Here's an overused but catchy phrase: Data is the new oil. While this phrase might not be entirely wrong, it is often misleading to a point that makes it dangerous. Simply put: Data in itself is not necessarily valuable, just as raw oil goes through a whole series of steps before it becomes fuel for cars or plastic for food containers. So in that sense, data is just the same.
Data alone has little value
In all fairness, it is clear that there are applications developed in a data-driven way that deliver value—for example, image data used in medical scenarios. The benefits of at least supporting diagnostics seem clear. In industrial settings, the use of data is often much less straightforward. The data collected is often just a by-product generated within specific processes. This data may or may not be suitable for a given use case. However, the quality of an AI system often depends on data being gathered under optimal and comparable conditions. This is especially true in scenarios where the amount of data is limited. This approach is often referred to as data-centric AI. The data used makes all the difference.
Managers and teams might fall for a fatal type of error: Instead of starting from a business case and then seeing if they can gather the necessary data with the quality needed, they start with data and desperately try to find a use case. These endeavors are time-consuming and expensive. If you have no goal in mind, how would you ever know your data is good enough?
Don't roll the dice; act with structure
Asking what you could do with your data is not the right way to start with AI. Instead, you need a structured approach starting from concrete use cases.
Start with use cases from your domain. This is where you are the expert with all the needed knowledge at hand. Discussing essential questions in workshops makes sense: What are really painful steps in your process? Which steps - if automated - could help minimize spending? What are common quality issues that drive costs? What are the requirements customers often confront you with? With these concrete questions, first specific use cases can be derived. As a next step, you should look at the data: What data is already available? Is there additional data that can be collected with a reasonable effort? You will figure out that minor adjustments can often be sufficient to enrich your data and make it useful for first real-world implementations.