A data science project is always about improving the Business process. The improvement could be anything – increase in revenue, lag time elimination, customer retention, etc. Your data science project would always have the improvement as the primary KPI you would need to achieve.
The problem is, a lot of data science education I look upon is inadequate in addressing the business-understanding part. Most of the time, it is an afterthought where the students are directed to focus on the programming and machine learning parts. It’s a shame, because business understanding techniques would undoubtedly help students secure data employment more easily.
I am not talking about understanding the whole business domain because it takes vast experience, and every organization is unique. I am more concerned about the data scientist mindset, which only focuses on the technical part without considering the business. You could have the best model in the world, but if the model is not considering the business aspect, it is useless.
Business understanding is a process of understanding how our Data Science project affects the business; this means we need to gain as much information as possible to build our data science project. In real business, we could always discuss with the business user to gain a better insight. Example of information we could ask:
- What is the business target, and how do we measure this target?
- What is the business process like?
- Where is the dataset located?
- Would we need a machine learning model, or is simple statistics enough?
- Are there any privacy or ethical issues raised by using the model?
And many more questions we could ask. This is when switching gears is critical for our work.
Because I want everyone to have a business mindset, I want to give an example of project practice to improve your business understanding capability. The business understanding might not be as technical as programming or statistics, but you could train for it. That is why all projects presented in this article are simple, but we would see them from different angles. Let’s get into it.
Travel Insurance Customers Leads
The project we would handle is the Travel Insurance Customer Leads. Predicting which customers are interested in buying the product is one of the classic data science projects, yet it is still sought after by every company. Every company needs their customers as the revenue source; that is why the prediction project is applicable in all companies – in turn, a great project to learn business understanding.
For the customer leads project example, I would use a dataset from Kaggle. The dataset aims to build a prediction model to predict which customers are interested in buying Travel Insurance. Many people would initially want to jump on exploring the data and creating a classification model – but in real business, we would start from the business understanding part.
You might think, "I am not from the insurance industry and not interested in this field, how could I do the business understanding?" – True, for in-depth business analysis, you might need experience in this industry; however, it doesn’t mean we can’t practice and be creative in understanding a business we are not part of.
Business Assumption
Often, we would work together with the business user, but we don’t have a business user; we could ask about the business problem in our case. Let’s work with the limited information we had and try to build assumptions around the information. The business context we could get from the dataset is:
- The company is a tour & travel company that offers travel insurance packages.
- The new insurance package also includes Covid-19 coverage.
- The company wants to know which customers would buy the travel insurance package.
- The data used comes from 2019 customers’ historical data.
The business idea seems simple; the user wants a machine learning model to predict whether the customer would be interested in travel insurance or not. However, did you see any problem with this idea with the information we had? I can spot a few issues already.
For example, are we referring to a new or existing customer? What detail of the travel insurance package? How much would you sell for this insurance? Would the business have a specific campaign? And many more problems I could think of. You could also try to find others by yourself.
We could assume based on the information we have on hand and make a creative guess to answer the previous questions. Let’s say the customer is the existing customer, and the target customer is a frequent flyer with a high income. So, if the business already has this target, would the target prediction change? Absolutely. The flagging target definition becomes different and you need to re-define it. This is what usually happens in the real business world.
Business Understanding Practice
What I want to highlight here is that simple ideas are not that simple. It is easy to develop the machine learning model, but you always need to consider the business aspect. Here is a business understanding practice using this dataset I could give you.
- Try to assume the business process. Try a simple one if it is too hard, such as the business market target and the insurance price. It would be better if you did some research and assumed the whole process.
- Select features for modelling but with business justification. For example, GraduateOrNot, would it be alright to exclude someone from travel insurance based on their education level? It is only for practice, but we try to think logically when selecting features for the machine learning model.
- When selecting a machine learning metric, try to assess how it would benefit the business. For example, with an accuracy of 70%, how would it affect sales? If the prediction off by 30%, how would that affect sales? And by how much? Try to create a simulation to handle this.
- You might want to create a dashboard prediction or reporting analysis to show your data analysis capability and machine learning development. Try to have all the business information that businesses might want to know. Justify why you select these plots or information.
- Explainability vs Prediction Power. Could you have a machine learning model that explains why the prediction model create the existing prediction? You might have the best model, but explainability would help you explain the result for the business.
- Research the business problem from various sources if you are interested in the Insurance industry. It would help you create a data science project that answers the insurance business need. If not, try to find an industry that you are interested in.
Take your time to think about the business problem and be creative with the process. Once you have a grasp regarding the business understanding, it would be easier for you to move on with many kinds of projects. Showing off how good you are at business understanding would make you stand out in the eyes of the company and business user.
For material purposes, here is some reading material you might want to read and project examples that adhere to the business process.
GitHub – cornelliusyudhawijaya/Cross-Sell-Insurance-Business-Simulation: This project is to develop…
How to Set Technical Metrics Data Science Project for Business
Conclusion
While developing a machine learning model is essential for data scientists, we must remember that our work is used for the business. Developing our senses for business understanding thus becomes important to help our job and stand out from others.
In this article, I outline some practices for you to improve your business understanding.
I hope it helps!
Visit me on my LinkedIn or Twitter.
If you enjoy my content and want to get more in-depth knowledge regarding data or just daily life as a Data Scientist, please consider subscribing to my newsletter here.
If you are not subscribed as a Medium Member, please consider subscribing through my referral.
