Finding the right AI projects in the Enterprise

We are seeing business leaders around the world talk more and more on how AI is augmenting human expertise to improve business processes and customer experience. They are also increasingly experimenting with AI projects in their respective enterprises.

But it is also a well known fact that only around 10-15% of AI proof of concepts succeed. Business leaders who are just starting with AI are finding it really hard to pick the right projects. Below are three important criteria that can help identify the right projects and help them succeed in their AI transformation journey.

1)   Proprietary Data

 If we take a cue from how the most successful AI companies in the world behave, it is data and not algorithms that give them competitive advantage. You see these companies publishing papers about their algorithms, but never do you see them sharing their proprietary data, which actually gives them competitive advantage.

While architecting products, business leaders need to think a bit differently and instrument the products in such a way that helps them get the necessary data reserve. Needless to say, while you are collecting more and more information about your customers, it is also very important to be very respectful of their privacy.

If your enterprise is currently at a nascent AI stage, it is very important to pick projects and use-cases where the necessary proprietary data is available. It is also very important that you simultaneously keep building the proprietary data reserve that comes with the right kind of customer engagement with your product. Even if you are not using the data immediately, it is a resource that would make a huge difference later on.

2)   Machine Learning Infrastructure

You need machine learning infrastructure to develop and deliver AI projects. The infrastructure includes data preparation pipelines, model training, prediction, and monitoring.

Although, a lot of these ML operations are getting commoditized, if you are just starting with AI, it is important to pick your low hanging fruit projects. Projects that won’t require you to put in place the entire ML pipeline and work with only a subset. For example, you might find a pre-trained machine learning model on the internet, which means you won’t have to train the model yourself. You only have to prepare your dataset, and use the pre-trained machine learning model to make predictions on your dataset. You can simultaneously, incrementally keep on building your ML infrastructure.  

3)   Light Sandbox Environment

It is a good approach to start these projects as proof of concepts in a sandbox environment. We should look at AI projects where we can start at a super light PoC level, where we are just dumping our past data (after masking) into the sandbox environment. We can then run machine learning simulations on the data and spew out some results. We can compare the results of how things were before, and will be after this implementation. Based on which we can decide whether to take it to the next incremental sprint or not.

Of-course, so that it does not just remain as an interesting project, there has to be a real roadmap from the beginning on how we are going to PoC our way to scale and make way into business.