A featured contribution from Leadership Perspectives: a curated forum reserved for leaders nominated by our subscribers and vetted by the Construction Tech Review Advisory Board.
OpenJaw Technologies
Powering Customer Centricity with Machine Learning
One of the most exciting aspects of this for a data scientist is that machine learning enables most of this change.
But the real paradigm shift here from a retailing perspective is the move from ‘ProductCentric’ to ‘Customer Centric’ retailing. Every time we experience a personalised offer, recommendation or experience online that is tailored specifically for us, an individual customer, we are witnessing Customer Centricity in action.
Why is Customer Centricity Better?
Every retail business is ultimately interested in selling more products to more consumers at the highest margin possible. One of the most effective ways of doing this is to acquire a deeper understanding of the interests, preferences, and needs of consumers and then to use this knowledge to more effectively target and match consumers with available products.
In the online retailing world, Amazon is a master of this approach. Amazon will analyse your history of transactions and your search activity on their web-site so they can match you with "look-alike" peer-groups to determine the optimal path for conversion, up-sell or cross-sell.
There are now countless other online retailers following Amazon’s lead in pursuing a Customer Centric approach to retailing using Machine Learning, across a wide variety of traditional industries, spanning travel, fashion, and financial services.
So let’s take a quick look at some of the most popular Machine Learning algorithms and predictive models that are powering this shift to Customer Centric retailing.
Machine segmentation with clustering algorithms
Clustering algorithms perform a type of machine segmentation that is given a history of customer transactions expressed as features such as ‘recency of purchase’, ‘frequency of purchase’ and ‘average spend’, the algorithm (usually a variant of ‘K-Means’) will find clusters of customers that are similar across these dimensions.
Collecting large quantities of customer data and applying Machine Learning to generate predictive insights is an important, foundational step. But these insights must be actioned by downstream systems to impact real change in the business
This is very useful for strategic insights, such as identifying groups of customers that are ‘loyal & frequent’ versus ‘risk of churn’, but can also be used for specific tactical initiatives, for example, a cluster of customers that were big spenders in the past, but have not purchased for a long time may be a very good group for a ‘win-back’ marketing campaign, a staple for many digital marketing teams today.
Ranking customers with customer lifetime value models
These models can take various forms, but most work with features like recency, frequency and spend to assign a monetary value to a customer. This normally spans about 18 months, 6 months of which are in the future, so the models usually are predictive in nature. There are many applications for customer lifetime value in retailing, for instance, it may be used to determine a discount level in a personalised offer or to rank customers to vary the speed of response in customer service scenarios or to inform how much to bid for online real-estate in a retargeting advertising campaign.
Predicting purchase probability with propensity models
The Machine Learning algorithms used to parameterize propensity models can vary widely (spanning logistic regression, random forests and deep learning to name a few), but they all tend to align to a similar hypothesis. This hypothesis is based on the principle of ‘propensity to purchase.’
But what does this word 'propensity' actually mean? Let's take a simple example to illustrate. A person may have a high propensity to purchase travel insurance if they are naturally risk-averse, purchased travel insurance in the past, are currently browsing online for insurance products and fit a particular demographic profile that can afford it. In many cases, a person may not be consciously aware that they have a propensity to purchase something, but the propensity model can predict it – that is the spooky part!
Some of the more sophisticated retail banks and insurance companies have used this approach to great effect over the past few years to increase the average product holding of their customers. Other industries, such as travel, are catching up fast and are now starting to combine the fast-moving ‘intent’ data found in online search and browsing activity with the slower moving transactional and demographic data needed to build accurate propensity models.
From insight to action
Collecting large quantities of customer data and applying Machine Learning to generate predictive insights is an important, foundational step. But these insights must be actioned by downstream systems to impact real change in the business.
For example, if you want to win back customers that were high-value in the past, but have now churned, then using your email campaign tool, powered by Clustering insights that precisely identify the most appropriate customers to target may be an optimal course of action. Or, if you have an e-commerce platform, then Propensity insights can be used to generate real-time personalized offers, with customer lifetime value used to tailor pricing.
The future of retail is Customer Centric. Powered by Machine Learning.
