance before sending

Requirements

• Pro or Enterprise plan: AI features are not available on Core
• AI provider configured: either platform-managed (included in your plan) or Bring Your Own Key (BYOK)

Step 1: Choose your AI provider

Go to Voice & Channels → Connections → AI Provider in your dashboard.

Platform-managed (recommended):

• Warpflow provides the AI, no API key needed
• Cost is included in your plan subscription
• Usage counts against your AI Signals limit
• This is the default. If you just want AI to work, leave this setting alone.

Bring Your Own Key (BYOK):

• Enter your own API key
• You pay the AI provider directly for usage
• Usage still counts against your AI Signals limit for metering
• Use the Test button to validate your key before saving

Step 2: Configure brand voice

AI replies follow your brand voice settings. If you haven't already:

1. Go to AI & Automation → Brand Voice
2. On the Voice Identity tab, select a persona (or keep the one from your industry preset)
3. Review tone selection and vocabulary settings across both tabs

See Brand Voice for details.

Step 3: Review Signal Guard

Signal Guard is the guardrails system that prevents AI replies from making inappropriate claims or promises. Your industry preset loaded default guardrail rules, but review them:

1. Go to AI & Automation → Signal Guard
2. Check that the AI Judge is enabled (recommended for healthcare, legal, and financial businesses)
3. Review prohibited phrases and escalation keywords
4. Add any custom compliance rules specific to your business

See Signal Guard for details.

Step 4: Enable AI in your routing rules

AI classification opens up new routing conditions:

You can add these conditions to existing rules or create new ones.

To have AI generate a reply, add a generate_reply action to a rule using the visual rules builder. The AI will write a contextual response using your brand voice and the message context.

Step 5: Configure AI routing

AI routing controls what happens after a reply is generated: should it auto-send or be held for human review?

1. Go to AI & Automation → Routing Rules and click the AI Routing tab
2. Configure auto-send and confidence threshold settings
3. Enable fallback replies if you want AI to respond even when no rule matches

See Set up AI routing for the full guide.

Step 6: Test AI features

1. Go to the Test Runner
2. Paste a realistic customer message
3. Run the test
4. Check the results for:
   • Classification: did it detect the right intent and urgency?
   • Lead score: does the tier feel right?
   • Generated reply: does it match your brand voice? Is it appropriate?
   • Signal Guard evaluation: was the reply approved, flagged, or blocked?
5. Check the Work Queue for held drafts. Drafts show why they were held (confidence too low, PII detected, or auto-send disabled). This helps you tune your AI routing settings.

What happens when AI is off

If AI features are disabled or your AI signal limit is reached:

• The pipeline still works. Rules based on channel, source, and business hours continue to fire.
• Templates still send. Template-based actions don't need AI.
• Only AI stages are skipped. Classification, scoring, and reply generation stop. Rules that depend on intent, sentiment, or score conditions won't match.

The pipeline never drops signals. It degrades gracefully to rules-only mode.

Tips

• Start with platform-managed AI. It's the simplest setup: zero configuration.
• Monitor your AI Signals meter on the Plan & Billing page. If you're consistently near the limit, consider upgrading.
• The AI judge is worth keeping enabled. It catches things that simple pattern matching misses, especially in healthcare and legal contexts.
• Test, test, test. Run different types of messages through the test runner to see how AI handles them before going live.