AI-Powered Business Intelligence

Chapter 2. From BI to Decision Intelligence: Assessing Feasibility for AI Projects

In the preceding chapter, you learned how ML capabilities could drive business impact. But to create a roadmap of prioritized use cases and make an informed decision about which use cases to pursue as a priority, we need to consider another dimension of criteria: feasibility.

This chapter dives much deeper into the fundamentals of ML to enable you to assess the complexity and overall feasibility of a given AI use case. We will explore feasibility based on three main topics: data, infrastructure/architecture, and ethics. As a result, you will be able to create the first version of your AI-powered BI use case roadmap.

Putting Data First

AI projects require a different mindset than classic BI projects. Most BI projects are done in a relatively straightforward manner, often following the traditional waterfall model: define the metric you want to show, design the data model, integrate the data, and make sure it works (which is often hard enough). Iterate if necessary. Job done.

The main difference in AI projects is that—even under ideal circumstances—the outcome is highly uncertain. We simply do not know whether an AI model will work with our data and be good enough to deliver value until we test it with real data.

For this reason, AI projects typically require multiple, shorter iteration cycles in an Agile-like project framework such as the cross-industry standard process for data mining (CRISP-DM), ...