This is an excerpt from the upcoming Designed Analytics Report: Avoiding The Third AI Winter. The report will publish on 07/08.

———————————————————————————————–

I’m differentiating deep learning and generative AI in a specific way here. At the core, generative AI is a form of deep learning. However, deep learning can be used for generative AI tasks as well as to address non-generative business challenges. Hence, whenever I use the term deep learning in this report, it is the application of deep neural networks for non-generative AI tasks. I’m using this approach so that we can demarcate such applications from generative AI applications.

One strategy to avoid the third AI winter is to understand how existing machine learning algorithms tie with AI approaches like deep learning and can then be coupled with generative AI methodologies to build commercially attractive solutions. As discussed before, one of the key reasons behind AI’s popularity is that machine learning models have been able to deliver. This means that machine learning practitioners generally have a good general understanding of how algorithms can be leveraged to address problems within the industry.

Remember that the driver behind the failure of machine learning or advanced analytics-related initiatives does not pertain to the algorithm itself. Any solution, however simple it may seem, is generally driven through a combination of people processes and technology. When an analytics project is successfully implemented, the successes go beyond that solution. The success percolates into other aspects in the form of learnings the organization has extracted through that project.

So let us take a step back and understand that when an analytics project is executed, what are the different forms of learning or capabilities that an organization develops in addition to the solution itself.

Let us say that you just implemented a solution that allows you to understand which of your existing customers is a better fit for specific marketing campaigns. Let us say you are a financial services provider in a few years. Your campaigns are generally geared towards selling additional financial services products. Let us say you leveraged a machine learning algorithm to define customer segments that will respond better to specific campaigns based on specific campaign attributes.

In the entire exercise of developing this kind of solution, the least challenging part, in my opinion, is at least in this specific use case where these algorithms have been leveraged frequently, is the algorithm development. The more important learnings come from many different perspectives. Let us start with the data side first. When you develop any form of analytic solution, you obviously start by understanding your data landscape.

The more complex the analytics approaches that will be leveraged, the more complex generally is the data requirement. For running some SQL transactions or SQL-based analytics, you might be able to tap directly into a database that already exists. That may not be the case when you want to perform something more advanced, like a machine learning-based algorithm.

So you may need more data points and more attributes. The good news is that many of these data points or a majority of them may have been documented during your ML projects and can also be leveraged when you are actually planning to leverage more advanced AI algorithms like these learning algorithms.

The next example category is implementation bottleneck.