A receptionist serving several property-management clients must keep each client's resident data and actions separate. Correct wording cannot compensate for a retrieval query or credential that points to the wrong account.
Carry an explicit client identifier through the call session, property lookup, record search, attachment access, ticket creation, and staff notification. Establish that context from a trusted route or approved mapping rather than a caller-supplied account name alone.
Enforce boundaries below the prompt
Scope database and API access to the intended client. Keep credentials and destination mappings isolated and validate the client-property relationship before every consequential action. A prompt instruction to “use the right client” is not a substitute for access controls.
Logs, test fixtures, exports, and notification channels also need the boundary. A correctly isolated booking path can still leak information through a shared debugging view or an alert sent to the wrong manager.
Work through a fictional shared building name
Two management clients each have a property called Riverside House. A caller reaches Client A's line and requests maintenance at Unit 5. The system must search only Client A's authorized property context.
If the details do not match, it should use an unresolved-property route rather than falling back to Client B's records. A seemingly helpful cross-client search can expose resident information and create work in the wrong portfolio.
| Isolation test | Required result |
|---|---|
| Same building name across clients | Only intended client candidates |
| Foreign record identifier supplied | Access denied or safe review |
| Wrong credential mapping | Action blocked visibly |
| Shared staff dashboard | Role-scoped records |
| Failure alert | Correct client recipient only |
Test denial paths deliberately
Successful same-client calls do not prove isolation. Include negative tests that attempt to read or act on another client's identifiers. Verify that failures do not return private record details in error messages.
Shared infrastructure reduces operating cost and repeated maintenance, but it requires disciplined access boundaries and review. Separate deployments simplify some isolation concerns while adding configuration and support work. Choose the model you can verify and maintain.
Review agency client isolation for the general technical pattern and include attachments and exports in the acceptance scope.
Questions about resident-data separation
Can one shared prompt handle several clients?
It can hold common guidance, but client facts, credentials, retrieval, and action destinations still need explicit enforced boundaries.
Should failed matching search another client's records?
No automatic fallback should cross the authorized client boundary. Use a review process within the intended scope.
What proves isolation is working?
Documented negative tests, scoped permissions, correct destination evidence, and review of logs, notifications, attachments, and support access.