AI and Employees: Crafting Successful Teams

Editor’s Note: The SCM thesis Human-Machine Teaming in AI-Driven Supply Chains was authored by Christoph Herrmann and Libin Huang and supervised by Dr. Maria Jesus Saenz ([email protected]). For more information on the research, please contact the thesis supervisors.

AI adoption increases, but project success often lags. Over the past five years, artificial intelligence (AI) has grown from a buzzword to real-world business applications in supply chain management among the world’s largest companies, covering areas such as demand planning and predictive maintenance.

By applying cognitive functions that mimic human thinking, AI provides tremendous opportunities as it continually learns from human decision making. However, a staggering number of firms abandon their AI projects or consider them failures, leading to billions of dollars in write-offs of such projects. In 2018, a McKinsey analysis showed that a predominant source of AI project implementation failures lies in the lack of interaction and teaming between the artificial intelligence and the employee.

Distinct capabilities are needed to make AI projects a success

To prevent failure, it is important that organizations build human-machine teaming capabilities. Transparency is required to facilitate the trust from human employees; authority balance ensures that the human team member remains in control and the AI takes over only where appropriate. Secure interaction establishes that the teaming process is protected and employees consider the application ethical, reliable, and secure. These capabilities facilitate mutual learning between the AI and the human employee — the growth in abilities between both partners — which in turn determines the AI project’s success.

However, the importance of these capabilities is dynamically dependent on the decision context. Prior research suggests that depending on the degree of risk of the decision (e.g., financial impact) and the degree to which the AI’s design is open or closed (i.e., whether unsupervised learning is present), human-machine teaming capabilities are configured differently.

Finding the right balance of capabilities

Analyzing 22 AI projects in the field of supply chain management, it was found that successful AI projects show seven fundamental teaming capability traits when breaking down the capabilities described above into 24 underlying indicator concepts. To build transparency, they emphasize the ease of understanding of AI outputs (explainability) and ensure easy interoperability between the human and AI. To facilitate authority balance, humans must have the “final say” (directability), while the AI corrects human slips and errors and achieves a reduction of the employee’s mental workload. To enable secure interaction, employees especially expect acceptable social conduct and secure means of interaction with the AI.

As AI projects are undertaken in a decision context characterized by more risk and an open AI design, additional traits, such as synchronized feedback loops and higher levels of information sharing between the AI and the human team member, become prerequisites of success. The framework enables supply chain managers to assess their decision context, and then build the right teaming capabilities. However, managers should also continually monitor how decision contexts change as their AI project evolves and which new capabilities need to be built.

Analysis of the projects also established additional learnings for practitioners. For example, it was revealed that employees do not necessarily need to understand the AI’s algorithms mathematically but need to grasp the intuition behind the AI’s outputs, which can be facilitated by visualizations and easy-to-use dashboards. Moreover, many AI projects still fall short on unleashing the power of human feedback.

Rather than providing binary feedback from the employee to the AI (e.g., through thumbs-up or thumbs-down buttons), firms should offer richer feedback capture mechanisms, potentially by leveraging natural language processing. Similarly, successful projects automatically track whether the AI’s recommendation was implemented by the employee and use this input for the algorithm to learn.
Ensuring the trust from employees in the AI effort is of utmost importance. In this context, it is crucial to take seriously employees’ fears of the AI taking over their jobs.

Successful firms focus on designing AI applications that enhance human capabilities and communicate actively how the AI makes their jobs easier. To facilitate this, it is crucial to involve employees early on in the design process of the AI project. Utilizing cross-functional teams with representatives from data science, IT, supply chain management but also the actual users of the AI solutions eliminates fear and enhances the usability of the outcome.

In the process, outperformers also show flexibility in project management which means they set out for a clear and ambitious vision but also iteratively improve a minimum viable product based on user feedback to reach individual user milestones.

Taking a capabilities-driven approach based on the AI project’s decision context and considering the managerial recommendations, you can ensure more successful teaming between humans and AI.

Every year, around 80 students in the MIT Center for Transportation & Logistics’s (MIT CTL) Master of Supply Chain Management (SCM) program complete approximately 45 one-year research projects. The students are early-career business professionals from multiple countries, with two to 10 years of experience in the industry. Most of the research projects are chosen, sponsored by, and carried out in collaboration with multinational corporations. Joint teams that include MIT SCM students and MIT CTL faculty work on real-world problems. In this series, we summarize a selection of the latest SCM research.

SC
MR