Grupo DEACERO

As you may have noticed, in recent years everybody has been talking about 'analytics and data science.' Every company wants to experience the golden age of machine learning algorithms and fancy prescriptive projects. But here's the thing, even if you're not yet in the analytics enthusiast's tribe and only recently started working or approaching this area, I’m sure you’ve noticed that almost no one is ready to seize the opportunity. Everybody is expecting magic tricks or a ‘one fits all’ solution.

Spoiler and advice, this is not magic and for sure, a preconceived solution won’t exactly fit your company's needs by feeding your data 'as is.' Instead, it could lead you to spend a lot of money and get poor results that your board won't love, so here is the advice, think it twice before believing in everything your preferred vendor tells you.

Yes, if you’re reading this, I want you to know that you’re not alone, you’re not the only one who thinks it’s crazy to invest millions paying some vendor that assures you they'll use neural networks with your company's Excel files and expect they'll get some drastically different results from your logistic regression. It’s just not the way this works.

These high expectations are not entirely your manager’s or the vendor’s fault. The whole social media is talking about AI and the ‘generative pre-trained transformer’ (GPT) algorithm, as well as the super-skills that it appears to have in any field. But what you need to understand and explain if needed, is that the development of this kind of tool took lots of years, money, and effort from brilliant minds exclusively dedicated to the algorithm development.

I want to be clear; I’m not trying to discourage you from investing in these kinds of projects. I just want to help you realize that if you’re going to start a project of this nature, you’ll need to have your feet on the ground and set reachable goals according to your company’s maturity, capabilities, architecture, and most valuable opportunity areas. Each company has different characteristics regarding its digital environment, so they have different values to capture.

"I want you to know that you're not alone, you're not the only one who thinks it's crazy to invest millions paying some vendor that assures you they'll use neural networks with your company's Excel files and expect they'll get some drastically different results from your logistic regression."

 If you're working in a digital native company, I am sure that your project's path is going to be natural. You'll certainly have the architecture, capabilities, and structure to get the best of it in a fair period. However, if you’re working in a non-digital native company that's migrating to a digital environment or using systems designed more than 20 years ago, you’ll have to work harder in the main projects developing algorithms and ensure that you’re considering all the required pieces to get the job done.

Now, I want to share some useful tricks I would have loved to hear a few years ago; there are ways of getting things done with a defined structure and basic steps that can make your life easier. I've seen these steps work in different industries among my projects and for some colleagues too who are in this same work area. It’s not rocket science; the key is to maintain a methodical and detailed approach during the process.

Understand the value of capture; although some projects could sound awesome, you need to remember that you should always tackle the most valuable. Can this project save you money? FTEs? Or can it lead to generating more earnings than the others in your pipeline?

● Do your research & seek inspiration; benchmark what your competitors or other industries are doing, which processes they’re using, inputs, and pieces needed to develop their tools, algorithms, or projects. Unless you’re trying to discover new algorithms or methodologies, you don’t need to make it all, take sources of inspiration, and make educated decisions.

● your problem and define the expected outcome; before starting your project, ensure that you understand the problem, involved actors, and influences, as well as to have a defined expected outcome. If you don't follow this step, there will be a big risk of getting nothing.

● Keep it simple, make a Pareto, and keep the trash out.; I'm sure you've heard this, sometimes less is more, so evaluate efforts, resources, and results. Sometimes a logistic regression could give better results than an overcomplicated neural network and keep your life out of the stress of maintaining multiple databases that contribute marginally to an improved outcome.

● Take care of your data & create padlocks (automated if possible); select carefully what are you going to feed to your models and generate processes that ensure the quality of the data you're using. These processes can also be applied to your models. For example, if you already have some expected outcome values, set operation limits and alerts that react to them, this way, you'll be aware that something is happening if something goes wrong. 

● Ensure accountability; involve functional experts and make them own these projects. Most of the time, these kinds of projects are developed by service areas like "Business Intelligence", and, if there is no accountability from the functional owners, even with great outcomes or suggestions, your project can just pass by and become forgotten.