Guide
How do approval gates keep AI drafts from reaching clients?
Every RealtyOS agent run says whether a person must approve it and why. Client-facing drafts, marketing content, and deadline changes wait in an approval queue, and compliance-sensitive output waits for an owner or broker admin. Nothing reaches a client until someone approves it, and each decision is recorded in the audit log.
What every run returns
Whether a task starts from a button on a record, from a question to the agent team, or from the system, the result has the same parts:
- A summary of what the agent found or drafted
- Recommended next steps
- The draft, when the task writes one
- A confidence score; below 0.6 the run asks the reviewer to check the facts
- Whether approval is required, and why
- The records involved
- An entry for the audit log
What waits, and for whom
The approval policy looks at the task and at what the agent actually wrote:
- Client-facing drafts, marketing content, and transaction deadline changes wait for a person.
- Compliance-sensitive output, and anything that reads like legal, tax, or financial advice, waits for an owner or broker admin.
- Fair Housing findings, texts without consent or inside quiet hours, and low confidence add reasons and raise the risk level.
Rules that only add gates
A brokerage's approval rules can add gates, such as auto-draft only. They can never remove the compliance gate.
Sending only after approval
Approved emails and texts are sent by RealtyOS, and nothing is sent before the approval. In the live demo, approved messages go to a demo outbox instead of real people.
A record of every decision
Agent runs, approvals, edits, sends, and exports land in an append-only audit log that owners, broker admins, and team leads can filter and export.
Questions
Can a draft be edited before it is approved?
What happens when the AI is unsure?
Can a brokerage turn the approval gate off?
Last reviewed by AgentivX Solutions