Are Most AI Startups Doomed to Failure?

Artificial intelligence is one of the most hyped and promising fields of the 21st century. I'm pumped about the potential it holds and think we've barely begun to explore it. However, it feels like there's a rush happening, similar to a gold rush, and not everyone will strike it rich.

AI has the potential to transform various industries and domains, from healthcare to education, from entertainment to finance. That is unquestionably exciting.

But we know that not all AI startups are destined to succeed. In fact, most of them are doomed because they lack defensibility and differentiation.

According to There's An AI For That, there are 12,208 AIs designed for tasks from creating music from images, to startup advice. They can't all be helpful, right? Just how many of these AIs will still be around next year?

It all comes down to one factor: solving problems for users.

The Binding of APIs

One of the main reasons why most AI startups are doomed is that they simply glue together AI APIs and create UIs.

These startups do not have any proprietary technology or data that gives them a competitive edge. They rely on existing AI services and platforms, such as Google Cloud, Microsoft Azure, IBM Watson, or Amazon Web Services, to provide the core functionality of their products.

They then add some user interface (UI) elements to make their products more appealing and accessible to the customers. Is this approach sustainable? For the few, perhaps, but what happens if Google decide to drop their current iteration of AI for another one, or Open AI increase the price for every API call?

It's a risky, short-term strategy that won't work out for everyone.

Fundamentally, these startups have to pay a high cost for using the AI APIs, which reduces their profit margin and scalability. Second, they have no control over the quality and reliability of the AI APIs, which may change or degrade over time. Third, these startups have no defensibility against the competition, as anyone can use the same AI APIs and create similar or better UIs.

Another reason why most AI startups are doomed is that even if they have a better UI, competitors can easily copy it.

A Good User Interface Is Not Enough

UI is not a strong source of differentiation or defensibility in the AI industry. UI is a superficial layer that can be replicated or improved by others. For example, suppose a startup creates a chatbot that uses a natural language processing (NLP) API to answer customer queries.

The startup may have a sleek and intuitive UI that makes the chatbot more engaging and user-friendly. It looks wonderful, and the small amount of users on board are full of praise.

The praise is short lasting, however, as this UI advantage can be quickly eroded by other startups or incumbents that can use the same or a different NLP API and create a similar or better UI.

What does this mean? Simply put, UI alone is not enough to ensure the success of an AI startup.

The same logic applies to the underlying technology of AI models like ChatGPT.

From ChatGPT to T5 and Beyond

ChatGPT is a popular and powerful NLP model that can generate coherent and fluent text based on a given prompt. It is based on the GPT-3 model, which is developed by OpenAI, a research organization that aims to create and promote beneficial AI. ChatGPT is used by many AI startups and products, such as Copilot, Replika, and Jarvis.

However, these models have no real moat and can be replicated by any large internet company. For instance, Google, Microsoft, and Facebook have their own versions of large-scale NLP models, such as T5, Turing-NLG, and Blender, which can perform similar or better tasks than ChatGPT.

These models are constantly evolving and improving, as the AI research community and industry are pushing the boundaries on a daily basis.

Building the best version of an AI model is also not sustainable, because the technological frontier of the AI industry is constantly moving.

So What Can Founders Do?

The only way to create lasting value in AI is to have continuous innovation.

This means that an AI startup must have a clear vision, a unique value proposition, a strong team, and a robust research and development (R&D) capability. An AI startup must be able to identify and solve real-world problems that are not addressed by existing solutions.

Have an honest look at the majority of AI startups out there and ask yourself, "Are they really solving real world problems?"

An AI startup must be able to create and leverage proprietary data and technology that give them a competitive edge. They must also be able to attract and retain talented and passionate people who can drive the innovation process and conduct and publish cutting-edge research that advances the state-of-the-art of the field.

These are the things that make an AI startup stand out from competitors and guarantee its success in the long run.

Do most AI startups face failure? Probably. Many believe that the overall startup failure rate is around 80%, so it's reasonable to assume that AI startups follow a similar pattern.

Unless there are ways in which these startups can show defensibility and differentiation, users will have little need to interact with them. Then consider that they rely on existing AI APIs and platforms, which are costly, unreliable, and vulnerable to competition.

While a UI might look slick, a UI is merely an interface to a system that might not solve any pertinent problems. Focusing on UI is a superficial and easily replicable layer.

The only way to succeed in AI is to have continuous innovation, which requires a clear vision, a unique value proposition, a strong team, and a robust R&D capability.

If you aim to thrive with your AI startup, here are the key elements that generate long-term value in the field of AI.

‍