Salisbury documents the work required before enabling embedded administrative AI

The customer problem is a queue of embedded features without a funded route to evaluate them. Ask the CIO, HR, finance, procurement and staff representatives which feature addresses a documented service delay, who can authorize its use, and what evaluation resources exist now. A bounded engagement could inventory one administrative product and design a pilot decision record. The value hypothesis is better prioritization and less unexamined activation, subject to validation. Establish current status before approaching Salisbury or a peer institution. This historical report supports discovery questions; it supplies neither an available sales opportunity nor evidence of savings from any named product.

Begin with one disabled administrative feature and document the tenant configuration, input fields, permissions, outputs and downstream actions. Require a test environment, a business process baseline and approved synthetic or minimized records. Verify whether the feature only drafts a suggestion or can alter a system of record. Cloud, local and hybrid choices require separate data-path analysis; a license entitlement is insufficient evidence of suitability. Proposed proof of value should compare correctness and review time against the current process, test unauthorized access and demonstrate disablement. The source provides no implementation specification or security test results to reuse as assurance.

Have an administrative service owner sponsor the pilot, with IT responsible for configuration and HR or finance responsible for business correctness. Dependencies include allocated reviewer time, procurement evidence and a usable training plan. Train staff on exceptions and escalation before introducing the feature into routine work. Require a governance decision before activation and again after material supplier changes. Proposed acceptance is a completed access test, documented owner for each data flow, and a quality-and-time comparison covering representative exceptions. Agree thresholds before testing. Risks include hidden review work, staff uncertainty and allowing a convenient feature to outpace oversight.