1. Agent Configuration
- System Prompt: Have you stress-tested the prompt? Does it handle edge cases (e.g., user asks about competitors, user is angry)?
- Greeting: Is
speak_firstconfigured correctly?- Inbound: Usually
true(Agent greets caller). - Outbound: Usually
true(Agent states reason for call).
- Inbound: Usually
- Voicemail: Is
voicemail_detectionenabled if you are making outbound calls? Do you have a specificvoicemail_messageset? - Language: If supporting non-English, have you selected a compatible Voice ID and LLM model?
2. Telephony & Network
- Number Assignment: Is the correct phone number assigned to the agent?
- Fallback: If using Twilio BYOC, have you configured a fallback URL in Twilio in case Butter AI is temporarily unreachable?
- Concurrency: Check your plan limits.
- Starter: 5 concurrent calls.
- Pro: 25 concurrent calls.
- Action: Upgrade if you expect call bursts.
3. Knowledge Base & Tools
- Context Window: Are your knowledge base documents concise? Large/irrelevant docs increase latency.
- Tool Timeouts: Do your custom tools respond in under 3 seconds? If not, the agent might time out or awkwardly fill the silence.
- Error Handling: What happens if your tool API returns a 500 error? Does your system prompt instruct the agent on how to handle failures (e.g., “I’m having trouble accessing that right now”)?
4. Operational
- Webhooks: Have you set up a webhook listener to track call outcomes (
completed,failed)? - Billing: Do you have auto-recharge enabled for your credits to prevent service interruption mid-campaign?
- Monitoring: Do you have a process to review call recordings/transcripts daily to spot quality issues?
5. Compliance
- Recording: If
recording_enabledis true, are you legally required to play a “This call is being recorded” message?- Tip: You can add this as a pre-connect announcement in Twilio/Telnyx or instruct the agent to say it immediately.
- Identification: Does your agent identify itself as an AI? (Required in many jurisdictions).