Introduction
As data scientists, we’re taught to work with numbers and use mathematical models to come up with predictions.
In practice, however, data science is very different from the material taught in an academic setting.
More than knowledge of complex algorithms, you need to be able to quickly derive value from massive amounts of data.
There are many opportunities for data professionals out there — in the field of medicine, finance, insurance, and security. To really add value as a data professional in these industries, you need to have domain knowledge.
You can only find meaningful patterns in data and come up with a valuable observation if you understand the context.
Steps like data collection and feature selection require a lot of human judgement. You need to understand the importance of each variable to the problem at hand. Otherwise, your model will be uninterpretable and meaningless.
For example
Let’s take a simple example in the marketing domain.
You are working at a large restaurant chain — Restaurant ABC. You need to build a clustering model to look into different customer segments. There are some variables that are more important to your model than others.
You are more interested in customer traits — like the amount they usually spend, number of purchases made, competitor brands they frequently visit, etc.
You are less interested in attributes like gender and age. While they do add value to your model, you want your segments to be separated mainly based on behavioural attributes.
This means that you’d have to add weights to different variables, so that variables of higher value have a larger impact on your model.
The understanding that some variables need to have higher weightage comes from domain knowledge. This improves the quality of your model and makes it more applicable to the problem you are trying to solve.
Why Learn Data Science for Marketing?
The biggest complaint that companies have with data scientists is the inability to derive value from the models built.
While these models are technically sound and use state-of-the-art algorithms, they simply don’t tie in to the business requirement. These models are often treated as black boxes, and the outcomes are inexplicable.
The insights derived are not actionable, and putting a model like this into production is risky.
If you are a data science aspirant who is yet to land a job, it is a good idea to acquire domain knowledge before applying.
It is easier to gain domain knowledge in fields like marketing or business analytics as compared to industries like healthcare or finance.
It will be a lot easier for you to land a job in these fields if you have knowledge of both data science and marketing.
There is high demand for people with a blend of these skills — people who have enough marketing domain knowledge to understand the process of acquiring new customers, along with enough technical knowledge to come up with data-driven solutions.
In this article, I will provide you with some resources you can use to learn data science for marketing. I will also break down the job scope of a data scientist in marketing, and the kinds of models you will build on a daily basis.
Learning Resources
If you have already taken a beginner level data science course and have a basic understanding of machine learning models, you can consider taking this Datacamp track.
It consists of 7 courses, and takes you through concepts like analyzing marketing campaigns with Python, sentiment analysis, customer churn prediction, market basket analysis, and A/B testing.
One advantage of this course as compared to other marketing analytics courses is that it teaches you concepts like data collection.
Often, when working in fields like marketing, you won’t be handed a pre-made dataset to analyze customer behaviour. You will need to source for external data through techniques like web scraping or with the use of APIs.
This Datacamp track contains courses that cover a broad range of topics in marketing analytics — ranging from external data collection to making business decisions with machine learning models.
2. Marketing Analytics — Coursera
The Datacamp course above was broader, and covered a wide range of material. It is a good introductory course to the field of data science in marketing.
However, if you want to dive deeper into the field of marketing and understand concepts like customer lifetime value, retention, and the various business metrics, this course is ideal for you.
This marketing analytics course doesn’t include programming or the use of machine learning models (except for regression analysis).
It is a course geared towards individuals who want to understand the domain better, along with the types of analysis that can be done for different business use cases.
I suggest taking this course if you already have a solid foundation of programming and machine learning. This course will provide you with the ability to tie your data science skills to the field of marketing.
Applications of Data Science in Marketing (+Tutorials)
Data scientists are able to come up with recommendations that aren’t always obvious to marketing professionals. They are able to find trends in data and showcase actionable insights that aren’t always intuitive to a human.
Some popular marketing data science applications include:
Market basket analysis
With market basket analysis, data scientists are able to identify a list of items that are frequently purchased together based on customer sales data.
This can then help the marketing team to come up with bundle packages and discounts.
Here is a market basket analysis tutorial you can code along to.
Customer segmentation
Customer segmentation is the practice of segregating a company’s audience into different groups. Each group will have shared characteristics.
With the help of data science techniques like clustering, customer segmentation can be automated and performed quickly.
These are usually distance based algorithms that can pick up on similar traits between audience sub-groups that a human might overlook.
Here is a customer segmentation tutorial I created that you can code along to.
Churn prediction
Customer churn is the percentage of customers that stopped using a company’s product during a certain time frame.
A churn prediction model is a data science model that is created to predict if a customer is going to churn. Based on this prediction, the marketing team can then shift their strategy to retain individuals who they are about to lose.
As I mentioned above, model explainability is a very important aspect of data science.
If you have a model that functions as a black box algorithm and is able to predict accurately when a customer is going to leave, but you aren’t able to understand why, then the model is of no use.
Here is a simple customer churn prediction model that you can code along to.
Channel optimization
Channel optimization is a process through which the marketing team usually comes up with techniques to improve customer experience and increase interactions.
An example of this could be optimizing hashtags on a company’s Instagram post, selecting the right keywords, and coming up with targeting mechanisms to gain new leads.
Data scientists can help in every step of this process, by coming up with recommendations based on what worked well in the past.
Sentiment Analysis
Sentiment analysis is the process of understanding user interaction with a brand, and gauging the underlying emotion behind statements they make.
This is useful when done on a large scale — for example:
Restaurant ABC wants to understand the pain points of their competitor brand, so they can leverage on this to acquire new customers. The best way to do this is through sentiment analysis.
Collecting reviews of people who have visited their competitors can help them understand what consumers are struggling with, and how they can address these issues.
I created a sentiment analysis tutorial using data collected from Twitter, and all the codes are available for you to follow along.
Natassha Selvaraj