Product descriptions:
Baylor University Enrollment Management helps students make informed decisions about affording and continuing their education. Serving 19,000 students and evaluating 50,000 applications annually, the division fields hundreds of calls daily through its contact center. These conversations about financial aid and student accounts are high-stakes, often influencing enrollment and retention. Manual review couldn’t scale with the volume, leaving the student voice trapped in recordings. With Agent Bricks operating on the Databricks Platform, Baylor now reviews 100% of calls, delivers faster coaching to staff, and surfaces structured voice-of-student data securely across campus.
Manual reviews left the student voice trapped in call recordings
When students or parents call Baylor University with questions about financial aid, deadlines, or bills, conversations can often become emotional. “It’s a direct line to the student voice,” said Kyle Van Pelt, Director of Process and Governance in Enrollment Management at Baylor University. Kyle views the contact center as the institution’s closest real-time connection to what students are experiencing.
A small team of roughly ten representatives handles 100-300 calls per day. Monitoring these calls for quality and coaching opportunities is essential, but having a dedicated quality assurance person was not a viable option. Reviews happened only after escalations, which could be days after the original conversation. Even a dedicated QA hire would cover only about 5% of calls, leaving feedback inconsistent and too slow to be useful. Positive interactions often went unrecognized, while problematic patterns could spread unchecked. This was especially true with a partially remote team, where oversight was more complex.
Calls also contained high-signal data about student concerns, objections, and decision factors. When a parent expresses hesitation about cost, when a student asks about a competing school's offer, when applicants raise the same concern week after week, those patterns could inform recruiting strategy, marketing messaging, and institutional research. Instead, Baylor’s richest voice-of-student insights were trapped in recordings that few people could analyze. “The data is inaccessible because it’s not captured in a structured format,” Kyle said. “And I cannot take Family Educational Rights and Privacy Act (FERPA) information and throw it out into the ether. I need to do it in a secure environment.”
Agent Bricks converts recordings into structured data for coaching
With those constraints in mind, Kyle and his team built an agent workflow using Agent Bricks, a Databricks toolkit for building AI agents within a secure, governed environment. The team connected Baylor's phone system via API, feeding call recordings and metadata directly into a medallion architecture on Databricks. There, the data is organized and enriched with initial processing so Agent Bricks can present sentiment trends, topics, question-and-answer pairs, and coaching signals to end users.
Two agents and a Genie space work together to surface insights. A Knowledge Assistant gives representatives instant guidance from policies, procedures, and documentation, reducing routine interruptions and supporting new staff training. A Multi-Agent Supervisor evaluates each call against standard operating procedures, identifying where guidance was accurate and where it diverged. The system links findings back to the original interaction with timestamps and source citations. “All your homework is done,” Kyle said. “The transcript is surfaced with timestamps, and it cites the actual sources.”
Supervisors can explore the data themselves through Databricks Genie, querying calls using natural language and filtering by sentiment shifts or compliance flags. Unity Catalog makes this kind of open access safe, providing the governance required to handle sensitive student data. Kyle described it as “using a calculator in a locked room” where Baylor brings the model into a secure environment, and controls all the data the model can see. Row-level security managed through Active Directory groups ensures different departments view only the calls relevant to their work. “Databricks makes it so you just need to know how to write a good prompt," said Logan Guardiola, Data Engineer.
From 5% sampling to 100% visibility in minutes
Agent Bricks shifted review from occasional escalation responses to an operational process covering 100% of calls. Insights from every conversation are securely available across the organization, with role-based access and governance that share process and product signals, not individual rep performance. Supervisors generate a daily report summarizing volume, sentiment shifts, and examples of both strong and problematic interactions. The first time Kyle generated a report, it took about two minutes to write and run the prompt.
The impact is most visible in coaching. Instead of feedback arriving days later, representatives receive timely guidance tied to specific moments in their calls. Supervisors can also surface and reinforce strong performance. “I can ask for the five most positive calls and tell those agents they did a great job,” Kyle said. “Improving quality of life for contact center staff has been one of the best good news stories of this project.” Baylor gained visibility into every call, something that wouldn't have been feasible with manual review at any staffing level.
Structured call data also gives Baylor visibility into its own processes. By aggregating the actions representatives take across calls, supervisors can spot recurring pain points and targets for automation that would be invisible at the individual call level. Genie extends this further by putting call data directly into the hands of end users, who can query it on demand without developer support to assess the impact of new communications, special offers, or policy changes as they happen.
From here, Baylor plans to apply the same analysis to email communications and connect agents to student account data, enabling agents to evaluate whether guidance was correct for each student's specific situation, not just whether the representative followed policy. “They want to talk to a person,” Kyle said. “Great AI can help that person be super effective and level up that experience for the representative and the student.”