How to use propensity modeling for web personalization

For decades, marketers have used propensity modeling to develop clearer campaigns. With targeted messaging, brands can better deliver resonant messaging to customer segments.

Propensity modeling lets marketers speak to customers in ways that ring true.

Today, the need to build propensity marketing models has never been clearer, especially when building personalized web experiences There are more messages bombarding customers and more noise to break through.

There is also more data available from which to build these models. The key for marketers is to develop the right strategy, using the right data, to deliver the right messages to the right customers.

What is propensity modeling?

Propensity modeling dates back to 1983. It’s a statistical process that tries to predict the likelihood that visitors, leads or customers will complete a specific action.

It is a modeling technique that attempts to account for independent and dependent variables that can affect the desired behavior. For marketers, it helps predict which leads and customers are mostly likely to respond to messaging and campaigns.

Armed with propensity modeling tools, marketers are able to target emails, social media and other campaigns to those customers that have the highest likelihood — propensity — to respond.

By determining the probability customers will respond to particular acts, brands can better target different subsets of their base. It can be a powerful way to manage costs and prioritize segments.

What’s more, marketers can then track and analyze the predictive modeling and refine future campaigns to reflect actual results.

Propensity modeling can have a powerful impact on your business. It can help develop better product recommendations for existing and potential customers. It can recommend content to move web visitors through a buyer journey towards a purchase. And, of course, it can help segment your customers and targets for specific products and services, needs, messages and campaigns.

Building a propensity model for web personalization

Not all propensity models are created equally. Many brands struggle with propensity modeling for several reasons:

• Quality. Many CRMs and marketing automation platforms today come with pre-built modeling tools. While these are scalable, they are often not robust, relying only on some basic information to build the model. For example, they may factor in customer information and some transaction data, but do not consider activity and the complete transaction history
• Static. Too many models are based solely on static indicators. They do not incorporate new learning, actions and results to refine the model. With machine learning and artificial intelligence tools so readily available today, brands can build models that use more data and results, identify patterns and evolve over time
• Misuse and Execution. Interpreting the data that propensity models generate is important. Too many brands stop using propensity models because they do not commit to the analysis needed to fully leverage the opportunities. In addition, operational changes may need to be made – what data are collected, what factors are tracked – to build and use effective models. If companies do not act quickly to identify predictions and act, they will not fully leverage the opportunities

Developing successful propensity models

What does an organization need to do to start building effective propensity models for web personalization? Here are a few tips.

1. Set clear goals

What are you trying to achieve with your model – more leads, more sales, more engagement? Your models can be built to achieve multiple goals. Your goals should be accompanied by perceived values for those goals. A set of goals that have varied goals is more valuable, allowing the model to optimize the segments to meet each goal. The key is to understand the goals, goal values and the data points necessary to make the model work best.

2. Know data attributes

Propensity models need good data to provide the best insights. The more pertinent data points the model can use, the more likely the web personalization will achieve the desired goals. To build good propensity models, however, your brand needs to be sure that it is using contextually relevant data. You want the data points that are most salient to the stated goals.

With so much data available today, marketers must also be mindful of the need for agility. You do not need to use the same data for all your models. In fact, thinking carefully and creatively results in better modeling.

3. Worry about data pipeline production

Your modeling will work best when the data pipeline is built to collect and use the necessary information. Your data pipeline production needs to be efficient at each stage – ingestion, validation, storage, updating and usage. When building propensity models, marketers also need to evaluate business processes to ensure that the data pipeline delivers what the models need.

4. Think about scalability

While propensity models can and should be modified over time, you want to have practical and scalable modeling. You need to think about how to scale your models so that you do not take a one-and-done approach, which is wildly inefficient. Effective models will be able to meet the needs of multiple predictions and business goals while being easily adapted and revised.

Propensity modeling in a data-rich environment can have a transformational impact on how customers engage with your website. Building and using the models well will position your marketing work for better ROI and better financial outcomes by giving customers experiences on your website that are resonant and relevant to their needs.