Table of Contents
In December 2001, the Segway was unveiled on US television to an audience primed for a revolution. Its backers predicted 10,000 units a week within a year and talked openly about redesigning cities around a two-wheeled scooter. Over its entire commercial life the company sold somewhere around 140,000 units. The engineering was genuinely impressive: hundreds of sensor readings per second, balanced by control loops that reacted faster than a human could fall over. None of it mattered.
That gap between technical brilliance and market success gets picked apart endlessly in post-mortems, and most of them miss the point. Products rarely die because the technology didn’t work. They die for reasons that are duller, more predictable, and far harder to engineer around. The reasons why some promising technologies fail tend to repeat across decades and industries, which is what makes them worth studying.
The Problem Was Real. The Purchase Wasn’t.
Juicero raised roughly $120 million for a $699 countertop press that squeezed single-serving juice packs. The machine was beautifully made, with a QR scanner that checked the pack hadn’t expired. Then journalists discovered you could squeeze the same packs by hand and get nearly identical juice. The press wasn’t solving a hard problem, it was solving a problem the customer had already solved with their fingers and a kitchen counter.
Google Glass ran into a different version of the same wall. The hardware worked. The problem was that wearing one in public made other people uncomfortable, and the people wearing it weren’t sure what they were supposed to do with it either. A technology can be functional and still answer a question nobody was asking out loud.
There’s a test that cuts through most of this: describe the moment the customer reaches for the thing. Not the feature list, the moment. If you can’t picture the moment in specific detail, the market probably can’t either.
Timing: Being Early Looks Exactly Like Being Wrong
AT&T introduced the PicturePhone at the 1964 World’s Fair. Callers could see each other across a dedicated link between New York and Washington. It failed, badly, because nobody wanted to sit in a booth, pay a premium, and be seen mid-sentence with no warning. Fifty years later the same idea arrived in everyone’s pocket and won immediately, because the surrounding pieces finally existed.
Webvan launched in 1996, raised around $800 million, built automated warehouses, and filed for bankruptcy in 2001. Online grocery delivery then sat dormant for the better part of two decades before becoming routine. The idea was never wrong. The roads, the phones, the payment rails and the customer habits hadn’t caught up.
Quantum computing is sitting in a similar spot right now. The physics is settled and genuinely strange, the way the rules that govern quantum systems bend intuition at every turn, but error correction at scale remains an open engineering problem rather than a finished one. That doesn’t make the investment foolish. It makes the timeline the thing you’re actually betting on.
The Last Mile Rarely Belongs to the Inventor
A technology needs a support system to be useful, and inventors almost never own it. Hydrogen fuel-cell cars have driven well for years. The buying decision collapses the moment you look at the map and count the refuelling stations. Concorde was a technical triumph that burned fuel too fast to price a ticket that could cover costs.
The gap between a working prototype and a product people buy is usually filled by logistics, distribution, service networks, standards bodies and finance rather than by cleverer engineering. Understand how new technologies move from research to market and you start to see that the hard work happens after the demo, not before it.
Unit Economics Have a Way of Making Decisions for You
Solyndra built a genuinely clever cylindrical solar panel. It took a $535 million federal loan guarantee, then went bankrupt in 2011 when crystalline silicon prices collapsed. The technology worked; the cost curve it was racing against didn’t wait for it. MoviePass hit a similar wall from the opposite direction, selling tickets at a monthly price below the cost of the tickets, and reported losses north of $300 million in 2018 before the whole thing fell apart.
Before the money runs out, most of these failures were visible on a spreadsheet. A few questions tend to expose them early:
- Does the cost per unit fall meaningfully as volume rises, or are you just buying revenue?
- What is the true cost of acquisition once the novelty wears off and paid marketing begins?
- Which single supplier, price or regulation, if it moves, breaks the whole model?
- Does the first version create a compounding advantage, or does each sale start from zero?
Building for scalability from the first release is unglamorous work, but the businesses that skip it tend to discover the omission at the worst possible moment, right when demand finally shows up.
Lock-In, Standards and the Cost of Asking People to Switch
Betamax is the classic case, and it’s not the one people remember. Sony’s format arguably produced a better picture, but the first VHS decks recorded two hours where Betamax managed one, which meant you could tape a whole film without swapping cassettes. Sony also kept tight control of licensing while JVC signed up everyone. Better lost to more convenient, and the same story repeats with QWERTY keyboards, AC versus DC, and every app store you’ve ever been locked into.
Asking people to switch means asking them to abandon sunk costs, relearn habits, and accept that their existing files, colleagues and equipment may not come along. Products that win usually find a bridge across that gap instead of demanding a leap.
Hidden Risks That Turn Into Headlines
Not every failure is commercial. Some technologies succeed in the market and fail more slowly, in ways nobody priced in. Staff wiring unapproved AI tools into company systems looked like harmless efficiency until security teams realised what was flowing out through them, a pattern explored in this look at shadow AI and the security breaches it creates. By the time the risk is visible, the tooling is already embedded.
Materials follow the same arc. Few twentieth-century inventions looked more useful than cheap synthetic polymers, and the century-long rise of microplastics is what happens when nobody asks where a product goes after it’s thrown away. Governance and lifecycle thinking are part of the technology, not paperwork bolted on afterwards.
What the Survivors Actually Do Differently
The survivors aren’t smarter or luckier in any consistent way. They tend to do three unglamorous things well. They ship something modest that works and learns from real users, rather than perfecting in private. They design for the system around them, whether that’s charging networks, app stores or regulators. And they keep the cost of each new customer below what that customer is worth, even when growth looks slower than a competitor’s.
The Prius is a useful example. It wasn’t the first hybrid, wasn’t fast, and wasn’t beautiful. It was reliable, it needed no new infrastructure, and it got cheaper to build every year. Twenty-five years later the name is a category. That’s a much less romantic story than a launch on morning television, and it’s the one that keeps working.
