Repeat Customer Prediction of New E-Commerce Customers

Determining The Likelihood of Repeat Customers

The one-time buyer problem is one of the most widely known challenges for retail e-commerce brands, as it is difficult to identify the potential repeat customers from only one purchase instance. E-commerce companies often acquire new customers in large numbers through intelligent marketing strategies, but it’s imperative to identify which of these customers are likely to repeat purchases.

Our objective of building a repeat buyer prediction model into LXRInsights, NetElixir’s proprietary customer analytics platform, is to anticipate potential repeat customers among the one-time buyers in the last two months’ duration. Our solution is twofold: one, brands can start accruing loyal customers, and two, retaining existing customers costs less than gaining new customers.

Understanding Your Customer’s Buying Behavior

Through LXRInsights, we track customer’s paths-to-purchase from the first point of attraction to purchase to determine what behavior leads them to conversion. The transaction data we collect includes the time and day that the customers made their purchase, the number of items in the cart, the average order value of the purchase, the referrer or the channel that led them to purchase, the device, and more. We interpret customer profiles from the transaction data that unveils the customers’ purchase patterns. Based on this profile, we use a machine learning model to predict whether a first-time buyer will repeat their purchase. 

Request a full demo of LXRInsights for a holistic understanding of how to grow your business responsibly with data-driven insights.

How Your Customer’s Purchase Journey Trains Our Machine Learning Models

LXRInsights needs key features of the customers’ online shopping behavior to train the model accurately. The accuracy of the model depends on the relevance of the features generated. The following list describes some of the features generated for this model:

1. Pageview information such as the total number of pageviews or average daily pageviews made before completing a purchase.
2. Visit action counts to determine the number of website visits per timeframe of the purchase cycle, from first website visit to purchase.
3. The order information, which includes product diversity, number of items in cart, and order value.
4. The top five browsers, top ten first and second referrers, top landing day of the week, top landing hour, and top checkout hour.

For a sense of how we segment this customer data, download our FACES 2021 report, which catalogs the online shopping behavior of high-value customers across ten e-commerce industries, such as food and grocery and online gifting.

Machine Learning Model

Machine learning algorithms come in handy to accomplish our objective of predicting potential repeat customers. By following the ritual of training ML models, we separate the data into two parts —Training Set to train the model and Test Set to test the model’s accuracy.

Pseudo Code of the Model

Below is the algorithm our model uses to predict the likelihood of repeat customers among first-time buyers:

1. Collect data from the past thirteen months.
   
2. Label the customers with repeated purchases as “Repeat Customer” and the customers with single purchases as “Non-Repeat Customer” for a baseline of your brand’s existing customers.
   
3. Chart relevant features of each segment’s path-to-purchase as outlined above.
   
4. Separate the data into two parts as Train and Test data.
   
5. Train the model by using a set of ML algorithms.
   
6. Measure the accuracy of the model by using Test Set data.
   
7. Select the model with the highest accuracy.
   
8. If the accuracy value is high, share the list of anticipated potential repeat customers.
   
9. Provide the features of the predicted customers that could be used to retain and retarget them.

Conclusions

We use the past transaction data to generate the relevant features of the customers and train the model. By understanding your customers’ buying behavior, your brand can better ascertain repeat customers among your current one-time buyers. You can then generate more personalized campaigns to re-engage these potential repeat customers with your brand and possibly convert more loyal customers.

Visit LXRInsights to request a demo of our customer analytics platform to better understand your new and repeat customers’ online shopping behavior.

References

1. Guimei Liu et. al. “Repeat Buyer Prediction for e-commerce” 2016, KDD August 13-17, San Francisco, USA.
2. https://www.custora.com/blog/four-steps-to-solve-one-time-buyer-prob