Technical Product Managers in the Brave New World of AI and Machine Learning

by Sri Phani Teja Perumalla

The rapid growth of artificial intelligence (AI) and machine learning (ML) across various sectors has radically disrupted the IT industry. This shift has created unique challenges and opportunities for technical product managers (TPMs) who handle the technical, ethical, and regulatory complexities of IT work. By adapting quickly, TPMs can capitalize on new trends and innovations to increase their value to their companies. This recent evolution has established TPMs as vital players in successful AI and ML product launches. The responsibilities of a TPM are typically more focused and extend beyond those of a conventional product manager. They often include managing data cycles, selecting models, and focusing on technical design and architecture. AI and ML are integral to a TPM’s work since 47 percent of TPMs already use these technologies.

Cross-functional collaboration

            Diverse teams of engineers, data scientists, and legal personnel support AI/ML products. It’s imperative for TPMs to oversee the cross-functional collaboration between these groups and serve as intermediaries by aligning disparate teams toward common AI/ML product goals. TPMs work closely with data scientists to understand model limitations, performance metrics, and desired outcomes. They also collaborate with engineers to ensure model deployment and system integration. On the legal side, AI/ML products are highly regulated and subject to compliance with safety and ethics regulations and guidelines. Especially in regulated industries, these products must comply with laws regarding data privacy, bias, and transparency. TPMs work with legal teams to address these requirements early in development.

‘Garbage in, garbage out’

            TPMs ensure data used in AI/ML products is high-quality, pre-processed, and maintained throughout a product’s lifecycle. This includes overseeing data ingestion, cleaning, transformation, and storage, ensuring it aligns with the model’s performance requirements. By applying the ‘garbage in, garbage out’ principle, TPMs recognize that even the most advanced models will produce flawed outputs if the data is unreliable. Because of this, they implement stringent standards to prevent low-quality data from undermining the model’s accuracy and overall effectiveness. TPMs often make a consequential choice when deciding between deploying a supervised or unsupervised ML model, which can introduce errors, biases, and other consequences. This decision must consider the business problem, the ML model’s performance metrics, and computational efficiency, which can enable scalability and accuracy.

Compliance challenges

            In industries like healthcare, finance, and insurance, regulations require that AI models and their decision-making processes are transparent and understandable. Compliance standards, such as the General Data Protection Regulation (GDPR) in the European Union (EU), mandate that AI-automated decisions are explainable to affected individuals. TPMs explain these important AI/ML aspects to the requisite stakeholders. Tools like Local Interpretable Model-agnostic Explanations (LIME) or Shapley Additive Explanations (SHAP) help TPMs create simplified explanations for complex models that stakeholders can use to clearly understand how the AI tools function. TPMs can also host question-and-answer and educational sessions to clarify how AI tools align with regulatory standards. Regular updates are crucial to informing stakeholders of model changes or performance adjustments. When stakeholders—from clients and regulators to non-technical executives—understand how these models work, trust in AI’s fairness and accountability is established. This is especially critical in industries where AI’s use directly impacts people’s lives and financial or health security.

AI product management ethics

            Another critical responsibility for TPMs is upholding fairness and mitigating bias in AI models throughout the product development lifecycle. They ensure the datasets used represent diverse groups and help identify and control any sources of biases. Data scientists support these efforts by employing techniques like resampling or various bias-detecting tools to improve the fairness of models. This process is essential in fields like finance or hiring, where AI models can unintentionally reinforce social inequalities if trained on historical data that may contain biases from past human decisions or societal inequalities. As such, it’s vital for TPMs to work with data scientists to prevent these AI models from perpetuating discrimination. One tool that can help is Google’s PAIR (People and AI Research), which allows developers to explore different concepts of fairness in ML systems. TPMs can use Google’s What-If Tool, which provides an interactive way to explore these fairness options. Developers can see the trade-offs of each fairness approach and make more informed, ethically guided decisions. This responsibility for sound ethics continues beyond the initial launch of an AI product since ongoing monitoring and adjustment of real-world applications is required to correct emerging issues.

            Ensuring fairness in AI healthcare diagnostic tools is another critical responsibility for TPMs in the healthcare industry. In this field, the implications of biased AI can be severe because biased models may lead to misdiagnoses or inequitable treatment across demographic groups. For these reasons, it’s critical for TPMs to ensure that diagnostic tools perform consistently and accurately across different populations. For example, since biases in training data could lead to disparities in healthcare outcomes, a diagnostic AI tool used in radiology must provide accurate results for all skin tones or demographic backgrounds, Fairness tools and frameworks like IBM’s AI Fairness 360 measure and improve impartiality in diagnostics, ensuring compliance with ethical and regulatory standards.

Emerging skills and tools for TPMs

            Key skills for TPMs include database management and knowledge of supervised and unsupervised learning. Database management skills are essential for guiding structured and unstructured data across platforms. Successful TPMs oversee data handling, including data storage and retrieval, and ensure data integrity within systems, which supports efficient AI model training and deployment. It’s also important for TPMs to be familiar with supervised learning models, where data is labeled to train the model. This includes tasks such as classification and regression, standard in applications like predictive analytics and customer segmentation. Additionally, deploying unsupervised learning techniques, such as clustering and association, helps TPMs oversee AI projects where patterns or groupings are extracted from unlabeled data. This knowledge allows TPMs to evaluate when each type of learning is appropriate based on the business problem.

            AI and ML have widely disrupted the IT economy. Now, TPMs face new responsibilities and opportunities that, if properly leveraged, can elevate their careers and companies above the competition. When TPMs adapt and learn how to translate new policies for company stakeholders, establish new data-driven key performance indicators (KPIs), and ensure optimal data quality, they can increase the success rate of new product launches. Companies with an eye on the future of the IT industry would be wise to ensure their TPMs are adequately trained to take advantage of these impressive new technologies.

About the Author:

Sri Phani Teja Perumalla is a product director with a multinational banking and financial services holding organization. He has a proven track record of launching high-profile consumer-facing products and platforms. He has a Bachelor of Technology degree and a master’s degree in international business. Connect with Sri on LinkedIn.