Towson announces talent-platform access with an AI contract assistant

The customer problem is finding relevant staff candidates and assembling defensible contract comparisons. Engage athletics HR, the athletic director, university procurement and counsel. Ask which searches or comparisons are slow, how data are sourced today and how stale information is corrected. A bounded engagement could test retrieval against a small authorized contract set with known answers. The value hypothesis is faster verified research, subject to measuring correction effort. Do not promise more diverse hiring, better retention or cheaper compensation from this announcement. Separate interest in a contract assistant from permission to automate selection or contact candidates.

Establish whether the assistant retrieves clauses, summarizes documents or generates unsupported comparisons; the announcement does not resolve this. Require document-level permissions, source links, version dates and logging before connecting any campus system. Test omissions, conflicting amendments, comparable-role definitions and unsupported answers against human-reviewed cases. Include prompt-injection tests using untrusted document text and prevent the tool from taking hiring actions. Evaluate hosted, hybrid or local feasibility only after obtaining supplier architecture and data terms. A proof of value should compare verified-answer accuracy and total review time against the current workflow, with explicit abstention for missing evidence.

Athletics HR should own a limited rollout with counsel validating contract questions and IT administering access. Inventory approved users and data purposes, prepare a baseline question set and train staff to check citations. Hold governance checkpoints before adding confidential documents, expanding candidate searches or exporting records. Proposed acceptance criteria include source support for every accepted answer, successful access-revocation tests, documented correction requests and reduced total research time without lower accuracy. Monitor adoption through verified task completion rather than login counts. Stale compensation data, misleading comparisons and excessive candidate-data access may outweigh convenience; preserve manual research and review routes.