Facilitating the collaboration of multiple groups is a critical element in problem-based learning (PBL). In face-to-face learning environments, PBL facilitators require sufficient information about a group's progress and collaboration in real time to make decisions about when and how to facilitate. The capacity of facilitator is limited as PBL scales up. Online PBL settings can mitigate this challenge by presenting data from multiple groups to support facilitators using orchestration technology. A learning analytics dashboard with visual displays is one type of orchestration technology. This study examined 10 PBL facilitators' use of such an orchestration technology to assess the collaboration patterns of multiple groups. Think-aloud protocols were collected from PBL facilitators as they used the technology to understand group collaboration patterns, which were illustrated through learning analytics visualizations. The think-aloud method enabled the PBL facilitators to verbalize their thought processes, and content analysis was conducted to analyze the transcripts. This study found that the expert facilitators made consistent formative assessments and strategically selected the most relevant visualizations to examine knowledge co-construction and group collaboration. In addition, they used multiple visualizations to calibrate and confirm their understanding of the students' learning and group processes, rather than relying on one visualization. Understanding how facilitators use the information generated from a learning analytics dashboard to assess collaboration in an online PBL environment is critical for better support of online facilitation and the design of orchestration technology.