Sharpen your Business sense as a Data Scientist
These methods would help you improve business sense
I have argued that business sense is one of the most important skills that data scientists should have. We data scientists are employed to solve the business problem, and we should know what problem we would solve with our data science project.
We can’t just suddenly use a machine learning model, right? This is why business is so crucial as a Data Scientist. You need to understand how the business in more detail. Keep asking a question like,
“What kind of business question exactly we want to solve?”
“Would we even need a machine learning model?”
“What kind of attributes related to the business problem?”
“How is the business strategy and practice within and outside of the company?”.
Asking questions is the first step to understanding the business, but how to sharpen our business sense even further?
Recently, I had a sharing session with Dphi to talk about how business sense is important for data scientists and how we could train ourselves to sharpen the business sense. You could visit the whole session on this link.
To summarize my sharing session, I want to outline the important points of sharpening our business sense as data scientists.
CRISP-DM methodology as a basis for sharpening business sense
The cross-industry standard process for data mining or CRISP-DM is an open standard process framework model for data mining project planning. This is a framework that many have used in many industrial projects and proven successful in the application.
Why is CRISP-DM good to sharpen the business sense? Each data science project on the CRISP-DM is affected by the business decision. Thus, using these steps, we can train our business sense.
The process of CRISP-DM is divided into:
Business Understanding
Data Understanding
Data Preparation
Modeling
Evaluation
Deployment
However, I am arguing to start with the first five and leave the deployment phase to increase our business sense.
Business Understanding
The Business Understanding phase is to understand what the business wants to solve. Typically we ask the following questions:
Determine the business question and objective
What to solve from the business perspective, what the customer wants, and define the business success criteria.
Situation Assessment
Assess the resources availability, project requirements, risks, and cost-benefit from this project.
Determine the Project Goal
What are the technical data mining perspective success criteria?
Develop Project Plan
Try to create a detailed plan for each project phase and what kind of tools you would use.
Data Understanding
Data Understanding shows everything you could understand about the data and relates it with the business question.
Collect Data
Acquire data and explain the data source choices.
Describe data
Examine the data format, number of rows and columns, field identities, and available features.
Explore data
Describe the relationship between data, visualize the data, and be creative. What is important is your data exploration could verify the business question.
Verify data quality
How is your data quality? Many missing values? Is the data collection appropriate enough?. Make sure that the data is past your quality threshold.
Data Preparation
After you understand the data you have, it is time for Data Preparation. We did this phase to prepare the data for the modeling phase.
Data Selection:
Select the dataset, columns, and/or rows you would use. The way you filter data should reflect the business question.
Data Cleaning:
Garbage-in, garbage-out — what happens if you did not clean the data properly. Cleaning data takes a lot of preparation and data understanding — you need a reason to correct data or impute new data.
Feature Engineering
Feature engineering you might think is interesting or helpful. Many business questions get the answer from feature engineering.
Data Formatting
Formatting data when you need it. For example, convert the categorical value into numerical value or vice versa. Don’t forget to state your reasoning.
Modeling
We would develop our machine learning model/product in this phase to answer the business question.
Model Selection
You might want to experiment with many models. It would be desirable to explain why you selected a certain algorithm.
Test design
Justify the reason for how you designed the test.
Model development
The development should consider how the business question and industrial scene would be as well, such as “Would the model I develop be possible in the business,” “Is the resources I need to develop this model is costly?” etc.
Model Assessment
Set your success technical metrics and choose the best model(s) viable for solving the business question.
In a real-world working environment, we don't try to achieve perfection. What we want is a “good enough” model.
Evaluation
The Evaluation phase is different from the Modeling technical evaluation. This phase evaluates the model concerning the business indicator and what to do next.
Evaluate results
Would the business success criteria be met using your model? which model(s) would you choose?.
Review process
Review your work process. Was anything missing? Were all phases executed? Need more time? Try to Summarize your findings and correct anything if required.
Determine next steps
Based on the previous tasks, decide if the model is ready for deployment, needs more iteration, or creates a new project.
This is the step that many data enthusiasts overlooked the most. This is because the study guideline would rarely be taught any business evaluations.
They are the steps that might help you understand how to sharpen your business senses as a data scientist. Below is a reading material that might help you understand how business is important for a data scientist.
https://towardsdatascience.com/learn-the-business-to-become-a-great-data-scientist-635fa6029fb6
https://towardsdatascience.com/a-design-thinking-mindset-for-data-science-f94f1e27f90