About Our Automated Reply System

Last updated: 14 April 2026

ContactReplyAI is an automated reply system used by Australian trade businesses to handle customer enquiries. This page explains how it works, what AI does and doesn't do, and how you can opt out.

How It Works

When a customer contacts a business that uses ContactReplyAI — by SMS, email, web chat, or WhatsApp — the business sends that customer an automated reply using ContactReplyAI's software. The process has two parts:

  • AI reads your message. An AI system (Anthropic's Claude) classifies what you're asking and extracts relevant details such as your phone number, address, or the type of job you're enquiring about.
  • An approved template replies. The reply you receive is drawn from a template that the business has reviewed and approved during onboarding, with details like their pricing, service area, and business hours filled in.

The AI does not write your reply — your reply comes from a template the business has signed off on. The AI's job is to read your message and pick the right template; the template does the talking.

Customer notice: If you are a customer who has received a reply from a business using ContactReplyAI, the business has sent you an automated reply, generated using ContactReplyAI's software. The business owner reviews these conversations and can respond personally at any time. If you'd prefer to speak with a person, reply asking for a callback or contact the business directly.

What AI Will Not Do

  • Will not generate novel replies. All outbound messages are drawn from templates the business has reviewed and approved. The AI does not write fresh prose for each reply.
  • Will not invent prices or services. Your details (prices, services, hours, trade) are filled into approved templates. The system won't invent prices or services the business doesn't offer.
  • Will not make binding commitments. Appointments, firm quotes, and enforceable promises are not made by the automated system. The business owner must authorise those directly.
  • Will not access external systems. The system cannot look up real-time availability, check live pricing, or verify anything beyond the business's approved templates and configured details.
  • Will not give professional advice. No legal, medical, financial, or detailed technical advice beyond general information the business owner would share with any customer.

Emergency and Safety Responses

When a customer's message contains keywords indicating a potential emergency (gas leaks, flooding, fire, electrical hazards, personal injury), ContactReplyAI uses pre-written, professionally reviewed safety templates — not AI classification. These templates are reviewed quarterly against current guidance from emergency services (000 Australia) and relevant industry safety bodies (WorkSafe AU).

Tier 1 — Immediate life-safety (gas, fire, electrocution)

Response: emergency number + evacuation instructions. Sent immediately, bypasses the normal reply pipeline.

Tier 2 — Infrastructure actions (burst pipe, trip breaker)

Response: standard homeowner-level guidance ("turn off mains water") with an "if unsure, don't proceed" caveat.

Tier 3 — NEVER given by the automated system

Gas infrastructure interaction, electrical fault diagnosis, structural assessment, any licensed-professional-only procedure. Response directs you to call a qualified professional directly.

AI Model Used

The AI component is Anthropic's Claude — specifically the Haiku model for inbound message classification and detail extraction. Message content is sent to Anthropic's API for this processing. Anthropic's commercial API terms (as at the date of this policy) prohibit the use of API inputs and outputs to train Anthropic's generally available models. Anthropic may retain and use limited metadata for abuse detection and trust & safety classifiers. We review Anthropic's terms annually and will update this policy if the position changes.

Outbound replies are rendered from the business's approved template library by deterministic code — not AI — with your details filled in.

CRAI Portal Chat Assistant (Marcus)

The ContactReplyAI customer portal at contactreply.app includes a built-in chat assistant named Marcus. Marcus is CRAI's own customer support and onboarding helper — he assists portal users (subscriber businesses) with setup, configuration, and product questions. He is entirely separate from the automated reply engine described above and does not send any messages on behalf of tradie businesses or any other third party.

Marcus is powered by a large language model (Anthropic's Claude Sonnet, accessed via OpenRouter). When the portal chat is active, the following context from the current portal page may be sent to the language model provider to generate a relevant help response:

  • The current page URL and page title
  • The onboarding step you are on (where applicable)
  • Values from visible form fields on the page — partially redacted before transmission: email addresses are reduced to first character + *** + domain (e.g. j***@gmail.com), phone numbers to the last three digits only (e.g. ***491), and ABN values to the placeholder <ABN>

Full email addresses, full phone numbers, and full ABNs are not transmitted to the language model provider. Conversation history with Marcus is limited to the last 12 turns of your session. OpenRouter and Anthropic's API terms prohibit the use of API inputs for training their publicly available models.

Accuracy and Safeguards

  • Template constraint: All outbound replies come from templates the business has reviewed and approved. If no approved template matches your message, a reply is held for the business owner's review rather than invented.
  • Confidence scoring: The AI classifier produces a confidence score for each message. Low-confidence classifications are flagged for the business owner to review before any reply is sent.
  • Human override: The business owner can take over any conversation at any time.
  • Pause button: The automated system can be paused entirely — all subsequent messages require manual response from the business owner.
  • Content filter: Outbound replies are screened for safety issues (defamation patterns, licensed-trade safety claims, identity misrepresentation) before sending.

Opt-Out

If you are a customer receiving automated replies and you wish to stop:

  • SMS: reply STOP, UNSUBSCRIBE, or CANCEL to any message. Stops further SMS from that number automatically and permanently (unless you later re-consent).
  • Email: reply with "unsubscribe" — the address is recorded as opted out.
  • Web chat / WhatsApp: send "please don't send automated replies" — this pauses the automated system for your conversation.

Opt-out records are retained indefinitely to meet Spam Act 2003 s18 obligations, even after the business's account is closed. See the Privacy Policy for detail.

Jurisdiction-Specific Compliance

  • Australia (primary): Compliant with the Spam Act 2003 (inbound-triggered only, no unsolicited commercial messages) and the Privacy Act 1988.
  • EU / UK: The service is offered exclusively to businesses registered in Australia. EU AI Act Article 50 transparency obligations are addressed by first-reply disclosure in outbound messages and by this page.

Contact

Questions about how the automated system works: