5 Predictive Analytics Models and How to Use Them

Businesses are better than ever at collecting customer data, but unless you have the tools and processes in place to make the most of it, its potential will pass you by. Predictive analytics offer a number of ways to guide your future planning and anticipate the long-term outcomes of your business with confidence.

In the marketing department, you can use predictive analysis to create tailored key performance indicators (KPIs) that align with the company’s goals. Applying the right model to your marketing efforts is at the center of success, so it’s vital you understand the nuances of each one.

1. Clustering model

As a marketing manager, you’re probably familiar with customer segmentation and the clustering model facilitates this process with the use of an algorithm. Instead of creating the segments personally, this form of predictive analysis does it for you automatically. The beauty of auto-segmentation in this way is that it goes way beyond anything a human would be able to achieve.

Clusters are made up of cluster DNA, which are all the different variables being analyzed. From first order revenues to the number of days between orders, these variables help to build up a highly accurate picture of your customers. Cluster DNA can look very different from one cluster to another, demonstrating the functionality of this model.

The most commonly utilized clustering algorithms are:

Behavioral clustering

Understanding how your customer behaves when making a purchase will help you tailor your marketing in a way that they’ll respond well to. This means taking everything into account from the channels they choose to use right through their typical order values and how frequently they make purchases from your brand.

Product-based clustering

By getting a clear picture of the groups of products your customers buy from, you can send them content that will tap into these interests. Some customers will only purchase a single type of item, but others will make selections from various departments, so it’s worth using an algorithm that can recognize this. It will also allow you to stay clear of areas they have no interest in at all.

Brand-based clustering

Grouping customers together into the brands they’re interested in is an effective way to help inform where to target new product releases. Algorithms that identify the crossover between multiple brands will enable you to shape your strategy accordingly, delivering the content and offers that will appeal to the widest possible audience.

2. Propensity model

The propensity model is the one that many think of when they hear the term predictive analytics. It’s like a time machine, making true predictions about the future behavior of your customers. There are a wide range of factors you can learn from this model, including:

• Predicted lifetime value
• Predicted share of wallet
• Propensity to engage
• Propensity to unsubscribe
• Propensity to convert
• Propensity to buy
• Propensity to churn

You may wish to rely on a selection of propensity models used together for the greatest effect. For example, this can lead to getting a good return on investment (ROI) on a customer that is both highly likely to churn but also offers good predicted lifetime value. You just need to design a customer win back campaign to fit.

3. Recommendation model

It’s hard to imagine a time when the recommendation model wasn’t a key tool for platforms such as Amazon, which routinely rolls out its “customer who liked this product, also liked…” feature. This is achieved through collaborative filtering, with the algorithm using all the previously buying and browsing data collected on a customer to make such recommendations.

There are multiple ways the recommendation model can be implemented. The first is upselling at the time of purchase, while cross selling can also add value to a basket before it’s taken to the checkout. Next sell recommendations occur once the purchase has already been made and are often placed in the confirmation email or in subsequent follow-up communications.

4. Slider model

Getting a heads up that a customer is likely about to slide across to another business is incredibly valuable, because you can do something about it. While churn sees your audience stopping purchasing from you altogether, sliders can go either way between you and a competitor, meaning you lose out on potential sales on a regular basis.

Using this predictive analytics model to target sliders with loyalty programs can ensure they remain a high value customer and are not splitting their spending with another company. The slider model is particularly useful in competitive retail markets.

5. Outliers model

Overlooking anomalous data is tempting, but putting the outliers model to work can offer more insight than you might expect. It’s used extensively within the retail and financial sectors, and can be a good indicator when something isn’t quite right. This can be anything from a product failure through to a fraudulent transaction.

Identifying the amount, location, time, purchase history and the nature of a purchase presenting as an outlier can help to pinpoint why it has occurred. This can give a more well-rounded picture of a brand’s audience and better understanding of how things can go wrong as well as right.