Enterprises today are inundated with massive volumes of unstructured data, ranging from call recordings and video footage to customer feedback and supply chain signals. This treasure trove of business intelligence is often left untapped due to the complexities involved in analyzing it. Historically, unstructured data has been challenging to navigate due to its inconsistent formats and the difficulty in deriving actionable insights. However, when properly managed and centralized, this chaotic data can serve as a crucial asset for training advanced AI models, enhancing their accuracy, adaptability, and contextual understanding. A notable case study is the Charlotte Hornets, an NBA team that successfully harnessed previously neglected video footage to identify a promising new draft pick.
The Charlotte Hornets faced the significant challenge of transforming raw, unstructured game footage into useful insights. To achieve this, they employed advanced AI techniques, including computer vision, which allowed them to analyze gameplay from less-visible leagues that typically go unnoticed by traditional scouts. Jordan Cealey, Senior Vice President at Invisible Technologies, emphasized the growing importance of computer vision in the AI landscape, stating that it enables the analysis of data sources previously considered too cumbersome. By utilizing techniques such as player tracking and movement analysis, the team extracted vital kinematic data, which informed their recruitment process. The insights gained through these methods not only identified a player whose skills perfectly complemented the team’s needs but ultimately led to that player being honored as the Most Valuable Player at the NBA Summer League.
The Hornets’ experience highlights several crucial lessons regarding the effective use of unstructured data in AI applications. First, it is imperative to prepare unstructured data for AI models through well-designed data pipelines and management practices. As Cealey points out, effective utilization of unstructured data depends on having structured data that is ready for AI consumption. Organizations may need to seek partnerships with technical experts to tailor AI models to their specific contexts. The traditional model of lengthy consultancy engagements may be insufficient in the fast-paced world of AI. Instead, the emerging Forward-deployed Engineers (FDEs) model, which connects engineering capabilities directly with the customer’s operational environment, offers a more responsive approach. By working closely with clients, FDEs ensure that AI solutions are built with a deep understanding of the unique challenges and requirements of each organization. For the Hornets, collaborating with FDEs allowed them to refine their models and create a high-quality dataset, essential for validating and enhancing AI performance. In summary, the successful integration of unstructured data into AI initiatives hinges on careful preparation, contextual understanding, and the ability to adapt models to specific use cases, ultimately leading to more effective data-driven decision-making.
Source: Using unstructured data to fuel enterprise AI success via MIT Technology Review