Artificial intelligence (AI) has found its way into almost every solution in the supply chain as a way to boost sales and margin, while sparking intrigue. In the race to gain a competitive advantage, or just to remain relevant, many organizations are rushing to adopt AI in ways that can sometimes be superficial, ineffective, and ultimately fail to deliver real value.
For example, in the space of supply chain applications, only a small fraction of AI applications deliver true value through solid return on investment. AI is a reality, but many solutions aren’t able to address company needs to effectively address, let alone solve, real supply chain challenges. It’s key to look past the noise to ensure any implementation is able to deliver on its promise.
The AI value gap
Despite significant investments in AI-powered supply chain solutions, most organizations struggle to realize meaningful benefits. The disconnect is often driven by the hype surrounding AI, where excitement can overshadow actual value. A Gartner survey showed that Generative AI is being deployed by 72% of supply chain organizations, with only modest results for productivity and ROI to date.
Challenges exist on both the vendor and buyer sides. First, is the vendor properly representing its expectations on the desired AI functionality, clearly defining the pain point to be addressed, detailing the expected value targets? Do supply chain leaders have enough insight to distinguish between genuine innovation and sophisticated marketing, or understand how to derive the most value from their tools?
Decision makers may sometimes fall prey to buzzwords, overlooking demonstrable performance improvements. Supply chain planning requires precise, context-specific calculations within complex and volatile environments that current AI models can’t reliably deliver without specialized design, or without taking significant risk. A misalignment with supply chain complexities can leave users with value reduction instead of value add.
The AI doesn’t add up
The mathematical complexity of supply chain operations presents a significant challenge for general-purpose AI applications. Supply chain management requires sophisticated algorithms capable of processing multiple signals simultaneously, including demand patterns, industry trends, seasonality, and broader market forces.
Generative AI can’t (yet) manage these issues alone. Big data can provide guidance to create better demand models that account for multiple dimensions and probabilities. But, effective decision-making also requires modeling management objectives and supply factors; both of which are highly volatile and unclear. Generative AI systems that aren’t trained on properly processed and calibrated data across these areas tend to hallucinate, producing unreliable outputs, leading to user distrust and system failure. Without trust, these systems will ultimately fail.
AI as a decision support system
AI can act as a support system for decision-making, to help companies make better decisions through forecasting, planning, and executing more accurate product decisions while pricing to value and managing inventory levels dynamically.
The most successful AI implementations are collaborative in nature, designed to enhance decision-making rather than replace it. Many inter-connected decisions cannot all be optimized at once, though. Managers should start from short-term or real-time decisions, which are easily observable and improved upon.
Through this process, they can gain confidence on strengths and weaknesses of the systems and understand the reasoning rather than blindly accepting it. Only then can they gradually move on to medium-term and long-term decisions, such as planning, where the learning journey is longer given the intrinsic delay in the feedback loop.
User friction and transparency
Even if an AI solution can show improvement in key areas, its value becomes negligible if users don't understand or trust it. Consider this mathematical formula: Value = Impact / Friction.
Many implementations fail because they do not provide demonstrable value. Even those that do, risk being abandoned due to workflow friction: complexity of adoption, inappropriate training, lack of transparency. Without transparency, users will tend to override a system, ignore its recommendations, or revert to manual processes, removing any potential benefits. The focus should remain on measurable outcomes while ensuring the interface builds trust through clear explanations of recommendations.
More importantly, users should focus on expanding their capability at translation, which means formulating clear and action-defined instructions that a machine can appropriately interpret and ultimately understand.
Take for example the Harvard Business School case of Pittarosso, “Artificial Intelligence-Driven Pricing and Promotion.” Here, company managers learned how important and difficult it is to define an objective function that a machine can efficiently implement. What “good” looks like, after all, remains for us to define, as machines cannot navigate the ultimate ambiguity of context required for setting effective goals. Humans, albeit empowered and augmented by AI, remain ultimately the key success factor.
Identifying the value
Decision makers can separate valuable AI implementations from superficial ones by evaluating:
● Performance: Solutions should improve specific metrics, and do so quickly. Leaders should request evidence of performance improvements in environments similar to their own.
● User experience: A solution should clearly explain recommendations in language that makes sense to supply chain professionals, creating transparency and trust.
● Value in use cases: Focus on well-defined and narrow challenges, especially initially, rather than broad capabilities across numerous functions.
Valuable solutions should demonstrate deep integration with supply chain-specific solutions, rather than standard AI capabilities. This implies demonstrable decades of research and development effort rather than just empty promises. With these evaluation criteria in mind, decision makers can focus on solutions that deliver actual results.
Beyond the buzz
As AI continues to evolve, leaders need to evaluate solutions based on their ability to address specific challenges. The most valuable implementations demonstrate value through clarity and performance: impact divided by friction, mathematical sophistication with user-friendly design.
By focusing on function, value and a return on investment, organizations can avoid costly solutions that don’t deliver.
Success will belong not to those who implement the most AI, but to those who identify and deploy the right AI applications with a specific, measurable goal in mind.
