Low Effort High Impact Data Science Projects
As a product manager, how can I use Data Science to help increase my company's bottom line, and get a bigger bonus.
Epic: Knowing the Technology Landscape
Part of succeeding in Product Management is understanding all possible technologies at your disposal. A good PM keeps their ear to the ground for innovations that they can co-opt for themselves.
Today, that new frontier all PMs should be looking to adopt into their products is Data Science. To educate the Pivot to Product ahead of bonus season, we have jungle Data Science Expert BowTied_Raptor to educate us on some easy effort projects that can translate to big wins.
Raptor - take it away!
If you are in charge of a data team, or you are the manager for a business that’s trying to figure out how you can incorporate data science and make a real impact on the bottom line, then this post is for you. One of the biggest frustrations that most managers have with data science is that in the past Data Science & Machine Learning promised the world, and very quickly fell flat on it’s face. Time and time again, you will hear machine learning enthusiasts talk about how their model is phenomenal and can beat the current State of the art (SOTA).
So, in this post, we will have @BT_Raptor help us out and see what projects he recommends which are easy to setup, and deliver a solid high impact on the revenue line for a business.
If you want to learn more about what Data Science is, click here.
Low Effort High Impact Data Science Projects
Sentiment Analysis
When you have a product you’ve released to your customers, one of things you want to make sure is that you have continuous feedback from the customers to the product development team. If the business in question is large enough, you cannot take the feedback from the customers and keep auto emailing the product development team. What you will need to do is take a look at the reviews you are given, and use some sort of Natural Language Processing (NLP) in order to categorize it, and figure out how to implement the requested changes, or improve the product in a smart fashion.
This is where Data Science can come in handy. If you are using R, the library you would be interested in is called stringi. For Python, you would be interested in Natural Language Toolkit (NLTK). If you need some help setting up libraries, then click here.
Here is a list of areas where Sentiment Analysis can help you impact the bottom line:
Pinpoint what areas need more development
Analyze the sentiment on your competitors
Analytics on a new potential product compared to an old similar one to simulate customer feed back.
💡White Belt Note: You can even find some out of the box models to run your data through if you don’t have Data Scientists on staff. Check out Amazon Comprehend
Customer Segmentation
Customer Segmentation refers to the process of separating your customers into different categories based off of similar attributes that they have. A few common ones that most Data Science teams employ is demographic data such as local economic neighborhood, average age of customers, customer loyalty level, etc…
In most businesses, you typical have a human that has strong domain knowledge on whatever it is that the business does, and typically they are responsible for doing the segmentation. The problem is that this relies 100% on their domain knowledge, what happens if they missed some information because they thought it wasn’t important, or what if they have some sort of a bias in the analytics?
This is where machine learning comes into play, and can help you out. Machine learning offers clustering algorithms which are un-supervised such as k-means clustering. Unsupervised machine learning algorithms means you do not have to provide the algorithm with some sort of a target y-column that it is trying to predict. Instead, it is given a set of data, and it has to make the analytics entirely on it’s own with minimal instruction.
Unsupervised clustering algorithms are useful for customer segmentation because since they are not given any sort of target y-column to try to predict, they have very little underlying biases, so typically these algorithms can find insights that pretty much no-one would have thought about.
Recommender Systems (Price & Upselling)
Recommender systems refer to algorithms that can predict the future preference of a set of items for a customer, and recommend them items that they are likely to purchase. Think of it like amazon or ebay, whenever you make a purchase, there is an algorithms of a sort in the background watching your every little move. Where you click, how long you’ve been on the page for, if you go back a page, etc… All of this information gets passed onto an algorithm and this algorithm will try to figure out which items you are very likely to purchase, and try to recommend them to you.
If you are interested in utilizing recommender systems, a great library for them is called Surprise.
That’s all for today.
If you want to deep dive into Data Science, be sure subscribe to Bowtied_Raptor substack Data Science and Machine Learning 101
Looking to break into Product? Check out below!
Product Manager - Brooklyn Data Co - Fully remote - $98,228/yr - $144,453/yr
Product Manager Eletronics Retail - TheoremOne - Remote - $130,000/yr - $155,000/yr
Got a product resume you need to up-level? I’ll review your resume, then talk tactics on a call.
Are you struggling with a product problem today? I’ll hop on a call and help you put forth the game plan to solve your issues.