Sometimes you get a very clear feeling of déjà vu.
These days, almost every executive meeting ends up in the same place: AI. And in most cases, people actually mean one thing: Generative AI. Chatbots. Copilots. Content generation. The urgency is real. The fear of missing out is real. And somewhere in the room there is always that unspoken sentence: “If we don’t do this now, we’ll be left behind.”
I’ve heard that sentence before.
In 2000, the buzzword wasn’t AI. It was e-commerce. I was working at a global company, a traditional building materials business, right when the dot-com wave hit industry. Investors and analysts started asking: “What is your e-business strategy?” ExCom did not have a convincing answer. And under pressure, the company launched a global e-commerce initiative. It was fast, big, and very visible.
That’s the first parallel. Hype cycles create external pressure. External pressure creates programs before the organization really knows what it wants to achieve. And a lot of consultants who are happy to further fuel the hype.
And that leads to the biggest mistake, then and now. Starting with the technology instead of starting with the customer.
There is a second mistake too, and it matters just as much. You can create a great customer experience that is simply not financially sustainable. If your service promise isn’t funded by a viable cost structure and healthy working capital, it breaks the moment the business gets stressed. In the end, supply chain must do two things at the same time: serve and fund. Customer value and economics are linked, whether we like it or not.
How I got pulled into e-commerce, and why it matters today
The story is almost absurd, but it’s true.
A corporate project manager was appointed. He came from finance, not from commerce or IT. He searched internally for anyone with e-commerce exposure. In my HR file he found a note saying that earlier in my career I had implemented SAP Internet Transaction Server twice (one of the first SAP e-commerce solutions). He flew to the Netherlands and asked our local CEO to onboard me into the global program.
The CEO pushed back. “He’s my IT director. I just hired him. I’m paying for him.” The project manager replied:
“Money is not a problem. I have €28 million to burn this year.”
If you’ve lived through a hype cycle, you recognize the pattern. The budget exists. The pressure exists. The story needs to exist. But the fundamentals like customer value, process, data, and economics are often vague. Should you experiment anyway? Absolutely. That’s how you learn. The mistake is not experimentation, it’s jumping from experiment to enterprise-wide rollout without understanding what actually worked and why.
Two years later the program was scaled back dramatically, and not because the technology failed. The technology worked fine. It failed because we never clearly defined which customer problem we were solving. We had built a solution looking for a problem, instead of the other way around.
That is why I’m cautious when I see companies today treat GenAI pilots as an AI strategy. A pilot is not a strategy. It’s a starting point, at best.
The dot-com lesson most leaders forget: integration and economics always win
During dot-com, many companies believed a digital front-end would magically improve the business. B2B marketplaces were the perfect example. One portal, less friction, transparency, automated procurement. On paper it looked undisputed.
Reality was less romantic.
The technology was never the hardest part. The hard parts were governance, data definitions, integration with back-end systems, and basic economic logic. A marketplace wasn’t valuable because it was digital. It was valuable only if it solved a customer problem better than existing channels, while also being economically sustainable.
Here is what killed many dot-com B2B initiatives. Not the internet itself, but the operating reality underneath. Systems didn’t integrate end-to-end, so the promised ROI never appeared. Governance was complex and slow because too many parties owned too many decisions. And relationships mattered more than price transparency, especially in B2B.
The technology was visible. The foundations were not. And in the end, the foundations decided the outcome.
That lesson applies directly to AI.
A clever model is not a strategy. And GenAI is not “AI in general.” It’s one powerful category. Highly visible, easy to demo, and easy to overuse.
Don’t confuse AI with GenAI: the EDI parallel
One thing I find striking today is how narrow the AI conversation has become. When many executives say “AI,” they really mean Generative AI.
But AI is bigger. Especially in supply chain and operations, a large part of what we call AI has existed for decades. Machine learning for forecasting and anomaly detection. Optimization and simulation. Operations research algorithms, many known since the 1980s.
In the past, we often didn’t scale these approaches. Not because the math was missing, but because compute power and infrastructure were expense or simply not available. With cloud computing and modern platforms, that constraint has changed. Many techniques that were theoretically possible become operationally feasable now.
I like to call this non-generative AI. It’s AI that improves decisions and execution, where generating content isn’t the main output.
This is where the dot-com era offers a useful analogy.
In the early 2000s, many companies thought “digital” meant websites and marketplaces. But in B2B, customers often wanted something else. Reliability, speed, accuracy, and integration. In many cases, the most valuable digital solution wasn’t a flashy portal at all. It was EDI.
EDI wasn’t sexy. It didn’t make headlines. But it connected systems, reduced manual work, reduced errors, and improved service. It also improved the economics. Fewer errors and fewer manual steps reduce rework and expediting. Better reliability reduces the need for buffers. Over time that improves cost and working capital.
Today, GenAI is the flashy website. Non-generative AI is the EDI. Less visible, but deeply integrated, and often much closer to measurable value.
One warning, because it matters:
Smooth output is not the same as correct output.
GenAI can be very useful, but it needs guardrails. The confidence of the output often exceeds its accuracy. That’s a dangerous combination if you don’t build in checks.
And to be fair, AI is not dot-com copy-paste. Some players are really profitable, just not the ones building the models. NVIDIA sells shovels and makes billions. OpenAI digs for gold and loses billions. The technology is better than in 2000. The mistakes organizations make adopting it? Exactly the same.
Five dimensions, not three
Next, I spent fifteen years at Medtronic leading supply chain transformations. One pattern kept coming back. When transformations failed, the root cause was almost never the technology. It was everything around it.
Most people know the classic framework: People, Process, Technology. It’s useful, but incomplete. Over the years I’ve added two dimensions that are just as critical: Data and Policy.
Without clean data, even the smartest AI model produces garbage. And without clear policies regarding decision rights, escalation rules, stewardship etc. changes don’t stick. People go back to doing what they always did.
So when I look at any transformation now, I check five things: People, Process, Technology, Data, and Policy. If one of them is weak, the whole thing is at risk.
This sounds obvious when you write it down. But in practice, I still see companies spend 80% of their attention on the technology and wonder why adoption fails.
The question that separates winners from losers
If you want AI to create differentiation, the boardroom question should not be: “What is our GenAI strategy?”
A better question is: ‘Where does the customer feel pain today, and what is the service, cost-to-serve, and working-capital impact of that pain?’ This isn’t just a question for sales or customer service. Planning, procurement, logistics, finance, basically every function should be able to answer how their work ultimately serves the customer. If that link is unclear, adding AI won’t make it clearer.
Real differentiation lives in outcomes customers feel. Stable availability. Predictable lead times. Fast and clear exception handling. Reliable promise dates.
To be sustainable, those outcomes must be funded. Customer value that relies on high cost or excess inventory will break under pressure.
This is the customer, cost, and cash triangle. Service, cost-to-serve, and working capital must move together.
A case from the field: certainty beats inventory
Let me make this concrete with a transformation I’ve seen in a regulated, service-critical environment.
The business had a very inventory-driven service model. The logic was simple: “If we hold inventory close to the customer, we can serve fast.”
Over time, it became unsustainable.
Working capital kept growing because inventory was used as insurance for everything. Demand volatility. Planning weaknesses. Slow processes. Unclear decision rights. Lack of trust in information. At the same time, service levels were only “okay”. Not terrible, but not great either. The company was paying the cost of a premium service model without offering customers a premium service in return.
That’s a dangerous place to be. You get the worst of both worlds.
The breakthrough came when we stopped debating inventory levels and started with a different question.
What do customers truly need?
The answer wasn’t “maximum inventory.” It was certainty. Certainty that products would be available at the moment they truly needed them. Not earlier. Not “somewhere.” Not “we hope so.” But reliably, predictably, and consistently.
Once you anchor the service promise on certainty, the whole operating model changes.
You move from “inventory availability” thinking to “promise reliability” thinking. That may feel like semantics, but it really matters. You redesign planning and replenishment rules around the moment of true need. You define decision rights and escalation logic for exceptions. You build transparency so teams trust the signals. And you stop using inventory as the universal solution for every problem.
This is where non-generative AI became powerful.
Instead of building an AI model that simply reacts to inventory availability, we introduced a model built around the service promise of certainty. The AI supported better decisions about supply allocation and timing, based on customer need moments, constraints, and trade-offs. It wasn’t about generating text or pretty dashboards. It was about making the underlying decision logic smarter, faster, and more consistent.
The outcome was exactly what you want from transformation. Customer satisfaction improved because service became more reliable and predictable. Return on investment improved because working capital came down and operations became more efficient.
I saw the same pattern at Medtronic. Our digital twin delivered $55 million in savings. Not because the technology was clever, but because we had first defined clearly what we wanted to improve. The technology was the enabler. The clarity about purpose was the foundation.
This is the pattern I see again and again. The best AI value comes when AI is embedded into a customer-driven operating model, not the other way around.
Transformation is a journey, not a step. AI becomes real only when it is embedded into how decisions are made, owned, and executed.
Three questions for your next AI discussion
If you want AI to be more than a headline, the path is not complicated. But it is demanding.
Before your next AI discussion, consider these three questions:
- Which specific customer problem are we solving, and how will we measure success? If you can’t answer this in one sentence, you’re not ready to start.
- Is our data quality good enough to support this? Bad data in means bad decisions out. No algorithm fixes that.
- Who owns the decision that AI is supposed to improve? If ownership is unclear, adoption will fail. Technology doesn’t replace accountability.
And be honest about the hard part. Most AI programs don’t fail because the model is bad. They fail because the organization doesn’t change.
What dot-com made painfully clear: new technology doesn’t cover up old problems. It puts a spotlight on them.
The bottom line
The dot-com bubble burst, but the internet didn’t disappear. It became the backbone of the modern economy. The winners weren’t the companies with the flashiest websites. They were the ones that integrated technology into a solid business model and operational excellence.
The same will happen with AI.
The hype will calm down. The polished demos with little substance will fade. What remains are organizations that did the hard work. Clean data. Robust processes. Clear governance. A service model that customers truly value — and that the business can actually fund.
So yes, embrace AI. But start with the customer. Make sure your service promise is sustainable. And remember that the foundations matter more than the technology.
Because in the end, technology is not the differentiator.
What you deliver to customers is.
Sikko Zoer is Senior Partner at Qwinn Business Partners.