Where AI helps—and where a person remains responsible.

1. Intended uses

• Answering public product-support questions from approved information.
• Classifying the topic and urgency of an enquiry.
• Drafting or delivering responses from approved business information.
• Asking configured qualification questions.
• Summarising conversations and identifying the next action.
• Offering booking options and routing exceptions to a person.

2. Disclosure

Where a person interacts directly with an AI voice or chat system, the production workflow should clearly identify that the interaction is automated unless it is already obvious. AI-generated or materially manipulated content should be labelled where applicable law requires it.

3. Human control

Customers must define escalation conditions. Complaints, uncertainty, vulnerable individuals, sensitive information, emergencies and high-value exceptions should be routed to a qualified person rather than handled as unrestricted automation.

4. Limitations

The public Support assistant is deliberately bounded: it answers from a verified help set and refuses to guess when the information is missing. Broader AI can misunderstand intent, produce incorrect information or act on incomplete business knowledge. BuiltBehind should use constrained instructions, approved knowledge, testing, monitoring and audit records, but no AI output should be presented as infallible.

5. Significant decisions

BuiltBehind must not be positioned to make solely automated decisions producing legal or similarly significant effects without a separate lawful, risk-assessed and explicitly approved design.

6. Feedback and challenge

Use the Contact & support route to challenge a support answer or report inappropriate behaviour. Every activated customer-facing AI channel should provide a practical way to reach an accountable person, correct information, contest an outcome and report inappropriate behaviour. That process must be tested before customer use.