Challenges in Artificial Intelligence Product Management

Original

Today, internet giants are diving into the AI revolution, while traditional enterprises still struggle with digital transformation. Project managers, product managers, executives, and entrepreneurs at the forefront of this era are all pondering the same question: What challenges will project and product management face in the AI era? Artificial intelligence product teams develop products. To clarify terminology, when we refer to "AI product management," we mean the development of AI-driven products.

Infrastructure vs. Applications

An important distinction lies in focusing on AI-driven applications rather than the underlying AI infrastructure involved in model training. This distinction parallels the difference between platform products and experience products. Platform products enable experience products. Both are significant, but the majority of AI product managers will oversee experience products-applications-so our focus here will remain on these.

The Nature of AI-Driven Products

Most products carry significant risks, and product teams are cross-functional, equipped with diverse skills to address these risks. Few products highlight the urgent need for robust product management more than AI products. An AI product leverages AI technologies to create problem-solving experiences for customers or businesses. The term "AI" encompasses both traditional AI (e.g., machine learning) and generative AI. These technologies enable functionalities such as intelligent recommendations, personalized experiences, or marketplace matching. Examples of AI applications include smart home devices using voice and natural language processing, fraud detection systems, and advanced generative AI capabilities like content creation, summarization, and synthesis.

AI products pose unique challenges in terms of risk. Product managers, designers, and technical leads must collaborate closely to find effective solutions. While AI product managers may not have machine learning scientists as core team members-especially in application-focused contexts-they must consult such experts. This collaboration is critical to leveraging underlying AI technologies effectively.

Technical Feasibility Risks

Generative AI is inherently probabilistic rather than deterministic. Traditional solutions typically produce consistent outputs for identical inputs. In contrast, generative AI systems may process billions of inputs, with model weights evolving through learning, potentially yielding different outputs over time. Some products and features align well with probabilistic solutions; others do not. This is a fundamental consideration. For instance, occasional mismatches in a personalized news feed may be manageable, but a product controlling insulin dosage must strictly adhere to medical guidelines.

AI product managers must ensure the technology aligns with the product’s purpose. This raises critical quality assurance questions: What is the acceptable error rate? What types of errors might occur? How will the product handle each error? Can user experience design mitigate errors?

Training data quality is paramount. Product managers must deeply understand the data and model training processes. All large datasets carry potential biases and limitations. While the ethical implications of data bias will be discussed under business continuity risks, AI product managers must grasp how these issues manifest in the final product. For many AI products today, the biggest hurdle is training data itself-insufficient quantity or quality to support viable commercial solutions.

When addressing technical feasibility, AI product managers must collaborate with technical leads and machine learning scientists to make optimal trade-offs. For example, high-precision models may require substantial investments in training data, processing power, and computational resources, impacting user experience, scalability, and cost. Technical debt and infrastructure must also be considered: Does the company possess the infrastructure to support AI products? Factors like data storage, processing capacity, and ongoing maintenance costs matter. Excessive technical debt can hinder scalability, technical feasibility, and business continuity.

Usability Risks

User experience is critical for any product but gains new complexity with AI. For AI products, UX design must clarify what the technology can and cannot do, as well as how the product works conceptually. Transparency builds trust and mitigates frustration when limitations arise. Traditionally, product managers rely heavily on designers to establish user trust. However, AI introduces additional constraints and complexities, many stemming from probabilistic outputs.

Users and customers need assurance about data usage and AI capabilities, which may require novel interaction paradigms. Designers and AI product managers must work closely to ensure AI experiences are intuitive, trustworthy, and easy to understand. In some applications, explaining the "why" behind AI decisions becomes crucial. This transparency fosters confidence. What level of explain ability is necessary to build trust? Similar to feasibility assessments, product managers and designers must analyze trade-offs affecting UX.

For example, a highly accurate AI recommendation system might slow response times, frustrating users, while a simpler model may sacrifice accuracy for speed. Balancing accuracy, speed, operational costs, and UX is essential.

Value Risks

Value remains a core risk. AI products promise immense value, driving global adoption. Yet many today are merely buzzword-driven. The AI product manager’s primary duty is to ensure AI-driven features deliver tangible incremental value-solving real problems better than existing solutions or addressing previously unsolvable challenges. Avoiding AI for marketing gimmicks or competition is critical. The product’s value must be clear and compelling.

As with complex features, value assessment requires combining quantitative evidence (e.g., A/B testing) with qualitative insights (e.g., user testing). Collaboration with product marketing is also vital to communicate value effectively. Marketing efforts must address user privacy and ethical data use, ensuring these points are clearly conveyed where appropriate.

Business Continuity Risks

Despite AI’s potential to deliver value, business continuity challenges are significant, and missteps in this area dominate headlines. For any product, effective marketing, sales, service, financing, monetization, legal compliance, and regulatory adherence are essential. For AI products, these risks are amplified.

From a unit economics perspective, AI products are nascent but costly. Additionally, questions around data sourcing, copyright, bias, and consequences of probabilistic recommendations persist. Companies are still grappling with legal liabilities and implications. Ethical considerations are increasingly urgent-beyond data bias, what are the legal and moral ramifications if users misinterpret results or models produce dangerous "hallucinations"? Probabilistic AI systems can save lives through superior accuracy or endanger them through errors. Companies must proactively address these issues.

AI product managers must also anticipate misuse by bad actors. Protecting company assets and reputation is part of business continuity. Depending on use cases, AI products may have societal or environmental impacts. Product managers must analyze these risks and collaborate with legal teams to safeguard customers and the company.

In summary, AI product managers bear full responsibility for navigating these business continuity risks. Success requires deep understanding of users, data, business, and markets. AI literacy is also critical-like mobile product managers, all product managers will eventually need foundational AI skills. In the future, most product managers will likely become AI product managers, necessitating knowledge of AI mechanics, risk landscapes, and mitigation strategies.