You are on page 1of 5

AskmeOffer Extension Assignment

Problem Statements
1. Strategy for analyzing user data related to coupon usage for Askmeoffers chrome extension.
2. Highlight factors and analytical methods to enhance the Auto Apply Promo Code Algorithm.
3. Discuss the data analytics skills to refine the algorithm further based on user experience and
hypothetical data.

Hypothetical Data
I explored the kapiva.in and beardo.in online as well as some famous websites like Amazon, Myntra, Ajio
and Flipkart. I Also tried using the coupons recommended by the extension. To recommend coupons
effectively, we need to analyze the user’s spending patterns and purchasing behaviors. The extension
fetches available coupon code prior to the transaction happens. This way, during transaction we can
recommend the list of available coupons that maximize the discounts for the user thus maximizing the
success rate of the coupon code.

We have following dataset for the analysis. Only features are listed –

Attributes of User What it Indicates


1. Age - numeric Denotes user age
2. Sex - numeric Denotes user sex
3. Income Range - numeric User income level per annum
4. Location - categorical User location – we can categorize it as Tier 1, Tier
2 and Tier 3 cities
5. Frequency of buying (Normal Days)- Indicates the frequency of buying from different
numeric online stores per month
6. Last product purchased – categorical Like electronic, fashion, food, body care, books etc.
7. Last Coupon code used – categorical It indicates the offers given to users like Free
Delivery coupon code, discounted price coupon
code, product wise or store wise
8. Social Media engagement rate – numeric How active is user on social media to recommend
the products
9. Festive Season purchase rate – numeric Indicates the products ordered during festive
seasons
10. Purchases during Mega sales like Amazon Indicates products ordered during these events on
Great Indian Festival or Flipkart’s Big online stores
Billion Day - numeric
11. Number of coupon codes used till dates - Indicates the total number of coupon codes used
numeric by the user till dates (aggregated from the user
profile on extension). Also, good indication of how
likely the user is inclined to use coupon codes

Above features indicate hypothetical data about users, and we can use it to predict the purchasing
behaviors and coupon code consumption pattern for a particular user. We can add few more features like
marital status, family size, education level, occupation, number of children etc.
Strategy of Analysis
The most important part of the analysis is a data related to user spending patterns and purchasing
behaviors. We should analyze the customers inclination and preferences towards coupon code based
promotional activities from the online stores. And to know whether the coupon code strategy makes the
user to buy more or less products.

Coupon code had a significant impact for business as it easier to entice user or customers to try
products at a discounted price. It also shows that coupon code helps the user to buy more and helps the
seller to earn profit. The effectiveness can be measured by analyzing the influences that is created by
coupon code in customer buying decision. It is the sales promotion activity; it is a concept that includes
certain factors which will make a product successful.

We are going to use this hypothetical data and by the application of statistical concepts and
machine learning algorithms we can predict the purchasing behavior, recommend the cross products
(cross selling means, if a user buys a laptop, we can recommend a laptop bag with special discount coupon
to entice the user for further spending and maximizing profits). We can also use correlation coefficients
like Pearsons’s correlation coefficient to find the correlation between numeric features and chi-square
analysis between categorical variables.

Factors and Analytical Methods


To enhance the effectiveness of the algorithms we should employ software engineering fundaments with
the blend of Machine learning to predict and recommend the right coupon code at the right time to the
right user.

We will be using following analytical methods and user data defined above to improve the auto apply
code algorithm –

Analytical Method Use


1. Pandas, NLP Pandas is data analysis library which we will used
for data preprocessing and data cleaning, NLP is
the technique to understand and process the
natural languages
2. Feature Selection and Backward After data processing we need strong features to
Elimination build predictive algorithms. We can use these
techniques to select a meaningful feature from
the feature store
3. Pearsons’s correlation coefficient, Chi- Pearsons’s correlation technique is used to find
square analysis, Feature Imputation the correlation between numeric features so that
we can eliminate features that are highly
correlated and thus reducing feature redundancy.
Chi-square analysis is used for the same purpose
for categorical features. Feature imputation
techniques are used to handle null values or
missing data available.
4. Coupon databases – MongoDB We have a ready heavy system, as we need to
process hundreds of users across a lot of online
stores against the thousands of available coupon
codes. We can prefetch the available coupon
codes from diverse websites and store it in a
dynamic database such as NoSQL MongoDB.
5. Clustering users into similar spending Clustering techniques are unsupervised machine
habits – KMeans Clustering, DBSCAN or learning algorithms used to classify a group of
Agglomerative Clustering similar users that shares similar attributes, so that
will be easier for us to collectively recommend the
available coupon codes
6. Recommendation Algorithms – These are machine learning algorithms.
Collaborative filtering and Content based Collaborative filtering method is based upon
filtering gathering and analyzing data on users’ behavior.
This includes user’s online activities and predicting
what they are going to buy next based on
similarity with other users.

Content based filtering methods are based on the


description of a product and a profile of the user’s
preferred choices.

Maximizing efficiency of the Algorithm and User Experience


The data quality is the most essential element behind the success of any machine learning and predictive
analytics strategy. So, the more data about the user we have the more enhanced our system will be.
Hypothetically we have a user’s data as defined above. Apart from data, the methods we are used to
preprocess and post process the data are also vital.

We can refine the auto promo code apply algorithm by considering following factors.

Data, Data and Data –

Acquire and process more data about the user and build a feature store from which we can select or
modify the features to fuel our ML models.

Coupon Code Databases -

In our databases we are storing the online stores websites and all the available coupon codes, its validity
range, type of the coupon code i.e., product wise, store wise applicable only for a particular website or
some conditional based like the purchasing price should be more than particular amount to apply the
particular coupon. Offline we can study the user’s data and cluster it based on similar users, geography,
spending patterns, products category etc. and based on these assign a group of users with potential match
of the available coupon codes and recommend it to the users. We can perform create set of combinations
and arrive at a specific combination that can maximize the conversion success rate of a coupon code.

We can also use these combinations to –


 build coupon recommendation systems or precompute the most suitable coupon for next
spending for users.
 Product recommendation system to recommend a next product that mostly user likes to purchase
based on history of the data available.
 We can leverage collaborative systems or content-based filtering systems to power our
recommendation system.

Recommendation Systems –

Recommendation systems can be implemented to recommend the products to the user based on user
history and data available, after recommending product we can also recommend the most suitable
coupon code benefits available beforehand to the user not at the time of checkout. This will greatly affect
the conversion of number of recommended items to actual items purchased. During the checkout we can
additionally display the cross /complementary items to the user like laptop and its bag as we discussed
before.

Feature Engineering and Selection –

Perform statistical techniques like Chi-square analysis and Correlation Coefficients to find the correlation
between different features and attributes of the user data, try different combination of features using
forward selection, backward elimination and train a set of Machine Learning models and perform the A/B
testing and productionize the highly accurate recommendation models to track user engagement with
extension, spending patterns and coupon code data.

User – Coupon code mapping to decrease latency–

Pull out the available coupon codes from database based on user’s online store website in real time or
prefetched it offline prior to the purchase event. This will help us to handpick the best coupon deals for
set of users and can be used in future during purchase to decrease the latency of the extension app.

Sometimes the user is applicable for multiple coupon codes and gets confused which one to apply, pick
the best one and apply automatically during checkout and recommend the others with different cross
products to increase the purchase amount. So, in this was we are not only applying the best promo codes
but also recommending others with optional products.

Conclusion
1. Quality and amount of Data will determine the success of the App.
2. Backed by strong data, we can leverage smart ML systems to recommend coupon code and
product item to users to maximize the success rate of the coupon codes.
3. Perform A/B testing by considering various ML models with different features.
4. Precomputation and selection of most suitable coupon code along with alternatives options
during checkout are necessary to decrease the latency issues.

You might also like