What Leads to AI Success?

The bubble has burst! Last week, an MIT report was released that states 95% of AI projects have “failed”. This has sent investors into a panic. But before you dump your NVDA portfolio let me give you some insight into what I see is happening here.

It all starts with Data Science. The key term here is “science”. We start with a hypothesis, perform experiments and form a conclusion. It has been commonly observed that 80% of data science projects “fail”.

Enter GenAI. Suddenly nobody wants Data Scientists anymore, it’s all about AI Engineers (I’ll give you one guess who these AI Engineers are). Now, the term “science” has been dropped, surely that means no more time wasting with hypothesis testing. Wrong. GenAI is a more immature field, meaning even more uncertainty in success, hence the 95% “failure” rate.

An interaction I had when I was in banking, gives a good insight into what “failure” means in this context. Our team had just decided not to embark on an ambitious project because of the risk of “failure”. I said to my line manager “even if this does not work, gaining that knowledge is not failure” and he replied, “I agree, but I can’t put ‘learning’ into our financial reports”.

This brings me to my next point. What is the best way to guarantee AI success? How do we combine the best of science with the fast paced, profit and loss driven world of business? From my observations it’s the following:

Data: Non-negotiable, the foundation of all things AI.

Talent: You can’t just rely on best practises here; you need intuition and knowing when/how to push back when bad ideas are brought to you.

Domain Knowledge: Most people sleep on this one. In an AI team, you need someone who has lived the pain point they’re trying to solve. Why do you think GenAI is so good at coding and math? The people building it are domain experts.

Passion: There is no simple answer to this. I’ve seen teams with all the above without passion. It’s sad to see the waste. What I will say is: Make sure management doesn’t insult your AI team’s intelligence.

I think it’s no coincidence that these points align closely with startup success, and that AI startup implementations were reported to see a success rate of over 60% in enterprise. As the winds change, expect to see a slow reshaping of AI teams as AI and Data literacy in business increases. In the meantime, the strong AI startups will become more appealing and more widely adopted due to their ability to effectively optimise the above four areas.

25 August 2025