July 8, 2021 at 10:44 AM
In Part One of this two-part blog, we started providing a short overview of just some of the propensity models that Principa has developed. In this Part Two, we continue to look at different types of propensity models available across the customer engagement lifecycle that are used to predict behaviour and solve business problems.
Propensity modelling attempts to predict the likelihood that a visitor, lead, or customer will perform a certain action. A statistical approach is followed - accounting for all the independent and confounding variables that affect said behaviour. The propensity score, then, is the actual probability that the visitor, lead, or customer will perform a certain action.
So, for example, propensity modelling can help a marketing team predict the likelihood that a lead will convert to a customer, or that a customer will churn, or even that a recipient of digital marketing will unsubscribe. Effective use of data and modelling can therefore improve performance levels achieved in a marketing campaign, resulting in an increased return on investment.
Predictive modelling is a very broad category and it is possible to build predictive models for a variety of behaviours.
Customer Engagement Lifecycle Stages: Acquisitions, Account Growth, Retentions and Reactivations
|Price Elasticity Model|
A method to predict the behaviour of consumers is the use of the price elasticity of demand. The price elasticity of demand attempts to determine the percentage change in the quantity demanded of a particular good or service, when the price of that good or service changes by a certain percentage.
The elasticity of demand helps companies predict changes in demand based on a number of different factors, including changes in price and the market entry of competitive goods.
|Churn Propensity Model|
While the model usually returns a value between 0 and 1, we use the word “propensity” to indicate how likely the customer is to churn, compared to other customers. In most cases, the customers are ranked based on the propensity to churn and split into groups, from the highest propensity to the lowest.
|Next Best Offer Model|
“Collaborative filtering” is the methodology of predicting which products a customer will be the most interested in, based on the preferences of other customers who bought or liked similar products, or the customer’s previous content preferences.
A more advanced analytical form of predicting the most appropriate product offer for a customer is through the approach of “reinforcement learning”, which tests different offers and learns from the response, optimising what is displayed individually to each customer. In contrast to collaborative filtering methods, this approach can quickly make predictions for new customers and products by learning in real time, as well as taking advantage of contextual data about the customer.
This approach allows you to train an individual prediction model for each of the products and services you want to recommend to your customers.
|Cash to Credit Model|
A cash to credit model helps an organisation to determine how much more a cash customer would spend if given a credit facility to enable spend, compared to having no credit facility. Since it is not possible to observe the same customer having and not having a credit facility at the same time, we make use of what is known as a matched-pair technique.
When using this technique, we find customers who are very similar across various profiles, then remove the effects of other “unwanted” variables, which then enables us to determine the impact of having a credit facility more accurately. Where there are no cash customers to match to credit customers for a particular profile of customers, Principa makes use of a regression technique.
By determining how much more a credit customer will spend when given the facility to do so, we can now provide better answers to a wide variety of questions, such as:
- How much more profit can we expect to achieve by bringing on additional credit customers?
- Will this additional expected profit be worth the trade-off of increased risk for their shareholders?
|Fraud Detection Model|
Machine learning models are able to learn from patterns of normal behaviour. They are quick to adapt to changes in normal behaviour and can also quickly identify patterns of fraudulent transactions. This means that the model can identify suspicious customers, even when there hasn't been a chargeback yet.
An algorithm uses customer data described by our features to learn how to make predictions e.g. fraud or not fraud. At the point of the transaction, the model gives each customer a risk score on a scale of 1-100. The higher the score, the higher the probability of fraud.
new customers can cost 6 to 7 times more than reactivating an old one, and the inactive customers are more likely to avail a product or service than people who never bought it before.
The key component associated with customer reactivation is knowing who to reactivate.
One of the analytical techniques to know this is through look-alike modelling. Look-alike modelling is an artificial intelligence technique which uses an ensemble of machine learning algorithms and big data analytics to find groups of people or audiences who are similar to a set of customers with known behaviour. Once this set of customers has been identified, you can apply the appropriate targeted marketing campaigns.
Building Advanced Analytical Capabilities
Principa was established in 1999 and has over the past 22 years developed a multitude of market leading solutions for our clients in South Africa, Africa and the Middle East; across the customer engagement lifecycle - where data analytics lies at the core of the solutions we deliver to the market.
Contact us, to find out how Principa can help you better manage your customers and Work Wonders.
Subscribe here, to ensure that you don’t miss out on any of our future blogs.