Predictive analysis is the use of data, statistical algorithms, and machine learning techniques to identify the probability of future outcomes based on historical data. Predictive models analyze past performance to estimate the likelihood of a customer exhibiting specific behavior.
What can predictive analysis do?
A 2014 TDWI report found that predictive analysis is primarily used to perform five tasks:
- Identify trends
- Understand customers
- Improve business performance
- Promote strategic decision-making
- Predict behavior
However, there are three types of predictive analysis that can be applied to marketing:
- Clustering models (segmentation)
- Propensity models (prediction)
- Collaborative filtering (recommendation)
Clustering models (grouping models)
Algorithms create customer segments based on user behavior. Segments are created automatically when users begin to exhibit similar behavior. Think of a cluster as automatic segmentation. There are several types of clusters:
Behavioral clustering (behavioral grouping)
This groups customers based on their behavior during the purchasing process. Behavioral clusters can be created by integrating different data sources: What are the customers' touchpoints? Which promotions do they participate in? How frequently do they purchase? What is their average order value? Algorithms can help determine the correct timing for sending marketing messages, as well as which message is most relevant to each customer.
Product based clustering (product-based grouping)
Within segments, it is possible to gather statistical data on which products those customers purchase. You can identify which products are purchased at different times of the year, which categories are ignored, and which colors are preferred. This information can determine which products are sent in emails or used to personalize the storefront, defining which products are displayed on the website's homepage for each customer.
Propensity models (propensity models)
Propensity Models analyze the likelihood of a user taking a future action based on the past behavior of other users. This model is the most widely used by marketing professionals because it helps project ROI and revenue based on historical data. Customer clusters are an integral part of the propensity model.
Predicted lifetime value (predicted lifetime value)
Based on the history of many other customers, it is possible to predict how much a customer will spend starting with their first purchase. LTV (lifetime value) can be broken down by acquisition channel, as the purchasing behavior of customers originating from social media differs greatly from customers acquired through inbound marketing activities. Propensity models enable marketing decisions, such as investment prioritization.
Predicted share of wallet (predicted share of wallet)
This model predicts the share of a particular product in the consumer's budget. In other words, share of wallet refers to the percentage of a consumer's wallet reserved for a particular category. For example, if a customer spends R$ 1,500.00 per year on footwear, you can predict the revenue potential within your existing customer base and create campaigns to capture that revenue.
Propensity to engage (propensity to engage)
This model predicts the likelihood of a customer clicking on your email links (for example). With this information, you can decide not to send an email to a particular customer who has a low probability of clicking on a specific campaign.
Propensity to unsubscribe (propensity to unsubscribe)
This model predicts the likelihood of a customer leaving your email list (for example) at any given time. Armed with this information, you can optimize email frequency. You can reduce the frequency of emails sent to the “high probability of unsubscribing” segment or choose to communicate with them through another channel.
Propensity to convert (propensity to convert)
This model predicts the likelihood of a customer accepting your offer. For example, companies can use this model in direct mail campaigns where they face high marketing costs.
Propensity to buy (propensity to buy)
This model predicts whether customers are ready to make a purchase. Widely used in inbound sales activities, it identifies when a customer is ready to make a purchase based on their previous actions.
Within the sales funnel, it is possible to address this without the need for an aggressive offer, especially in the later stages. Those who are likely to buy will not need large discounts, while customers who are not likely to buy may need a more aggressive offer, thereby generating incremental revenue.
Propensity to churn (propensity to churn)
This model determines the likelihood of losing a customer during the purchase process and also identifies which active customers are at risk, so you can add them to your watchlist or send a win-back campaign. Often, propensity models can be combined. For example, you may want to win back a customer who has a high probability of churning and a high purchase value.
Collaborative filtering (collaborative filtering)
Collaborative filtering is the marketing term commonly used for recommendation models. It analyzes transaction data from other purchases and recommends products that are more likely to sell.
Up Sell Recommendations (upsell recommendations)
Similarly, these recommendations are made after a customer has already purchased a product, and you can, for example, include them in the confirmation email. You can recommend products to increase the average order value at the time of purchase.
Cross Sell Recommendations (related product recommendations)
It calculates the likelihood of purchasing another (related) product by looking at the history of everyone who purchased this product. This model is much more effective when combined with other predictive analysis models.
Next Sell Recommendations (next product recommendations)
Likewise, these recommendations are normally made after a customer has already purchased a product, and you can, for example, include them in the confirmation email. These product recommendations are specific to each customer, so all purchase and browsing data must be considered, not just their most recent transaction. You can offer products immediately after purchase and automatically add them to the order.
