As Artificial Intelligence (AI) systems permeate society, organizations must prioritize building trust, managing risks, and adhering to regulatory standards. While some concerns regarding AI can be overblown, legitimate risks demand attention. Instances of racially-biased identification of shoplifters using facial recognition software or AI tools leading to sexist hiring preferences are just two examples that underscore the tangible threats posed by AI systems.

The retail industry in particular stands to benefit greatly from AI technologies, as they enable more effective and efficient customer interactions. Nevertheless, it’s essential to address concerns of bias, privacy, and fraud risks while preparing for upcoming regulations. In this piece, we walk through key considerations and strategies for AI risk management in retail, including navigating today’s evolving regulatory environment and effectively deploying AI systems while fostering customer trust. 

Navigating the AI Regulatory Landscape 

As international and domestic regulations surrounding AI evolve, organizations must proactively develop an internal culture and policy framework to adhere to these new standards.  

The recent EU AI Act classifies AI into four risk levels: minimal, limited, high, or unacceptable. While most AI applications will fall into the minimal or limited risk categories, this legislation imposes strict requirements on providers of high-risk AI systems and bans those classified as unacceptable. These include manipulative, deceptive practices and systems that categorize by race, sexual orientation, or compile facial recognition images. The goal is to ensure all people interacting with AI systems are informed and confident in their reliability. 

Within the U.S. specifically, the recent Executive Order on AI highlights the importance of labeling AI-generated content. Compliance with this order will entail new technological implementations, a possible increase in scrutiny from government regulators, and challenges for retailers navigating the complexity of labeling content generated partly by AI or for use in advertising or product descriptions. However, it’s also a means to enable transparency with customer-facing AI systems.  

Additionally, the Federal Trade Commission (FTC) resolution enables streamlined investigations into AI-related practices, emphasizing accountability and consumer protection. Legal action can and has been taken to hold organizations that use AI-generated content or algorithms accountable. While these measures can enforce a level playing field among retailers and catch bad actors, they also signal that agencies will protect consumers from companies’ negligent or irresponsible use of AI.  

Lastly, many state-initiated laws aim to integrate AI regulations into broader consumer privacy laws, further accentuating how widespread these regulations are and the necessity to ensure compliance at all levels.  

With all of this in mind, data and analytics leaders must stay on top of these regulations to ensure any AI tools comply with this complex and changing environment.  

Building Trust: Key Strategies for AI Risk Management in Retail 

The most effective way to accommodate the evolving regulatory environment is to proactively implement a risk management framework. Organizations can use a structured approach to identify, assess, prioritize, and mitigate risks while working toward their overarching AI objectives and goals. This framework is crucial in establishing customer trust and ensuring compliance with future regulations.   

Based on the National Retail Federation’s guiding principles and National Institute of Standards and Technology’s (NIST) AI Risk Management Framework, here are six essential strategies to help you navigate this process effectively:  

1. Education and Awareness: Educate stakeholders and staff on AI fundamentals, risks, and benefits through clear, accurate information, open and transparent dialogue, and training programs.   
2. Governance: Align AI risk management strategies with existing company policies and establish governance structures that support a culture of awareness, transparency, accountability, and risk mitigation.  
3. Risk Assessment: Conduct comprehensive risk assessments throughout the AI lifecycle, including mechanisms to monitor the use of AI tools by employees and customers. Leaders need to consider the decision rights and impact of AI tools on outcomes and training methods, the transparency of model decisions, and the management of any drift or unexpected changes throughout their use.   
4. Privacy and Security: Implement rigorous privacy and security measures to safeguard AI systems against cyberattacks, data breaches, and unintended consequences. Regular monitoring and updates are essential to address emerging threats effectively.  
5. Ethical considerations: Mitigate bias and ensure inclusivity in AI practices by promoting diverse representation in data sources, collection techniques, governance leadership, and development teams. Inform customers on how your policies will prevent or mitigate the risk of discrimination.  
6. Customer Experience: Integrate risk mitigation policies and procedures with business partners supporting your AI tools. Coordinate across teams and systems to deliver a smooth, transparent customer experience when they use your AI services. 

Implementation Considerations 

When implementing the strategies above, it’s essential to cater to the specifics of your unique market and customer base. While customer experience is a pivotal factor guiding your efforts and will validate the efficacy of your decisions and strategy, scrutinizing implementation details for transparency and privacy is equally important. 

The level of transparency and security required for your risk management framework will depend on the audience. While providing customers with a basic explanation of what the AI system does and how it identifies information (i.e. system properties and content labeling) may be appropriate, a regulator may require access to source code and documentation from developers on how the system functionality is validated. This variation of needs must be accounted for to optimize customer experience while having the processes and mechanisms in place to safeguard against litigation or audits. 

Privacy considerations, on the other hand, should be different for each use case. Variations in the sensitivity of the data involved necessitate different protection levels across diverse applications. While a chatbot for everyday product inquiries and suggestions might offer minimal data privacy guarantees, other systems handling customer financial data may enforce stringent encryption protocols and operate on dedicated servers. As such, implement your AI risk mitigation with nuances and formulate nimble processes to adapt and evolve to your specific use cases and requirements. 

Final Thoughts 

AI trust and risk management represent critical pillars of responsible AI governance. By embracing proactive strategies and leveraging impactful processes and resources, organizations can foster trust, mitigate risks, and navigate regulatory complexities effectively – ultimately enabling them to better serve their customers and achieve unprecedented advancements in productivity and innovation.  

Clarkston’s data and analytics experts can support your journey and empower your organization to safely deploy AI technologies that enhance customer experience. 

Contributions from Kevin McDonough