Propensity Modelling - Can We Predict The Future?
Wednesday January 18, 2017
Data analysis in sport and leisure has evolved significantly over the past decade. Gone are the days of mass marketing a product to your full database because it has historically been a best-seller or simply bombarding every fan with the full club catalogue in hope that one item will catch their eye.
Segmentation and other statistical techniques have allowed companies to profile contacts to understand their spending behaviours, but are we at a stage where we can we predict what products fans and customers will purchase before they do? How accurate can facts, figures and mathematics be to predict something as uncertain (and as irrational) as human decision making?
It all comes down to the laws of averages. For every group of season ticket holders predicted to renew their ticket for example, there will always be at least one who doesn’t. It is quite often that no amount of data could have predicted the shift in behaviour of this non-conformer, but if the majority of fans follow the predicted trend, is that not a measure worth taking advantage of?
The most accurate statistical models will collect as much information as possible about the target audience to understand why a customer will behave differently to others and use these changes in behaviour to improve future models. Thus, outliers will become less of an unwanted surprise which disprove models (and upset statisticians) and more of an opportunity to enrich datasets and further understand the customer base.
As mentioned, the key to creating the most accurate models is the quantity (and quality) of the data used to understand the target audience and to predict behaviours. It is difficult to build meaningful segments of customers based on a small number of attributes. Remember, correlation does not imply causation. Just because men purchase more football shirts than women, it doesn’t mean you should only target football shirt advertising towards men. In this example, an accurate propensity model will consider as many attributes of football shirt purchasers as possible, test the significance of each attribute and then produce a score for the entire customer base using only the significant attributes. So instead of marketing a product to a set of customers based upon one or two variables, while excluding the rest of the customer base and assuming the product will sell to those advertised to, companies can predict the probability of every customer purchasing a certain product based upon several variables and tailor their marketing strategies accordingly to maximise revenue.
Often sports & leisure clients look to propensity modelling to predict sales of merchandise or one off general ticket sales. However, we have seen that the most valuable application of these models is by predicting new membership or season ticket sales. It is these sales which not only generate the most revenue for clients, but also highlight the most invested fans or customers to upsell additional products.
By understanding key transactional metrics such as how recent transactions were made, how often they are made and how much is spent (e.g. traditional RFM modelling), alongside customer attributes such as age, distance from the venue and email engagement rate, our analysts have produce accurate and thorough propensity models. From these metrics, prospective new members/season ticket holders have been targeted with tailored marketing communications.
A leading example of where this methodology has been successfully applied is with one of our major European football club partners. By using propensity modelling, we segmented the clubs fans based on their likelihood to purchase season tickets and tailored the approach to marketing accordingly.
For example, ‘Hot’ prospects, those fans whose overall propensity score identified them as significantly more likely to purchase a season ticket, were targeted with a multi-channel approach which included a phone call from the club.
As a result, the conversion rate of the hottest prospects increased from the average of 2.5% to 10%, with over £400,000 revenue generated from new season tickets (not renewals) and subsequently more through the up sale of additional products.
So, can we predict what products a fan or customer will purchase before they do? Not with 100% certainty. However, through accurate propensity modelling and the right marketing campaign, we can certainly maximise revenue based upon our predictions.
To learn more about our Propensity modelling, and other Data and Insight services, please contact us via our enquiry form.