Agent Bricks Gives Baylor a Direct Line to the Student Voice

Overview

Baylor University's Enrollment Management division built an agent workflow using Databricks Agent Bricks to review 100% of contact center calls about financial aid and student accounts, up from roughly 5% manual sampling. Call recordings are fed via API into a medallion architecture on Databricks; a Knowledge Assistant agent gives representatives instant policy guidance, and a Multi-Agent Supervisor evaluates calls against standard operating procedures, citing timestamps and sources. Supervisors also query call data via Databricks Genie in natural language. Unity Catalog and row-level security via Active Directory groups govern access to FERPA-protected student data. The system generates a daily summary report in about two minutes and enables faster, more consistent coaching for contact center staff.

The challenge

Baylor's Enrollment Management contact center fields hundreds of calls daily about financial aid and student accounts, but manual review could not scale with the volume. Even a dedicated QA hire would cover only about 5% of calls, leaving feedback inconsistent and too slow to be useful, and the data was inaccessible because it was not captured in a structured format.

The solution

Kyle and his team built an agent workflow using Agent Bricks, connecting Baylor's phone system via API to feed call recordings and metadata into a medallion architecture on Databricks. A Knowledge Assistant agent gives representatives instant guidance from policies and documentation, while a Multi-Agent Supervisor evaluates each call against standard operating procedures, linking findings back to the original interaction with timestamps and source citations. Supervisors can also query call data through Databricks Genie using natural language. Unity Catalog and row-level security managed through Active Directory groups govern access to FERPA-protected student data.

Reported business value

Agent Bricks shifted call review from about 5% manual sampling to 100% coverage of calls. Supervisors generate a daily report summarizing volume, sentiment shifts and examples of interactions, which took about two minutes to write and run the first time it was generated, enabling full QA coverage without adding headcount.