In Kurvv’s quest to make the power of machine learning accessible to business users, we hare happy to say that our Life-Time-Value (LTV) bases Segmentation solution now also works for eCommerce customer additional to hotel (hospitality) customers. (limited to Shopify & Square users but more is on the way) Our LTV based segmentation models predict how valuable new customers and leads will be to your business over their lifetime, and group them into low, medium and high value categories. We do this by analyzing historical data to identify which attributes are most often associated with high-value customers.
Then, when you get new customers and leads which share those “high-value” attributes, we can predict their likelihood to purchase again in the future. These “predictions” has almost infinite ways it can be used. For instance, if you’re running marketing campaigns but have budget constraints, customer segmentation can help you more precisely target your marketing towards higher-value customers and reduce spend on lower-value customers.
If budget isn’t your primary concern and you want to increase scale, customer segmentation can give you a list of high-value users which you can use to seed lookalike (similar users) audiences with advertising partners. As well, customer segmentation can tell you which customers are likely to be more responsive to cross-sell promotions. Our segmentation product for Square and Shopify uses two different machine learning (ML) algorithms to classify your data, and automatically selects the best-performing model to use for the final output. Below is a brief description of how each algorithm work (along with links in case you’re interested in the math behind this)
KNN algorithms transform attributes about your customers into distance measures, and then create groups of the “closest” customers – that is, those with the most similar attributes. To learn more, check out this article here
Naïve Bayes uses traditional statistical methods – specifically Bayes Theorem – to group customers. Bayes Theorem basically looks at the impact of each attribute about your customers independently and measures the probability of that customer being of a particular value – high, medium or low – given that attribute and the baseline probability of each customer value. Finally it adds up the probability of all the attributes considered and makes a final determination. To learn more, check out this article here.
For Shopify and Square, the models train on your historic customer geographic data (city, state, zip) and transaction information (first product purchased, price point, etc.). Then those “trained” models are used to predict the lifetime value of new customer records as they’re created. The output dataset contains customer IDs, emails, opt-in/out information for targeted marketing, predicted customer value and the model’s confidence in the predicted value – everything you need to begin using the data to drive value for your business.
Machine Learning Basics with the K-Nearest Neighbors Algorithm- Onel Harrison
Introduction to Naive Bayes Classification – Devin Soni
Photo by NeONBRAND on Unsplash