Transcription: Transcripts are created using a combination of speech recognition tools and human transcribers, which may result in errors. Please refer to the corresponding audio for the accurate record.
Penny Crosman (00:04): Greetings and welcome to the American Banker Podcast. I am Penny Crosman. Many financial institutions are currently experimenting with various types of AI, such as machine learning, deep learning, and generative AI. However, the journey from conceptualizing an AI use case to successfully implementing it and achieving tangible outcomes can be quite challenging. Eric Siegel, a renowned expert in predictive analytics and AI, who previously served as a university professor, has recently authored a book titled “The AI Playbook.” Today, Eric will enlighten us on how to effectively translate advanced AI concepts into practical results. Welcome, Eric.
Eric Siegel (00:43): Thank you, Penelope. It’s a pleasure to be here.
Penny Crosman (00:45): Thank you. To begin with, what motivated you to produce a music video centered on predictive analytics?
Eric Siegel (01:02): My primary goal is to educate and raise awareness about this technology. There are two prevalent approaches to explaining the deployment of predictive models generated by machine learning. One focuses on high-level buzzwords and potentially exaggerated claims, which can be rather abstract. Conversely, I believe in offering concrete examples because the process of learning from data to make predictions and leveraging these predictions to enhance various large-scale operations, such as targeted marketing, fraud detection, credit risk management, insurance, and pricing strategies, is both captivating and essential. My book, “The AI Playbook,” emphasizes the necessity of bridging the gap between business and technology. This involves empowering business professionals to acquire a basic technical understanding, enabling them to engage meaningfully with AI initiatives. The key message is that effective collaboration is crucial for success. Currently, a significant number of new machine learning projects in enterprises fail to progress to the deployment stage due to insufficient planning and collaboration on the business side.
Penny Crosman (02:29): You mentioned that machine learning is considered mandatory. Could you elaborate on why you believe this to be the case?
Eric Siegel (02:49): Clarifying my statement, I believe it is imperative to comprehend machine learning, as its application is crucial in enhancing business operations. In a competitive landscape where standard business processes are becoming commoditized, and products exhibit similar features, the utilization of science to drive business improvements becomes essential. Business operations encompass numerous decisions, and the ability to predict outcomes effectively is paramount for informed decision-making. While neither humans nor machines possess clairvoyance, leveraging data for predictive purposes enables superior forecasting compared to mere guesswork. Machine learning facilitates more accurate targeting in marketing, improved credit risk assessment, and enhanced fraud detection capabilities.
Penny Crosman (03:55): In your view, why do most machine learning projects fail to progress to the deployment phase? Do you consider this a significant obstacle that needs to be overcome?
Eric Siegel (04:16): The primary reason for project failures lies in inadequate planning and stakeholder engagement. Despite identifying clear opportunities to leverage predictive analytics for enhancing operations, many initiatives falter due to stakeholders’ hesitance or insufficient technical preparations. The focus tends to be on the technical aspects rather than on the critical phase of integrating these predictive models into operational workflows. Bridging this gap necessitates rigorous planning and active involvement of business stakeholders. Without their active participation and commitment to the deployment process, valuable insights generated by machine learning models remain untapped.
Penny Crosman (05:51): Could you elaborate on the factors contributing to the reluctance or challenges faced by businesses in deploying machine learning models successfully?
Eric Siegel (06:04): The reluctance to deploy machine learning models stems from a combination of factors, including fear, bureaucratic hurdles, and a lack of comprehensive understanding. Implementing change management practices is essential to navigate these challenges effectively. While the technical aspects of machine learning are fascinating and innovative, the ultimate goal is to drive operational improvements. This requires a shift in focus towards integrating predictive models into business operations seamlessly. Stakeholders must actively engage in the deployment process to ensure that the potential value offered by these models is realized.
Penny Crosman (08:01): In the realm of financial services, the use of machine learning in critical areas like lending decisions, fraud detection, and marketing raises concerns about transparency, fairness, and bias. How do you view these challenges in relation to machine learning models?
Eric Siegel (09:03): Addressing responsible AI practices, ethical considerations, and potential biases in machine learning models is paramount. Discriminatory models and machine bias pose significant risks, especially in decision-making processes that impact individuals’ access to resources or services. Achieving transparency in model decisions is essential to identify and mitigate biases effectively. While there are valid concerns regarding model interpretability and fairness, there is a need to strike a balance between understanding model predictions and ensuring ethical practices in deploying AI solutions.
Penny Crosman (12:40): With the recent surge in interest surrounding generative AI and large language models, what do you consider to be the most practical and beneficial applications of these technologies?
Eric Siegel (12:40): Large language models excel in generating initial drafts for various content types, including text, code, and images. While there is considerable hype surrounding the capabilities of generative AI, it is essential to recognize its limitations. These models are proficient at mimicking human-like responses and producing coherent text across diverse topics. However, their primary function is to predict based on the training data and may not always provide accurate or definitive answers. Understanding the specific use cases and limitations of large language models is crucial for maximizing their utility in various applications.
Penny Crosman (17:42): For financial institutions with limited technical resources, relying on AI vendors for prepackaged solutions is common. What advice would you offer on selecting the right AI vendors, evaluating their offerings, and collaborating effectively with them, especially for smaller organizations?
Eric Siegel (17:42): When engaging with AI vendors, it is essential to prioritize consulting services over off-the-shelf solutions. Machine learning projects require a collaborative approach between the vendor, the business stakeholders, and data scientists to ensure successful deployment and operational integration. While smaller organizations may lack in-house technical expertise, active participation in the project planning and deployment phases is critical for achieving desired outcomes. By understanding the core principles of machine learning and fostering strong collaboration with vendors, even small institutions can leverage AI effectively to enhance their operations.
Penny Crosman (25:08): Thank you, Eric Siegel, for sharing your valuable insights today on the American Banker Podcast. And to our listeners, we appreciate your continued support. Remember to rate, review, and subscribe to our podcast on www.americanbanker.com/subscribe. This is Penny Crosman signing off. Thank you for tuning in.
