Factors Limiting the Application of AI in Supply Chain

Factors Limiting the Application of AI in Supply Chain

Because of the immense potential of AI, it’s easy to get deviated and undermine the primary challenges that companies face in adopting and using AI efficiently.

FREMONT, CA: The potential of artificial intelligence (AI) to transform the industries and its processes makes it an indispensable technology. Cost-cutting and streamlined processes are merely a fraction of what the technology has to offer. AI has immense potential in the supply chain with its ever-improving algorithms and potential opportunities in the supply chain that can be easily leveraged. For instance, with more than 90 percent of seafaring accidents resulting from human error, AI could cut down on such cases significantly.

Vehicles can also share important data such as speed, position, and heading using vehicle-to-vehicle communication. The information can alert autonomous systems and drivers against possible threats and collisions proactively. It can even alert the drivers about relatively lighter and less visible vehicles, such as motorcycles and bicycles. V2V, when equipped with AI, can reduce congestions and collisions by alerting pilots, drivers, and captains to anticipated collisions with other vehicles as well as traffic hazards. AI has a key role to play even in terms of innovation, productivity, and global economic growth. As per McKinsey & Company, the adoption of AI has the ability to deliver an additional global economic activity of over $13 trillion by 2030, which is about 16 percent higher cumulative GDP than today. It amounts to 1.2 percent additional GDP growth each year.

Considering the immense potential of AI, it's easy to get deviated and ignore the primary challenges that companies face in implementing it efficiently. Here are the major challenges and primary solutions:

Shortage of Clean Data

Computational processes require quality data and AI results in desired results if such data set is available to it. Machine learning (ML) needs huge volumes of accurate data to train its algorithms and to build predictive models. However, not many companies have the quality or quantity of data. Companies are required to enhance their data quality via effective master data management and by including real-time data into their processes and systems as often as possible. Multi-party, real-time digital business networks maintain an authentic record of the proceedings while continuously synchronizing external systems and ensuring that the processes are running on the latest information possible.

Compartmentalized AI is not an Effective AI

Supply chains are innately cross-enterprise and cross-functional, and the data required to operate them is scattered among external and internal partners. Companies that are trying to implement AI in a fragmented way without considering the big picture are likely to end up with poor results. Without access to relevant data, algorithms will have blind spots and will miss opportunities for execution and optimization.

It is important for the companies to include relevant operations, systems, and trading partners to strengthen the context, accuracy, and completeness of data. The goal must be to connect the entire supply chain with a real-time network, from its source to the end customer.

Black Box and Explainable AI

ML techniques such as decision trees and scorecards are easy to understand. However, neural networks are not so simple. Would the right course of action be to act on this data or allow the system to act autonomously?

For instance, Amazon’s experiment in using AI to recruit talent proved to be a biased one when the researchers noticed that the system was inclined towards hiring male candidates. It happened as the algorithms were trained predominantly on male data. The AI downgraded the female candidates and made decisions based purely on gender.

Thus the companies must be aware of how the algorithms work and how they arrive at a particular decision. Companies must also be capable of altering the algorithms to meet their desired needs.

Short-Term Optimization

Each process and change has a cost. If it’s not factored into decision-making, the result can be worse than if nothing had been changed at all. It is easy to lose track of the long-term effects of an action in case of a supply chain. Several solutions get into this trap by re-planning the complete supply chain resulting in unnecessary distortion in the system when a minimal change could have sufficed the purpose. Such a problem can be avoided when optimizations are used as a continuous process instead of occasional and are limited to impacting the least number of entities.

Weekly Brief