How PowerFrame uses AI
PowerFrame deploys AI agents ("AI Employees") that handle real-time customer interactions for service businesses. This page explains in plain terms what that means, what data AI sees, and what you should know before enabling it.
TL;DR: AI handles your inbound calls, iMessage replies, lead qualification, and review requests. It talks to your customers in conversation. It is accurate about 92–97% of the time. It escalates to humans when uncertain. If a caller directly asks if they are talking to AI, the agent discloses.
AI providers we use
| Function | Provider | What they see |
|---|---|---|
| Conversational reasoning | Anthropic (Claude) | Message content, your configured scripts and pricing |
| Voice AI (outbound calls) | ElevenLabs (voice synthesis), Retell (orchestration) | Voice output, conversation context |
| Voice AI (inbound calls) | Retell + Anthropic | Real-time transcription, reasoning |
| SMS / iMessage | Twilio (SMS), Sendblue / Beeper (iMessage) | Message content only |
| Postmark, SendGrid | Email content only |
None of these providers use your data to train their public AI models. All have contractual commitments prohibiting training use. Where available, we enable zero-retention modes.
Disclosure practices
Proactive disclosure: On first contact with a new caller, the agent identifies itself by name ("Hi, this is Sarah from ACME HVAC"). It does not volunteer that it is AI unless asked.
Reactive disclosure: If a caller directly asks "Are you a real person?" or "Am I talking to a bot?", the agent discloses: "I'm an AI assistant for ACME HVAC. I can help you book an appointment or transfer you to a team member if you prefer."
State-specific disclosure: In California, Colorado, and other states requiring AI disclosure, the agent follows local rules. Our configuration includes state detection from phone number and address.
Accuracy and limitations
Confidence thresholds: Every agent has a 70% confidence threshold. Below that, it escalates to a human on your team (via dashboard notification, SMS, or call transfer — configurable).
Known limitations: AI agents may:
- Misunderstand heavy accents or poor audio quality in voice calls
- Miss sarcasm, implied context, or highly specialized industry jargon not in their training
- Occasionally generate incorrect pricing if not strictly configured
- Struggle with customers who shift topics rapidly or interrupt
We tune continuously based on flagged conversations. Your Client Operations Manager reviews 20–50 flagged interactions weekly during month 1, then monthly thereafter.
What AI does NOT do
- Does not provide legal, medical, financial, or tax advice
- Does not make binding commitments beyond your configured pricing
- Does not process payments independently — it books appointments and logs quote requests, with billing handled by your existing payment stack
- Does not make hiring, termination, or disciplinary decisions
- Does not contact individuals on DNC lists
- Does not send SMS or voice calls outside of TCPA-compliant hours (8 AM–9 PM local time)
Human oversight
Every PowerFrame account has a Client Operations Manager assigned who:
- Reviews flagged conversations weekly
- Adjusts agent logic, scripts, and escalation rules
- Provides a monthly performance report with conversation samples
- Is reachable via Signal for urgent issues within business hours
Data minimization
AI agents only see data needed for the specific task:
- The Receptionist sees: your service areas, hours, emergency protocol, appointment booking rules
- The Qualifier sees: your qualification criteria, vertical-specific triage
- The Dispatcher sees: technician availability, skill matrix, service radius
Agents do not share data across accounts. Your configuration is isolated.
Bias and fairness
We design agents to treat all callers equivalently regardless of accent, gender, or perceived demographics. We audit for bias quarterly. If you notice biased behavior, report to safety@powerframe.ai.
Your rights regarding AI
If you are a customer of a business using PowerFrame, you have the right to:
- Know if you are interacting with AI (just ask)
- Request escalation to a human at any time
- Opt out of AI-only communication channels
- Access and correct information the AI has recorded about you (contact the business directly)
Questions
Questions about our AI practices:
Email: safety@powerframe.ai
Phone: (888) AI-POWER