The Hidden Risks of AI in Supply Chain: Adoption, Resistance, and Missed ROI

Nine in ten companies are currently investing in AI, yet only one in four achieves the ROI they set out to get. This gap comes down to organizational readiness. We recently hosted a webinar with Mauricio Dezen, founder of Logica Ops and a strategic Streamline partner with over 35 years of experience in supply chain leadership, to discuss why AI projects fail and what separates the ones that succeed.

The AI paradox: the ROI is real, but most companies don't capture it

According to McKinsey's State of AI 2025 and Gartner 2026, less than 4 in 10 organizations that invest in AI platforms can demonstrate measurable enterprise-level financial impact, and only 1 in 4 reaches their stated ROI objectives.

When implementations do work, the results are substantial. One business Mauricio implemented with Streamline was a $50 million company that freed $18 million in working capital within six to eight months by changing how it makes planning decisions. That company has since grown nearly threefold. 

In a second case, a company tripled revenues while reducing inventory from $40 million to under $30 million; without AI, maintaining the same inventory-to-revenue ratio would have pushed their stock above $100 million.

"The ROIs are obscenely high. I guarantee you, they are there. But you need to understand where you are and what you want to do."

Most organizations invest and still don't reach those numbers, for reasons that are almost always organizational.

Three types of AI adopters

Spreadsheet Survivors

The Spreadsheet Survivors implement an AI platform but never change the internal behaviors around it. Spreadsheets remain the real source of truth, and AI gets used as a reporting layer. The root cause is consistent across cases: the organization changed its software but not its decision-making process.

Beyond the wasted budget, the cost here is institutional. A failed AI project consumes the credibility needed to try again, and going back to leadership after 18 months of frustration with no results is close to impossible. In a field where one year of AI development is equivalent to five to ten in traditional business terms, that kind of delay puts companies meaningfully behind competitors who got it right.

"It's 90% human behavior and 10% technical. You have to change the culture."

Cautious Climbers

The Cautious Climbers make up roughly 60 to 70% of AI adopters, according to Mauricio. These companies test AI on a subset of products or locations before committing more broadly, which is a reasonable approach. Where they run into difficulty is in how they evaluate progress. Traditional forecasting optimizes for accuracy; AI-based planning optimizes for business outcomes such as working capital, service levels, and stockout reduction. When teams use forecast accuracy as the benchmark for an AI platform, they end up measuring incompatible outputs, and the AI appears to underperform when it is actually delivering something more valuable.

"When you compare AI tools with only forecasting, sometimes it looks like the AI's performance is worse. It's not. The output is different."

Benefits emerge gradually for cautious climbers, with bigger gains coming as trust and platform maturity build. Progress accelerates significantly when teams commit fully to a defined subset early on, whether that is one product group or one location, generating clear business outcomes that make the case for broader rollout.

Fast Adopters 

The Fast Adopters embrace AI recommendations early, shift planners from data manipulation to decision-making, and measure planner adoption rather than system utilization alone. This profile consistently produces measurable ROI within the first months. The $18 million working capital result described above came from exactly this combination: a new supply chain director with a transformation mandate, an engaged user, and structured SIOP support from Mauricio's team. Within weeks of go-live, the company cancelled 70 to 80% of outstanding purchase orders and moved from 30 to 10 weeks of safety stock.

"It's a combination of cultural change, a strong sponsor, and a clear understanding of what they can get from AI."

The vision gap

AI failures are set in motion before a contract is signed. Without a defined outcome in mind, companies select software before defining what they want to achieve, run through endless vendor demos and RFPs without a clear filter, and end up treating AI as a reporting tool rather than a decision engine.

A strong AI vision addresses four things: 

• the business outcomes to achieve

• the decisions AI will influence

• the process changes required

• the ROI targets

An ERP organizes data for humans to act on. An AI planning platform generates the decisions itself, covering what to order, when, and how much safety stock to carry. Recognizing that difference before signing anything is what separates organizations that extract value from those that end up with a more expensive status quo.

"If you don't know where you're going, AI won't get you there."

The executive champion

AI transformation reshapes people, culture, processes, and decision-making simultaneously, and none of these can be changed by software alone. Without a C-level sponsor, AI tends to get classified as a supply chain or IT project, middle management protects existing processes, users revert to spreadsheets, and ROI never materializes.

With the right sponsor driving alignment, implementation timelines compress and financial outcomes become visible across the organization.

"When you have that kind of sponsorship, it moves really fast. It can cut implementation from 18 months to 3 to 4 months. The faster you shrink that, the faster the company sees the ROI."

The success formula

Vision, executive sponsorship, behavior change, and technology are the four components Mauricio identifies as necessary for successful AI adoption. Most organizations focus only on the last one. The most successful invest in all four.

"I won't start an AI transformation without at least a draft mapping of all four."

Outcome targets also need to be framed as business results. Forecast accuracy is a metric, while reducing working capital by a specific dollar amount is a business outcome. That framing determines what the organization optimizes for and whether leadership will be able to recognize the return when it arrives. 

Waiting for a perfect plan is not a viable option either: one year in AI is roughly equivalent to five to ten in traditional business cycles, and organizations moving now are compounding that advantage quickly.