In recent years, AI has become such a hyped-up concept that all and their housecats want to use artificial intelligence. Moreover, it has become the silver bullet answer at the bottom bullet point of the slide deck. I get it. ChatGPT showed the amazing capabilities of large language models, and I am astonished almost daily by the capabilities and use cases. However, I want to discuss how we approach problems in business and how we can work smarter and use machine learning as a highly efficient tool when it is needed and not as a Swiss Army silver knife bullet.

The sad reality of ML in non-tech

Engineers working in tech and software sometimes get captured by the extraordinary speed of the technology. When ChatGPT was launched, thousands of startups rose to take advantage of the development and progress made. Furthermore, all the visionaries (dreamers) within the existing old-fashioned companies rushed to create new slides to present to management. However, regardless of the development of the technologies, machine learning projects face a significant challenge within those run-of-the-mill banks, logistic companies, and manufacturers.

It is about trusting the process.

Those old-fashioned companies have survived and thrived for a reason: hugely efficient, finely tuned systems with standardized systems. Those processes work fine, and much work is put into making those systems perform. Those processes are based on deterministic logic that has been iterated on and works well enough. In my favorite series of all time, The Wire, there is a saying on focusing on the money trail instead of the drug trail. "If you follow the drugs, you get drug addicts and drug dealers; if you follow the money, there is no way of knowing where you will end up." This analogy also works well for machine learning, which is probabilistic in its inherent nature. If you expect a deterministic, predictable, clearly explainable output, then using machine learning is a recipe for failure.

Use Math Stupid

The advantage of using mathematical thinking and business logic is that it is deterministic. You can explain why every input gets an output. Logic can usually be easily integrated within the systems and does not require an advanced setup with training or feature engineering. Sure, you won't get as good results as machine learning, but it is better than nothing, and it can be implemented quickly.

All machine learning projects should first be tried to be fixed with conventional logic; then, you should consider using more advanced tools. Remember, from systems thinking, all complex systems that work are based on simple systems that work, but no system that works was complex at the start.

You most likely don't have enough quality data.

Every company believes they have a lot of data, and that there is value just waiting to be had if you can utilize it. The sad reality is that the data available is not that valuable. It is old, scattered, and on so many different formats that getting value from it is not worth the effort. This is something that will need to be addressed before starting a journey towards using machine learning. Modern enterprise data warehouses are powerful tools for visualizing and storing data and leveraging it. Embarking on an endeavor to use data warehouses is also a fantastic opportunity to enhance the quality of your data models and ensure that even if you do not have the data foundation to do machine learning yet, you can set yourself up to have it in the near future. Awareness of the opportunities and taking measures to enable them will separate data-driven versus non-data-driven ventures.

First things first

To recap, there are many opportunities to exploit the data within your organization. You need to understand where the opportunities lie within your company and be aware that using simpler methods is the best approach.

Awareness is key; understanding where you are as an organization on the data plane is paramount to understanding what needs to be done and the key to a successful machine learning program.