Verified Author Share
Notes from a closed-door dinner with enterprise, mid-market and startup leaders — April 30, 2026
Supply chain AI adoption is failing because organizations are deploying tools without operational readiness. The five most common failure patterns identified at a closed-door dinner of fourteen supply chain, logistics and freight-tech leaders in April 2026 were: (1) top-down mandates without bottom-up readiness, (2) erosion of undocumented operational knowledge (“tribal knowledge”) as AI is trained on procedures that don’t match reality, (3) tool fragmentation where each business unit independently selects its own LLM or copilot, (4) commoditization of historic competitive moats like proprietary pricing data, and (5) widening skill polarization between high and low performers.
These observations align with broader industry data: RAND Corporation’s 2024 research found that more than 80% of AI projects fail — twice the rate of non-AI IT projects, and Gartner’s April 2026 research found that 57% of infrastructure and operations leaders have experienced at least one AI project failure. The dinner’s diagnosis was qualitative and human; the data confirms the pattern is structural and widespread.
On a Thursday evening in late April, we closed the doors and sat down with a group of supply chain, logistics, manufacturing and technology leaders. No slides. No vendor pitches. No agenda beyond a single question:
What is actually keeping you up at night?
For two and a half hours, the room argued, agreed, pushed back, and surfaced patterns that none of us had heard articulated this clearly before. By the end of the night, an uncomfortable consensus had emerged — and it had nothing to do with which AI model is best, which platform to pick, or what the next breakthrough will be.
This post is our attempt to share what we heard, in the same spirit the room shared it: candidly, with rough edges intact, and without the marketing gloss that usually obscures these conversations.
If you were in the room, this is a thank-you and a record. If you weren’t, this is a window into what your peers are actually wrestling with — not the version they share at conferences, but the version they share over wine.
The dinner was small and deliberately mixed — fourteen invited guests across five distinct vantage points. The largest single group was a contingent of freight-tech founders and CEOs — leaders of AI-driven logistics startups who spend their working lives selling into the very organizations that the rest of the room represented. Alongside them sat enterprise practitioners from semiconductor, telecom, CPG, and dairy supply chains, a mid-market 3PL leadership team, a boutique 3PL connector, and a few institutional voices from defense logistics, sports operations, and higher education.
Most of the sharpest pain points below — the refinery valve story, the retail VP confessing “we have no idea what we’re doing,” the mid-market companies running disconnected systems — were told by the freight-tech founders describing their clients and prospects, not by the affected organizations confessing in first person. The dinner was, in effect, a cross-diagnostic conversation: founders pointing at the problem, mid-market and enterprise practitioners selectively confirming or pushing back, and the institutional voices mapping the patterns to adjacent industries.
That pattern is itself a finding. The clearest view of an organization’s AI dysfunction often comes from the people trying to serve it.
One caveat worth holding in mind: the room had a clear center of gravity around freight, logistics, and adjacent supply chain operations. Retail was mentioned but not directly represented. Pharma, automotive, consumer electronics, and fashion didn’t come up at all. The patterns described below are most reliable for logistics, freight, manufacturing and CPG-adjacent operations — and should be treated as hypotheses, not findings, for sectors that weren’t in the room.
Every participant came in with a different industry, a different role, a different company size. They left agreeing on this:
The current wave of AI adoption is being driven from the top down, executed without operational clarity, and rewarded by markets that have stopped asking whether companies actually know what they’re doing.
That sentence took most of the evening to construct collectively. It is also the single most important takeaway. Everything else flows from it.
CEOs are mandating AI adoption to satisfy quarterly earnings calls. Middle management is being handed tools without a problem to solve. Operations teams hold the tribal knowledge that determines whether any of it works — and they are the ones being asked to use, and increasingly to fear, the technology.
The result is a fragmented, undermanaged, politically charged adoption curve where the loudest signal is fear of being left behind, and the quietest signal is whether any of the deployments are creating value. This matches what McKinsey’s November 2025 State of AI Survey found at scale: 88% of organizations now use AI in at least one function, but only 39% report any enterprise-wide EBIT impact from AI.
The image of the night came from a refinery operator’s experience, shared by one of the participants. Their AI-driven workflow was instructing morning-shift operators to open a valve all the way — exactly as the standard operating procedure said to do it. Every afternoon, the next shift would arrive and reverse it back to halfway. Why? Because opening it fully would fracture the pipeline.
The SOP was wrong. The tribal knowledge was right. The AI had no way to know.
This story is not really about valves. It’s about the gap between documented process and operational reality — a gap that exists in nearly every organization in the room, that nobody is talking about, and that AI is making increasingly dangerous to ignore.
When you train AI on the documented process while operators run a different one, you don’t just fail to capture value. You actively destroy it. The AI confidently produces outputs misaligned with reality, and the people who could correct it are the same people most likely to leave when their judgment is overridden. Gartner predicts that through 2026, 60% of AI projects unsupported by AI-ready data will be abandoned — and “AI-ready data” is exactly what undocumented tribal knowledge isn’t.
Takeaway: Process discovery is the new prerequisite for AI deployment. If your organization can’t draw its current workflow on a whiteboard — including the parts no one writes down — you are not ready to deploy AI on that workflow.
This pattern came up again and again, in nearly identical form across companies of very different sizes.
Under pressure to “do AI,” each business unit went out and picked its own LLM, its own copilot, its own agent platform. Marketing chose one. Sales chose another. Operations chose a third. Each team is now defending its choice as superior, the IT function is overwhelmed, and there is no enterprise-level governance, no shared knowledge graph, and no clear owner.
This is what happens when the speed of exploration outruns governance. By the time anyone recognizes the fragmentation, switching costs have already calcified, and political capital has been spent defending choices that should have been provisional. Independent research backs this concern: S&P Global Market Intelligence’s 2025 Voice of the Enterprise survey reported that the average organization scrapped 46% of AI proof-of-concepts before reaching production, with 42% of companies abandoning most of their AI initiatives — up from 17% in 2024.
Takeaway: If your organization has more than two AI tools in active use without a single accountable owner, fragmentation is already a problem. Centralizing later is twice as hard as preventing it now.
One of the more provocative threads of the evening was a debate about what AI will commoditize.
The argument went something like this: the moats supply chain companies have built — proprietary pricing data, opaque capacity information, performance metrics that customers can’t independently verify — are about to become transparent. When AI gives every buyer the ability to see every seller’s true on-time delivery rate, true capacity, and true cost structure, the basis of competition shifts entirely.
What replaces those moats? Service relationships. Judgment under ambiguity. The quality of human escalation when things go wrong. Integration depth that can’t be replicated overnight. Trustworthy interpretation of data. And — increasingly — whether your AI implementation has any humanity in it at all.
Takeaway: If your competitive position rests on information asymmetry, it is on a shorter clock than you think. The companies investing now in service quality and human relationship depth will outlast the ones optimizing for the moat that is about to disappear.
The educators in the room shared a pattern that should concern anyone planning workforce strategy.
The strongest students and operators are using AI to produce work that would have taken a full team months. They are not just faster — they are operating at a different altitude entirely.
The weakest are using AI to produce faster mediocrity.
The gap between the two is widening, not narrowing.
This has implications well beyond the classroom. It changes how you hire. It changes how you invest in training. It changes which roles compound value and which roles get absorbed. And it forces an honest conversation about what your workforce strategy actually is — because pretending the impact will be evenly distributed is no longer credible.
Takeaway: Plan your workforce strategy around the polarization, not against it. The organizations that pretend AI will lift all boats will be outcompeted by the organizations that explicitly invest in turning their best people into superhumans.
This was one of the sharpest reframings of the night. The word “pilot” signals optionality, soft commitment, and acceptable failure. That framing produces exactly the outcomes it implies.
The alternative — borrowed from operational doctrine — is to treat each AI initiative as a mission. Defined objectives. Real consequences. The assumption of success.
Organizations that adopt the mission framing report dramatically higher follow-through. The language is not cosmetic. It changes who shows up, how seriously the work is treated, and whether the organization actually pays the price required to make it work. MIT’s NANDA Initiative found in its 2025 “The GenAI Divide” research that only approximately 5% of AI pilot programs achieve rapid revenue acceleration — confirming that pilots, by default, are designed to underperform.
Takeaway: Audit your internal language. If your AI initiatives are called pilots, you may be authorizing the result you don’t want. Rename them. The behavior tends to follow.
The conversation kept circling back to a set of tensions that almost every organization is currently trying to resolve by doing both — and paying the cost in fragmentation. We counted six of them by the end of the evening:
None of these has a universally correct answer. What matters is making the choice deliberately, communicating it clearly, and aligning the rest of the organization to it.
The companies struggling most are not the ones making the wrong choice. They are the ones making no choice while behaving as if they had.
Two and a half hours of conversation produced more clarity on what is broken than on what to do next. That asymmetry is itself a finding.
The organizations that will win the next 24 months are not the ones with the boldest AI strategies. They are the ones with the most disciplined operational ones. Specifically:
Discover before deploying. Map the workflow as it actually runs. Surface the tribal knowledge. Close the SOP-vs-reality gap. This work feels slow; it is the only thing that prevents fast failure later.
Govern before scaling. Decide who owns AI deployment. Define the competency required to use it. Build the council that includes both IT and Operations with real authority. Do this before the third tool gets selected by a third team.
Integrate before optimizing. If your systems don’t speak to each other, AI cannot optimize across them. Integration is the unsexy precondition that most companies are skipping.
Mission, not pilot. Define the outcome. Commit the resources. Measure the result. Adopt the language that produces the behavior you want.
Plan the early win. Engineer the first visible success. Communicate it. Anchor organizational memory in it. The primacy effect is real, and you only get one chance at it.
Communicate honestly about workforce. Whatever your headcount intent, name it. Ambiguity is interpreted as the worst-case scenario. Honesty produces faster adoption than carefully managed silence.
Late in the evening, one participant raised a question that the room couldn’t fully resolve:
What does an ethically adopted AI strategy look like in a market that rewards speed?
The question kept returning, in different forms, all night — through concerns about data privacy, workforce displacement, the widening gap between haves and have-nots, and the loss of basic human interaction in customer service.
That question deserves more time than a dinner allows. It will shape, more than any individual technology choice, what the next decade of supply chain looks like.
We don’t pretend to have the answer. But we know it is the right question. And we know that the organizations willing to take it seriously — even when the market isn’t asking them to — will be the ones still standing when the dust settles.
Based on the closed-door dinner of fourteen supply chain leaders in April 2026, the five biggest challenges are: top-down AI mandates without operational readiness, erosion of undocumented tribal knowledge, fragmentation of AI tools across business units, commoditization of historic competitive moats, and widening skill gaps between top and bottom performers. These match broader industry data showing more than 80% of enterprise AI projects fail to deliver intended business value, twice the rate of non-AI IT projects (RAND Corporation, 2024).
Supply chain AI projects most often fail for three structural reasons rather than technological ones: poor data quality (38% of failures, per Gartner April 2026), unrealistic expectations of fast ROI (57% of cases), and persistent skill gaps. Within freight, logistics and manufacturing specifically, the dominant failure pattern observed in our April 2026 dinner was AI being deployed on top of undocumented or wrongly-documented processes — meaning the AI optimizes for a workflow that doesn’t actually exist.
Tribal knowledge refers to undocumented operational expertise that experienced workers hold in their heads but is not captured in standard operating procedures. In supply chain and manufacturing contexts, this often includes critical adjustments — such as opening a valve only halfway despite the SOP saying to open it fully — that prevent equipment failure or operational errors. AI trained exclusively on documented processes will systematically diverge from operational reality unless tribal knowledge is surfaced and incorporated.
The word “pilot” signals optionality, soft commitment, and acceptable failure — which often produces exactly those outcomes. MIT’s NANDA Initiative found in 2025 that only approximately 5% of AI pilot programs achieve rapid revenue acceleration. Reframing AI initiatives as “missions” with defined outcomes, real consequences, and the assumption of success tends to produce dramatically higher follow-through, according to leaders at the April 2026 dinner.
Neither alone. The consensus from the April 2026 dinner was that AI deployment requires a co-ownership model where IT builds the technical capability while Operations defines the problems worth solving. IT-led deployments tend to produce technically functional systems disconnected from real problems; Operations-led deployments tend to fragment into incompatible point solutions. The recommended structure is a joint AI council with veto authority on both sides.
According to RAND Corporation’s 2025 meta-analysis of 65 enterprise AI initiatives, 80.3% of AI projects fail to deliver intended business value. Gartner’s April 2026 survey of 782 infrastructure and operations leaders found that 57% had experienced at least one AI project failure, and only 28% of AI infrastructure projects deliver promised returns. McKinsey’s November 2025 Global AI Survey found that while 88% of organizations use AI in some function, only 39% see measurable EBIT impact.
The April 30, 2026 closed-door dinner included fourteen leaders representing freight-tech startups (the largest contingent), enterprise practitioners from semiconductor, telecom, CPG, and dairy supply chains, a mid-market 3PL leadership team, a boutique 3PL connector, and institutional voices from defense logistics, sports operations and higher education. Retail, pharma, automotive, consumer electronics and fashion were not directly represented and the report’s findings should not be extrapolated to those sectors without independent validation.
This was the first of what we hope will be many of these conversations. If you were in the room, thank you for the honesty. If you weren’t, and you’ve recognized your own organization in any of the patterns above, we’d like to hear from you.
The full intelligence briefing produced from the evening — including the verbatim conversation highlights, the industry-specific outlook, and the operational frameworks — is available to attendees and to those joining the next convening.
Until then: discover before deploying, govern before scaling, and don’t let anyone in your organization mistake a pilot for a mission.
Methodology: This post synthesizes findings from a closed-door dinner of fourteen invited guests held on April 30, 2026. The conversation was recorded and transcribed (with consent) and analyzed across thematic patterns. Direct attributions to individuals or specific employers have been withheld to preserve the candor of the original exchange. External statistics have been cross-referenced against published research from Gartner, RAND Corporation, McKinsey, MIT NANDA Initiative, and S&P Global Market Intelligence.
About Zallpy: Zallpy is a software development and AI consulting firm working with enterprises across the Americas. We host these conversations because we believe the gap between AI ambition and operational reality is the defining business problem of the decade — and because closing it requires the kind of honest exchange that doesn’t happen in conference rooms.
Event partner: This convening was held in collaboration with Plug and Play Texas on April 30, 2026.
Citations and sources: – Ryseff, J., De Bruhl, B., & Newberry, S. The Root Causes of Failure for Artificial Intelligence Projects and How They Can Succeed, RAND Corporation, RR-A2680-1 (2024). Available at: https://www.rand.org/pubs/research_reports/RRA2680-1.html – Gartner. AI Projects in I&O Stall Ahead of Meaningful ROI Returns (April 7, 2026). Available at: https://www.gartner.com/en/newsroom/press-releases/2026-04-07-gartner-says-artificial-intelligence-projects-in-infrastructure-and-operations-stall-ahead-of-meaningful-roi-returns – Gartner. Lack of AI-Ready Data Puts AI Projects at Risk (February 26, 2025). Available at: https://www.gartner.com/en/newsroom/press-releases/2025-02-26-lack-of-ai-ready-data-puts-ai-projects-at-risk – McKinsey & Company. The State of AI: Global Survey (November 2025). Available at: https://www.mckinsey.com/capabilities/quantumblack/our-insights/the-state-of-ai – Challapally, A., Pease, C., Raskar, R., & Chari, P. The GenAI Divide: State of AI in Business 2025, MIT NANDA Initiative (July 2025). Reported by Fortune: https://fortune.com/2025/08/18/mit-report-95-percent-generative-ai-pilots-at-companies-failing-cfo/ – S&P Global Market Intelligence. Voice of the Enterprise: AI & Machine Learning, Use Cases 2025. Available at: https://www.spglobal.com/market-intelligence/en/news-insights/research/2025/10/generative-ai-shows-rapid-growth-but-yields-mixed-results
This post was published in May 2026 and reflects the state of supply chain AI adoption as observed at that time. Last updated: May 2026.
Verified AuthorA strategic technology consultant with experience supporting CTOs, CIOs, and engineering leaders in identifying and resolving execution constraints that impact delivery speed, stability, and organizational agility. At Zallpy, he works within complex enterprise environments diagnosing bottlenecks across architecture, integrations, data flows, and delivery processes, helping teams uncover root causes and implement practical, sustainable solutions. He collaborates closely with technology organizations to improve system reliability, strengthen integration resilience, and increase delivery visibility, while also focusing on knowledge transfer and capability building so teams can sustain improvements independently and scale with confidence.
A strategic technology consultant with experience supporting CTOs, CIOs, and engineering leaders in identifying and resolving execution constraints that impact delivery speed, stability, and organizational agility. At Zallpy, he works within complex enterprise environments diagnosing bottlenecks across architecture, integrations, data flows, and delivery processes, helping teams uncover root causes and implement practical, sustainable solutions. He collaborates closely with technology organizations to improve system reliability, strengthen integration resilience, and increase delivery visibility, while also focusing on knowledge transfer and capability building so teams can sustain improvements independently and scale with confidence.
Verified Author
Verified Author
Verified Author