Machine Learning

Fuzzy Clustering

The Fuzzy Clustering model groups users in homogeneous groups based on the buying behavior of specific products or categories. For each customer, the Fuzzy Clustering model finds the top three clusters and their associated probabilities.

Fuzzy clustering enables customers to be part of multiple groups and calculates the probability of belonging to each cluster. The output for each scored user is the name, id, and probability for the first, second, and third most likely clusters.

Use cases

Identify micro segments in product clusters

Treat customers in the same cluster differently based on their probability of belonging to that cluster. For example, you can run two different campaigns for the customers with 0.98 and 0.25 probability.

Build segments that leverage a combination of clusters for granular targeting

Target customers that belong to multiple clusters. For example, you can create a menswear-accessory bundle deal that consists of the customers with 0.60 probability in menswear and >=0.25 probability in accessories.

More targeted product or brand campaigns

  • When customers access your website, show content based on their product affinity.

  • When creating a product-focused campaign, identify and target the customers who have previously purchased such products. These types of campaigns can help increase affinity for products that customers already prefer.

  • When launching a new product, identify an existing similar product, locate the customer cluster aligned with the existing product, and target customers in those cluster(s) in the launch campaign. This ensures a high response rate as you target the customers with similar product affinities.

Deepen product affinity and increase promotion spend effectiveness

Engage with your customers better without overspending. You can create surgical discounting or pricing tests. For example, you can give a lower discount for product X or similar products to the customers that belong to cluster(s) aligned with product X. If a customer already likes product X, you do not need to give a high discount.

Broaden customer’s purchased product mix

Improve retention, loyalty, and lifetime spend by encouraging customers to buy different products. You can create tests by providing higher discounts for product X or similar products to customers in clusters that are either negatively indexed towards product X or do not feature product X at all. When you convert such customers to purchase product X, you help improve retention and gain incremental revenue without impacting margin. As such customers would not have purchased these products prior to your tests, the benefits are substantial.

Nudge customers that only buy one category of products to buy into other categories

Use Metrics to identify the most common one-time purchased product. Based on that, you can create campaigns to encourage one product purchaser to buy other products, thereby improving their retention and lifetime spend.

A customer may not strongly belong to any specific cluster and instead have all the cluster probabilities almost equal. For example, you may want to treat your customers in the same cluster differently if one belongs in that cluster with .98 probability and the other with .25 probability.

Some marketing mediums are expensive. Therefore, you can use the fuzzy clustering model to further filter a cluster. Fuzzy clustering provides a way to group such customers. For example, you may want to target the customers with >=0.8 probability of belonging in a cluster. Alternatively, you can include more customers or target the customers whose first or second most likely cluster is a certain cluster.

Target audience

The Fuzzy Clustering model predicts the behavior of buyers.

  • A buyer is a customer with a transaction in the past three years.

  • The model is trained on the customers with more than two transactions in a particular time frame.

  • The customers with more than two transactions in a particular time frame are scored by being assigned to their top three clusters and probabilities.

  • The customers with one transaction are labeled as One-time buyer.

  • The customers with no transactions are labeled as Non buyer.

How does the model work?

For each contact, the model calculates the purchase behavior features in a specific category.

The model binds historical data multiple times using a range of values for clusters. It also generates various graphs and metrics, and uses them to determine the clusters that can be used for scoring. The goal is to have stable and meaningful clusters that exhibit reasonable statistical differences in the purchase behavior.

The name and probability of the first, second, and third most likely clusters are displayed in the CDP user interface.

Using the Fuzzy Clustering model

CDP displays the cluster names for the top three clusters in 360 > 360 Profiles. In addition, CDP displays the cluster names and probabilities for the top three clusters in Actions > Campaigns+, and Analytics > Metrics.

To learn more about how the Fuzzy Clustering model can enhance your marketing workflow, contact us.