1. AI is an optional layer
Lead intake, human assignment, manual qualification, notes, tasks, appointments, pipeline changes and sale closure must remain usable without AI. AI is off by default.
Switching AI off must not suspend a customer relationship or block a sales stage. Open work needs an accountable human owner or a visible manual assignment queue.
2. What the assistant does
Leadvera’s AI features are conversation summaries, response drafts, qualification suggestions and next-step suggestions. Each runs only after a workspace administrator switches AI on.
AI suggestions are produced only when AI is enabled and a provider is configured. They are not evidence of model accuracy, which is why a person reviews them.
3. Review before consequential action
A suggested response is not a verified fact. Review names, dates, prices, property details and contractual commitments before communicating with a customer.
Sensitive, disputed or consequential matters should be handed to an authorised person. AI must not independently make eligibility decisions about housing, credit, employment or similarly consequential services.
4. Minimise and approve data sharing
Only the information needed for the requested task is sent to the AI provider (Anthropic, listed in the subprocessor register).
Do not place passwords, payment credentials, identity documents or unrelated sensitive records into prompts or notes that AI may read.
5. What the assistant can and cannot do
The assistant reads the lead or conversation it is asked about and produces a draft, summary or suggestion. Messages it drafts are sent only after a person approves them, unless an administrator has explicitly turned off approval for an AI agent.
The assistant never closes a sale as won or lost, deletes or merges records, or changes its own permissions. Every AI action is recorded with the person or agent it acted for.
6. Keep work moving through outages
If AI is switched off or unavailable, the work goes to a human owner or the manual queue. Assignment rules keep working without the AI provider.
AI calls have timeouts and limits. A failure creates an actionable state and returns the work to a person, rather than repeatedly blocking a lead.
7. Explain and monitor use
AI drafts and suggestions are shown for review before anyone uses them, so teams know what to check and correct. Where communication rules require disclosure, include it in the message.
Review AI output against real cases and correct it when it is wrong; AI runs are recorded so they can be reviewed.
8. Manage changes deliberately
We tell customers before changing the AI provider, and workspace administrators can switch AI off at any time.
The agreement and privacy materials reflect the enabled use. Core CRM access remains available if the customer switches AI off.