Moats and AI
How will the rise of AI capability affect the ability of startup companies to build moats? What moats are defensible in an AI-enabled world?
I believe that at least three types of moats remain defensible: Brand, Network Effects, and Physical Integration. Beyond that it’s very unclear to me. Why do these three make the cut?
Brand: A Trust Moat #
A strong, well-respected brand allows people to economize on information related to quality of product or service. They know how the burger will taste, or that they can find the supplies they need, or that they will receive good advice. To the extent that people continue to make buying decisions, a trusted brand will differentiate a company in the market.
Perhaps decisions made by AI and not people will also integrate brand as a discriminator. Or perhaps decisions by AI will ignore brand and rely on other gauges. In the first case brand still matters. If other criteria drive the AI decision, then I would expect patterns of judgement to emerge that act as the equivalent of brand reputation, so that even of the basis of brand awareness has shifted, the phenomenon will remain.
Before people allow AI tools to make such decisions they must enjoy a high degree of trust. I suspect that will take a long time to develop. It also introduces the same question of brand quality at the AI layer.
I think that the economy of information represented in a brand will remain useful regardless of where the decisions are made, making it a defensible moat.
Network Effects: A Social Moat #
Network effects arise when the value a user gains from a system increases as more people use it. Switching away from such a system means abandoning the network. Smaller substitute networks are unlikely to deliver as much value, so the opportunity cost is high.
Social networks offer a case in point. Many people have tried to implement Twitter replacements, but none has overcome the Twitter network effects. Competition in network-effects arenas tend to produce winner-take-all outcomes.
The ability of AI to support fast-follow workalike capability may make it much more difficult to win these competitions. If your application demonstrates the viable network effects in an industry and I can quickly create a work-alike application, then I may overcome your first-mover advantage. I expect to see an increasingly competitive scramble for dominance in network-effects arenas, followed by winner take all. You will have a defensible position if you can get there, but the path there will be more difficult than ever before.
Physical Integration: A Non-Digital Moat #
Companies that integrate AI with the physical world – think construction, manufacturing, or hardware-dependent services – work with both digital and phyical constraints. Physical world interation requires robotics or people to carry out work. It may take place in environments subject to weather or other variance. These complexities will slow any AI-driven displacement.
Services that exist to provide a human touch are another version of this. Some examples are being a good waiter, pet sitter, or escort.
What Missed The Cut? #
A friend suggested that User Experience (UX) belongs on my list. I am not persuaded. His contention was that UX depends on deep understanding of the user, which the AI tools will not be able to generate, and that workalike systems will bear only superficial resemblence to great UX because they will lack the depth of understanding.
If the key difficulty undergirding UX is depth of understanding of the market, then I expect to see firms arise whose comparative advantage lies in gaining that very insight and then applying it with fast-follow builds. A firm like Rocket Internet already operates with a fast-follow business model.
What Did I Miss? #
Should something else make the cut? Am I off the mark with the three that I’ve chosen? What say you?