The progress of supply chains in adopting artificial intelligence varies widely. So does the return on investment. But according to one recent study, there’s one strategy that guarantees failure: doing nothing.

“In every era of transformative change, there are early adopters and fast followers,” says the report, “The Ultimate Guide to AI ROI,” from research firm Zero100. “And the chasm between those who act fast on AI and those who wait and see will be one of the greatest of our generation.”

Failure to move forward with an AI adoption strategy is “catastrophic,” the report adds, quoting a tech executive who spoke with Zero100 during its preparation.

AI’s long-term impact promises to be nothing short of revolutionary. “The global supply chains of tomorrow will be AI-empowered and fully digitized from end-to-end — a complete convergence of supply chain and IT that will fundamentally change the nature of supply chain work,” says Kevin O’Marah, co-founder and chief research officer with Zero100.

That said, business leaders’ commitment to adopting AI to date has been anything but consistent. Around 90% of large businesses have experimented with AI in their supply chains, the report says, and 29% say it’s an area for “heavy investment” over the next three years.

At the same time, just one-third of executives have a “strategic vision” for integrating AI and machine learning into supply chain functions, according to an Zero100 analysis of company earnings calls. And only a quarter of leaders are seeing “tangible returns” so far.

One could argue that the arrival of every transformative technology entails a certain amount of stumbles, wrong turns and dead ends. In the case of AI, business leaders are still struggling to puzzle out where in the supply chain it can deliver the best value in the immediate term.

To a certain degree, they’re emboldened by early successes in what can be termed “classical AI” — the application of mathematical algorithms derived from human-made rules. But that kind of AI has progressed in fits and starts over the decades, gradually taking hold in the business word. Only in the last year or so, with the emergence on the scene of generative AI and large language models — as embodied in such tools as ChatGPT — has the technology’s full promise become clear.

Early adopters may have been overly enthusiastic about AI’s short-term potential. O’Marah cites a “hype component” affecting business leaders who anticipated big productivity gains that were slow to materialize. Spurred by some early successes in limited applications of AI, “their excitement was a little bit ahead of what’s currently working,” he says.

That’s no reason to doubt the true potential of AI, he says. The system’s underlying dynamic is that it learns by doing. Every new iteration, therefore, is an improvement over what came before. When he worked for Amazon.com, O’Marah recalls, he was witness to “a constantly improving set of tools for allowing delivery vans to follow an optimal route.”

AI is also rapidly getting better in performing such key functions as truck load building and answering questions about transportation execution based on an ever-growing storehouse of experience, O’Marah says. “These proven methods of using traditional reinforcement learning in a machine-learning capacity are working. And have been working for a while.”

AI today can mine a treasure trove of content from social media — far more than any human could ever process — to anticipate supply chain disruptions such as strikes, port closures and natural disasters, O’Marah notes.

With the ultimate value to the supply chain of AI and machine learning not in serious dispute, the question becomes: How can it be adopted it in a way that realizes the fastest possible ROI?

Zero100’s report is intended as a playbook for achieving that end. Like any good consultant, the firm recommends that companies begin their journey by defining what outcome is driving their AI investment, and what they hope to achieve with the technology in the end.

“Are you aiming to leapfrog competitors, revolutionizing your supply chain with bold moves?” the report’s authors ask. “Or do you seek incremental innovation, nudging the edges of existing processes?... Clarity here will guide the scope of your AI investments.”

O’Marah says it’s important at the outset to establish a dedicated AI team within the supply chain organization, then collaborate with functional users of the technology. In that way, he says, they can draw on the wisdom of “citizen technologists,” using their invaluable experience to transition from planning to real-life application of the desired tools.

Developers can adopt “small language models” to address specific problems to be solved — for example, minimizing risk in the purchase of a particular material. Simultaneously, users need to begin training the generative AI model on the fundamentals of their operation. At a certain point, the model becomes sufficiently powerful to exceed the ability of humans to make complex decisions. “Once it’s there,” O’Marah says, “it’s massively scalable.”

Certain instances of GenAI have garnered skepticism in recent months by outputting demonstrably false information, or “hallucinating” absurd scenarios. (Nobody followed the recommendation of Google AI to put glue on pizza.) O’Marah says such glitches are to be expected as the model learns and adjusts to the real world. “Until you’ve debugged it, you’re going to run the risk of getting weird faulty statements.”

None of which should dissuade companies from progressing with AI adoption, provided that they do so in a measured way that couples a frank assessment of their current state with a long-term vision.

For all its dazzling allure, AI will never function effectively in a supply chain setting without full attention to the human element — leading to what the Zero100 report terms “an AI talent dash.”

“Supply chain leaders need an AI-literate workforce, which means ramping up hiring and upskilling for critical AI and machine-learning skills,” report says.