a hand holding a crystal ball of connected data


Imagine if you had a crystal ball which could tell you the likelihood for your customers to buy your products before they ever see an offer? Perhaps it would tell you what message would resonate with a customer before they ever see it? Wouldn’t it be amazing if you could predict your customer’s buying behavior and preferences? Perhaps you would make sure you tailor your messaging to speak to as many people with similar interests and preferences as makes sense. After all you have only so many hours in the day right?

As fictional as it may sound, predictive modeling is that exact crystal ball for businesses today. The term predictive modeling/analytics describes the use of data driven segmentation, predictive algorithms and even machine learning to predict consumer responses to your products and offerings. Businesses use customers’ historical data to predict their behavior in the future. In fact, real time data is also becoming more and more relevant for predicting behavior as technology allows us to capture and use that day as it comes in. After predicting the behavior, organizations can strategize for customer acquisition, campaign strategy and even client retention.


Businesses have been using customer data for decades. However, their use of that data was always to analyze past historical data to feed future business strategy. Although “B.I.” (business intelligence) has been around for a while its becoming ever important with new visual tools like Domo, Tableau, or Qlik which allow for easier comprehension of complex business performance metrics through easy to synthesize and comprehend visuals. . In the past BI feed the ability to retrospectively make adjustments focus on tactics for future improvement and hope for business growth. Fast forward to 2018 and replace the “hope” part with “prediction”. Today’s technology allows businesses to make predictions about future events. Here is how it works.

  • John buys a new smartphone from retailer ABC
  • Retailer ABC uses data from thousands of customers just like John who bought the same smartphone and predicts that John and those customers will buy a case for the phone and possibly other peripherals
  • Retailer ABC constructs a “cross sell & upsell model” to deliver marketing messages to drive further sales to a mathematically proven audience deliver the right message, at the right time, to the right person, in the right channel (i.e. social vs email)
  • Retailer ABC further uses John’s and his lookalike fellow customers (audience persona) to go out and find prospects who share similar data profiles to John as they are more likely to buy that specific smart phone
  • Retailer ABC also uses buying history and various other data to prioritize John and others who have really valuable lifetime purchasing value
  • Using prioritization Retailer ABC then scores all clients and prospects coming into their marketing funnel as likely to be or not to be their likely best buyers

Why predictive modeling and analytics make sense today is because businesses have the capabilities now more than ever to execute them. That capabilities includes lots of data being collected (big data), fast processing power, faster data systems, robust data analytics tools, and skilled data analytics professionals. All these components work together to predict possible future outcomes.

By 2022, the market for predictive analytics will grow by 22%. More importantly, nearly 36% of businesses believe predictive analytics is one of their most crucial investments. From product suggestions in the retail industry to illness forecasts in healthcare, predictive analytics has become vital to modern business’ success. What’s important is that even non-technical, traditional marketers are starting to utilize predictive modeling and stop using data as only a way to analyze past performance.


It is not possible for any organization to retain 100% of its customers forever. However, preventing existing customers from unsubscribing to your business is worth every effort. Mainly because it is 7 times cheaper to retain an existing customer than to acquire a new one.

Churn is the number of customers unsubscribing to your business in a set timeframe. Divide that by the number of subscriptions at the start of that timeframe and you have the churn rate. By using predictive models for churn rate, you can predict when a customer will unsubscribe to your business. More importantly, the data you have will tell you why the customer will unsubscribe.

Marketers can use predictive modeling to assign churn scores to customers to indicate their proneness to unsubscribe based on time, behavior data, and even customer segmentation profiles. To do that, they use predictive analytics on the data they capture from website searches, product purchases, customer surveys, CRM notes, etc. The more data they collect for predictive analytics, the better predictions they will have. For example, they could collect data from social interactions or sentiment analytics of customers to know their churn propensity.

However, when you have such huge volumes of data , it must go through some filters to be useful for predictions. Keep in mind that low-quality or unreliable data can result in false predictions. Bad or low quality data causes 38.5% businesses to take wrong decisions. Hence, it is critical to apply certain techniques to mine useful insights from the data.

There are various techniques that marketers can use to find useful patterns in the data available to them. Some well-known and useful techniques include anomaly detection, automatic clustering, segmentation algorithms and more. Let’s look at them in more detail.


  • Collaborative (Recommendation) Filtering
  • Propensity (i.e. Predictive of Best Customers) Modeling
    • AKA – Multivariate Modeling
    • AKA – Regression Modeling
  • Cluster (Segmentation) Modeling
  • Churn (i.e. Predictive of Lost Customers) Modeling
  • Classification (Decision Tree) Modeling
  • Forecast (i.e. Predictive of Revenue) Modeling
  • Sentiment (i.e. Predictive of Brand Perception via Social) Modeling

To start with an example collaborative filtering or recommendation engines are common in eCommerce. The purpose of this model is product and service recommendations to customers based on the history of their online shopping. Marketers can cross-sell and up-sell products to customers using this model. In fact, companies can even suggest the customer what next item he/she should buy based on their recent purchase. Predictive analytics allows you to put greater discounts or more advertising spend behind a targeted churn audience versus spending a significant budget in a targeted way.

In propensity modeling you first analyze the characteristics of customers by looking in to the past purchase and behavioral data (i.e. click-stream) and even append 3rd party data such as demographics or even pre-built consumer segmentation tools. Based on those customer attributes, you can predict, mathematically, the propensity of the customer to perform a certain action. Some of the models you can create include but are not limited to: propensity to purchase, propensity to engage, propensity to open an email, and propensity to unsubscribe just to name a few.

You use these models to segment your customers using a variety of data attributes. You could segment your customers based on variables like average order size, age group, interests, geography, and demographics.

Below is a basic example of what a cluster segmentation graph may look like. Using analytics, removing outliers, and defining common data attributes an analyst can create a visual represent and audience outlined is this example below with 4 different clusters:

A churn model helps companies know the factors that can cause a customer to stop using your product or service. It further suggests what steps the company can take to reduce the churn. It predicts a time-frame within which the customer will churn. Most importantly, it uses data to predict “who” will churn. Churn modeling can have a significant impact on the bottom line for high volume, large consumer businesses like telecom, automotive, and banking.


You can incorporate predictive modeling for your marketing success in these simple steps.

You must have a problem at hand to solve it. What is it that you want to predict about your customers? What are your existing marketing challenges? What do you want to discover from the past data of your customers? The answer to these questions will also decide what actions you will have to take to achieve your goal.

Predictive modeling depends entirely on data. Once you know the problem, you have to collect every piece of data to feed to your predictive models. Various sources for your 1st party data include website/ecommerce, CRM, social networks, CMS, and marketing automation tools just to name a few. At this point, having the right data management partner is critically important. At Cruz Street, for instance, we have professionals and expertise to work with your systems or our own to build a data repository for you to conduct your modeling and analysis efforts. We also provide analytical tools and analysts to help you make sense of your data collection challenge.

This is another point where you will need a data analyst. You can use a variety of software tools that help you with business intelligence and predictive modeling. Read an article on technology you may consider in this process. The right data partner can assist in successful deployment of your predictive models.

You need a team to continuously monitor the performance of your models. These IT and data professionals will work together to keep the infrastructure that supports your predictive models intact. At Cruz Street we have a data ecosystem, predictive analytics platform, and hosted model solution to help you achieve your goals.


Acquiring new customers and retaining the existing ones have to be more than finding patterns in the past data. The next step in marketing is predictive analytics i.e. predicting the next move of your customers and creating action plans accordingly. However, proper data management serves as the starting line to successful predictive analytics. When it comes to data management, many organizations feel apprehensive due to their teams, experience, and existing priorities. However, all businesses will eventually need an edge to compete in this ever data driven space. Let Cruz Street Digital help you as your chief data officer/scientist on-demand. .

Our experts can help you develop an infrastructure where high-quality data can be fed into predictive models we help you build to make the most reliable and accurate predictions of your customers’ behavior possible.

Related Posts