Innovator's Dilemma

The Innovator’s Dilemma is an explanation for why large and well-run companies get overrun by new entrants. Harvard economist Clayton Christensen described it in 1997 in a book of this title. His observation: these companies don’t fail because they make mistakes, but because they do everything right. They ask their most important customers what they need, and they improve their product exactly in that direction. A new technology that is initially worse and cheaper seems uninteresting in this context. Until it becomes good enough – and by then it’s too late.

Why good management becomes a trap here

The surprising core of the idea is that stupidity plays no role. A manager who puts money into a small, unprofitable niche instead of the business with paying regular customers is acting irrationally in the short term. He would have to explain to his superiors and shareholders why he is building a product with a low profit margin. Such proposals die almost automatically in the decision-making process at large companies.

That’s why the dilemma isn’t just a business history lesson, but a warning. It says: success itself creates blind spots. Whoever has a lot to lose defends it – and in doing so may be defending the wrong thing. Christensen distinguishes between two types of innovation here. Sustaining innovations make an existing product better, such as a sharper phone camera. Disruptive innovations attack from below, with a simpler and cheaper offering for customers who were previously not served at all.

A common mistake is to call every major change a disruption. Christensen himself protested against this. An expensive premium product that upends the market is not a disruption by his definition, but rather a very good attack from above.

The typical course of displacement

It starts with a product that is objectively weaker. The first digital cameras produced worse images than film cameras. Customers who valued quality rejected them, and established manufacturers took this as proof that the technology wasn’t to be taken seriously. Initially, it was sold to people for whom cheap and practical mattered more than perfect.

The crucial second step is this: the new technology improves faster than the market demands improvement. At some point it crosses the threshold at which it becomes sufficient for the majority of customers. From this point on, the market tips very quickly, because the new product is also cheaper or more convenient. The old provider can no longer catch up, because its factories, contracts, and cost structures are suited to the old world.

Christensen’s practical advice is therefore: found a separate, small unit that operates independently of the parent company. It should be allowed to be satisfied with small revenues and to compete with the main business. Only this way can one escape the compulsion to measure every idea against the standards of the current business.

The dilemma in the AI industry

In business news, the term currently appears mainly around search engines and chat programs. Google makes its money from advertising alongside search results. An assistant that formulates the answer directly saves users clicks – and thereby potentially costs exactly this advertising revenue. In addition, every generated answer is more expensive to compute than a classic list of results. A start-up without an advertising business doesn’t have this problem.

Similar patterns can be seen with chips and with software. Small, openly available language models are weaker than the expensive top-tier models. But for many tasks, such as summarizing emails, they are good enough and significantly cheaper. This is exactly the starting situation from which, according to Christensen, displacement arises.

Anyone reading company news can use the term as a test question: Is a company under pressure because it has to protect a business that is currently its best? Newspapers, music labels, and car manufacturers have also been described this way. The prediction is never certain – not every attacker from below prevails.