In this paragraph “Be Business my friend”, I am readapting the Bruce Lee phrase: “Be water, my friend” from the business logic point of view. Why? Because this is the first approach that you have to manage if you want to be compliant in your projects.
I read a lot of about digital transformation, Chief Data Officer (CDO), data science, deep learning, machine learning, IoT, the blockchain, and so on. Every day there are a couple of new terms in the game: The last one that I heard is MLOps. It was months ago, during the DevOps Talks Conference in Melbourne that I came across this term. Sasa Savic, Co-Founder, Telstra HPSE coined this new term by explaining how combining machine learning and DevOps constitutes – MLOps – the road to building intelligent services.
In order to go ahead with a digital transformation project, you need to work very closely with the business. When I was learning data science basics, I saw that I can improve a lot of my machine learning models. Technically, we have a lot of possibilities to improve our models even while the accuracy or validation metrics are growing. But if you didn’t use the correct features, your predictions score won’t be valid.
Let’s consider the example of soccer. What would you do if you had to predict the number of users who will be connecting to your online platform? You will have to keep a track of all the historical data that you obtain from all the databases or data lake provided to you.
However, the data provided may contain a lot of unnecessary information. So, how does one create an environment to retrieve the correct data? This is where you will have to work with the business department and not only with the IT department to include new tools or extract the data from this physical repository to digital.
Let’s consider another scenario in which you want to build a machine learning model about the number of supporters who want to consume one football match through your streaming platform. What would be your primary requirement to complete this project? You’ll need to manage data. You’ll need attributes such as competition, participating teams, classification, name, and the number of players injured, date of the match, team rivalry, tradition, budget, stars, and historical data.
Many of these attributes comment above that describe the observation, are not included in the business environment. You have to obtain them before and share with the rest of your specifications in order to start building your best model. I am sure, with the new features, your entropy value will be lower than before, and this will improve your model. In the next steps, you can clean data, standardize, normalize, discrete your data. You can apply technics such as dimension reduction Principal Component Analysis (PCA) and cross-validation.
Let’s consider another example. What happens if you are Muslim and during the holy month of Ramadan you receive an email from your favorite restaurant with a recommendation for lunch? Obviously, the model mentioned above stands invalid. One needs to spend a lot of time knowing business rules and applying them to their respective businesses. In simple terms, if you fail to bring the best version, your customers choose to discontinue you.
One has to understand the business rules to speak the same language that the customer speaks. Incorporating these rules can provide businesses with more expertise in retaining their customers as well as acquiring new ones.
I love data and digital transformation despite issues such as silos and resistance to change. However, when you believe in your own capacity and your alignment with your company and customers, you have a lot of possibilities to be successful.