Why Generative AI Without Strategy Is Doomed to Fail.
Generative AI is everywhere. Boards ask for it. Teams try it. Budgets appear. The promise is simple: faster work, better decisions, new ways of operating.
But where are the real results? The conclusion is disappointing.
A large study by Humlum & Vestergaard (2025) followed 25,000 workers across 7,000 companies in Denmark. The study looked at 11 knowledge-work roles most exposed to chatbots for text-based tasks: marketing, journalism, HR, software, finance. Supply Chain wasn’t included. Still, the pattern is clear. Adoption is massive. Time savings are small (a few percent). There is no detectable effect on wages, hours, or jobs. Two years in, the hype is huge. The impact is small.
We see the same in our daily work. The Dutch newspaper Financieele Dagblad warned about “workslop”: AI output that looks polished but clogs up inboxes. It does not help work move forward.
This feels new, but we have seen it before. In the late 1990s, the dot-com boom promised that a web shop would change everything. Capital flooded in. Then reality arrived in the early 2000s. Winners needed logistics, supply chain, customer experience, and business models beyond a web shop. Many companies disappeared. A few with real foundations (Amazon is the classic example) kept building and later dominated.
Generative AI can follow the same path: early excitement, lagging results, rising costs, and then disappointment. Unless we change course now.
Why results stay small
A few things hold organizations back:
- No strategic anchor.Pilots start from “we should do something with AI,” not from clear business problems.
- No process redesign.Companies place AI on top of old workflows. The old workflow wins.
- Data and governance are not ready.If your data is scattered or unstructured, generative tools produce confident mistakes.
- Easy use cases dominate.Internal comms, quick drafts, demo chatbots. These are visible but not mission-critical.
- Wrong expectations.Generative AI hallucinates. It struggles with math unless guided. It always produces an answer, even when wrong.
Don’t confuse generative AI with AI in general
Machine learning, optimization, and advanced analytics already create measurable value. They improve forecasting, inventory, routing, and risk management. Generative AI is different. It is powerful with language and ideas but immature on its own. The real gains come when we combine generative AI with proven methods and, most of all, with experienced people.
What leaders should do now
1.Start from outcomes, not tools.
A VP Supply Chain tells her team, “We need a generative AI pilot.” The room nods. Nothing changes. A week later, she reopens the meeting with a different sentence: “We must cut Days-in-Inventory by 10% this year without hurting OTIF. Where can AI help?” The conversation shifts right away. Forecast bias appears. Slow exception handling appears. Long planning cycles appear. Real pain points surface. Now the team can frame use cases that matter: accelerate exception triage, reduce manual reconciliation, surface early demand signals per SKU family. The pilot has a scorecard, not just a demo video.
2.Put the data house in order before you automate the house.
Legal and procurement want a chatbot to summarize supplier contracts. Great idea until you open the files. You find scans, mixed languages, legacy templates, handwritten notes. The team pauses. They spend a sprint building a simple pipeline: OCR where needed, consistent fields (term, termination, penalties, SLAs), a controlled vocabulary for categories. Only then does the chatbot come in. The output goes from “charming but random” to “usable and safe.” Audit can trust what it reads.
3.Change the way the work flows, not just the tool on top.
Planners live in email and Excel. Every day ends the same: 60 new messages, many late, all urgent. Someone suggests a “planning assistant” bot. Instead of placing a bot into the email storm, the team redraws the flow. Forecast updates, supplier delays, and demand spikes land in one shared workspace. Exceptions are flagged with clear thresholds. The assistant proposes next actions in that space, not in long email threads. After two weeks, the midnight inbox shrinks. Decisions move earlier in the day. The tool helps not from its own power but from the process change.
4.Let each method do what it does best, then connect them.
A CFO wants one “AI” to fix forecasting and inventory. The team explains a different path. Machine learning builds the forecast and tracks accuracy by product family. Optimization models convert that forecast into target buffers and safety stock policies. Generative AI then produces executive-ready scenarios: plain-English trade-offs, stress tests (“freight +20%”), and short board briefs with the math linked behind each claim. The CFO sees the chain end-to-end: rigorous models where numbers matter, generative AI where communication and speed matter.
5.Keep a human in the loop, not as decoration but as a control.
Procurement receives an AI alert: a tier 2 supplier looks risky due to negative press in local media. The category lead checks the sources. One article is real. Two are re-posts of the same rumor. One is a paywalled piece taken out of context. She asks for a follow-up: translate the original piece, confirm the factory location, look up shipment history for the last 90 days. Within 24 hours the picture changes. Risk is still there but different in nature. A hasty switch would have created more risk than it solved. Trust grows not from AI perfection but from people treating it like a colleague who needs review.
These stories are simple on purpose. They show a pattern: strategy first, data ready, workflow redesigned, methods combined, humans accountable. When this order holds, generative AI stops being generic.
Why this matters for supply chain
Supply chains are complex systems. Small errors cascade. The attraction of generative AI is clear: write reports, draft playbooks, answer questions, summarize noise. On its own, generative AI is fragile. The best approach is to use generative AI as the translator and accelerator on top of robust ML and optimization, never as a replacement. That is where we see speed and adoption improve without losing rigor.
Building on our earlier perspective
We said this before in Beyond the AI Hype: A CXO’s Guide: AI alone is not enough. Value appears when you integrate technologies into the way decisions are made and when experienced people stay in control. This article extends that view and adds a warning: do not let generative AI replay the dot-com script inside your company.
Closing thought
Generative AI has real power. But promise without foundation is just illusion.
Leaders who know what generative AI can do, and more important, what it cannot, will avoid the generic trap. Start from outcomes. Fix the data. Redesign the flow. Combine methods. Keep experts in charge.
Do this, and the hype finally crosses the results line. Fail to do it, and we repeat a very old story.