Reports of supply chains adopting generative AI are rich with stellar results: fewer errors, faster response times, leaps in productivity.

The problem: For the moment, such outcomes are rare. That’s because the number of businesses and supply chains to have taken GenAI beyond the pilot stage to full implementation is relatively small.

So finds a new report on AI readiness from GEP, a vendor of AI-driven procurement software, and the University of Virginia’s Darden School of Business.

According to the research, more than half of surveyed supply chain professionals are deploying AI in some manner, but fewer than one in 10 “have scaled AI pilots into enterprise-wide operations.” Seventy-four percent haven’t even gotten beyond the planning stage, or laid out a roadmap for proceeding. Clearly, there are some major operational hurdles that stand in the way of extracting the technology’s full value.

Michael DuVall, GEP’s global head of strategy, says researchers suspected something was amiss with AI, robotics and related tech initiatives nearly a year ago. “We were hearing that projects were stalling out, that clients were in a wait-and-see mode.” As many as 95% of AI investments were said to be failing.

That anecdotal pessimism led to the launch of the new report. Heading up the initiative was Tim Laseter, professor of practice at UVA Darden. He and his team drew in part on case studies from Amazon.com, logistics services provider C.H. Robinson, and Harvard Business School. In all, they surveyed some 180 senior supply chain executives, from the C-suite to vice president, director and manager levels, across 12 industries.

Turns out that the culprit behind the lack of AI progress wasn’t the technology itself. It was the sticking point that bedevils just about every new tech implementation: a lack of proper business processes to support the change.

Many if not most AI projects are driven from the top down — but they tend to sputter out as they descend the organizational chart, DuVall says. Energy dissipates; messaging becomes absent or vague, and everyday operating concerns intrude. In addition, many of the companies surveyed were attempting to layer AI over old and broken processes. Laseter adds that they were treating the effort as a routine software installation, rather than one involving “operational transformation,” and requiring a sharp focus on change management.

The result for many such projects is “pilot purgatory,” assuming the program even gets that far. Companies soon discover that working with GenAI on a corporate level is a lot different than an individual using ChatGPT to write a term paper or answer a trivia question.

The irony is that a technology that many believe will eventually replace human workers requires the expertise and participation of people to implement properly. Laseter says the success stories — the “performance elite,” in the researchers’ words — often involve the onsite presence of PhDs with deep knowledge of AI, coupled with individuals on the front line who are skilled in process management.

Tools and processes that prove critical to an AI undertaking include automated data cleansing, real-time dashboards and digital audit trails, the report says. That squares with AI experts who repeatedly stress that the technology is no more accurate than the data that’s fed into it.

Also of value is taking a “portfolio” approach, under which executives might be nursing along multiple AI projects on different timelines. A successful implementation within one function could serve as the basis for moving ahead with the next. In that way, businesses can spread out their investment and assignment of resources.

Researchers further found that organizations with the greatest success in scaling AI initiatives relied on the direction of a dedicated steering committee, consisting of experts from multiple functions and disciplines. A third of those lacking such a structure “had no systematic view of opportunities at all,” the report states.

Even the “performance elite” has room for improvement, as they struggle to absorb AI into the organization. “One of the things we found,” says Laseter, “is that even in the best cases, no one is knocking it out of the park on all dimensions.”

What’s often still missing, the report notes, is the failure to fully address stakeholder engagement and talent management, “suggesting the next competitive advantage will come from a better-prepared workforce, not a better model.”