Some say that marketing is both an art and a science. We’d say that it comes down harder on the science side – and that the effectiveness of model-driven marketing proves it.
If you’ve ever wanted to discover a new audience, capitalise on a new product, uncover the behaviours that drive your business’ success or learn how to eliminate customer churn, then a statistical model likely has the answers. It’s all a question of which type of model is the right one for you. That’s why we’ve broken down six of the most common types of analytical and predictive modelling for marketing, explained in real-world terms around how they can improve your company’s performance. But first, a quick note:
The importance of a commercial context
Predictive analytics for digital marketing can be an incredibly valuable capability for a business to develop – but it can also be very easy to lose sight of the commercial context and get bogged down in pursuit of the academics.
Here’s what we mean: let’s say your business is trying to predict the behaviour of a set of customers. A model is developed and, after some adjustments, has a 90% accuracy rate. In practical commercial terms, this model is fit for purpose – it is highly accurate and will inform the business in its marketing pursuits. However, the analysts behind this model decide 90% isn’t good enough. They continue to tweak and mould and adjust, spending 200 additional man hours on the model. The result? A model that is 91% accurate. Better, certainly, but far more resource-intensive than was necessary for commercial use.
Worse still, the new model has lost something along the way towards that 1% gain: it now lacks balance between predictive power (knowing what is going to happen) and explanatory power (knowing why it’s going to happen).
In this case, the business needs the model to have both, but this new, more accurate model has sacrificed the explanatory side. As a result, when the new model occasionally throws out an unintuitive prediction, the business is unable to figure out why. Marketers lose confidence in the power of the model, and the results, no matter how accurate, fall by the wayside. The commercial point of the model has been lost for the sake of a single percentage point increase in accuracy.
We’re mentioning this because it’s easy for any of the models described below to fall into this trap, especially when the modelling is performed by non-dedicated or inexperienced analytical teams. Even the ‘best’ model can fail to be fit-for-purpose if designed without the actual needs of the business in mind – whether that’s churn, acquisition, customer commitment or any of the other examples listed below.
In short, it’s important to work with business-focused analysts, no matter what model you decide to use. With that out of the way, let’s move on to the six models themselves:
1. Customer acquisition modelling
Marketing is not just about finding an audience - it’s about finding the right audience. Customer acquisition modelling revolves around identifying potential prospects who are likely to be “good customers”, as well as discovering traits that drive a higher likelihood of a population becoming customers. This allows marketers to better target their customer acquisition efforts, improving overall efficiency and ROI.
Example: A non-profit organisation has gathered a significant amount of personal information on potential donors through its website but is unsure which of these are most likely to convert into full donors. It uses a highly complex customer acquisition model, training it on historical data. The model provides a ranked list of contacts most likely to convert. Marketers then reach out to these specific contacts via email with a simple request to make a recurring donation each month.
2. Recommendation modelling
Recommendation modelling encompasses a suite of models designed to identify opportunities to improve the overall value of a given customer’s relationship through higher value or more diverse purchases. It includes upsell modelling, cross-sell modelling and next best offer modelling. The major differences between these models are the desired business outcomes:
- Upsell modelling deals with increasing purchase value. It relies on knowing which products are most valuable to the business and predicting and identifying which customers are most likely to purchase those products. Example: A retailer feeds customer information into a highly complex upsell model, which then ranks the customers most likely to purchase higher-value products. Marketers reach out to these customers individually, offering a discount on their next purchase of a specified high-profitability item.
- Cross-sell modelling deals with increasing purchase diversity. It relies on finding those customers who are most likely going to expand their purchasing behaviour to another subset of offerings. Example: A marketing firm feeds customer information into a highly complex cross-sell model, which then ranks the customers most likely to expand their product purchasing behaviour. Marketers reach out to these customers, offering a free trial of any of their other services.
- Next best offer modelling deals with encouraging overall purchasing value across the entire catalogue of offers. It relies on identifying which products or services are most likely to appeal to a set of customers and is generally more focused on customer needs. Example: A bank feeds customer information into a highly complex next best offer model, which associates customers with specific products they are likely to purchase. Marketers design a set of push notifications to be deployed through the online banking function to these customers, letting them know about the existence and benefits of their individually associated products.
3. Fast-track modelling
Fast-track modelling predicts which customers are most likely to become high-value clients over time. This is different to merely identifying high-potential customers overall, as it takes the influence of tenure into account as well - a cohort of new customers might not be high value now, but they could become so after a year or two. Once identified, these customers are then prioritised by marketing efforts designed to push them further down the value cycle.
Example: A credit card provider feeds the information of every new customer into a highly complex fast-track model, which determines the likelihood of this new customer eventually becoming high value. Should the new customer breach a given threshold of likelihood, they are set as a priority for marketing efforts. They are sent offers for lower interest rates over a fixed term sooner than other customers, encouraging them to make more regular, higher-value purchases – a behaviour which should persist after the fixed term concludes.
4. Churn modelling
Churn modelling is used to identify customers who are likely to stop doing business with a given company within a short amount of time. It is often used as a form of an early warning detection system, flagging high-risk customers that require outreach. An important point to note is that churn modelling is mostly used for companies that operate on a customer ‘binary’ - either a person is their customer, or they’re not. Power companies are a good example, as are telecommunications businesses (barring the minority of users who have more than one phone or internet connection).
Example: A power company uses churn modelling to identify current customers who are at a high risk of churning. Once flagged, these customers are prioritised for outreach marketing material, highlighting the benefits of staying with said company and offering a loyalty discount.
5. Slider modelling
Slider modelling is used to predict which customers are likely to “slide” towards doing business with a competitor, reducing a business’ share of wallet. It is similar to churn modelling, but where churn modelling deals with the ‘customer binary’, slider modelling is less discreet. It is the difference between discovering which customers are going to stop doing business with a company altogether (churn) compared to them preferring to do business with a competitor – or no one at all (slide). Companies that can ‘share’ customers with a competitor typically use slider modelling rather than churn modelling. Retailers are a common example.
Example: A petrol company uses slider modelling to identify which customers are likely to start spending less at their petrol stations and (presumably) more at competitors. The marketers reach out to these customers with fuel coupons to entice them back before their slide continues.
6. Customer commitment modelling
Customer commitment modelling aims to understand who the most loyal (and least loyal) customers are in an existing client base and rank them accordingly. It may also involve identifying the behaviours that are associated with higher levels of loyalty. This information can then be used to differentiate marketing messages depending on customer commitment, targeting the disposition and likelihood of each given segment more specifically. It could also be used to reward customers who are the most loyal, encouraging higher lifetime value and more word of mouth marketing.
Example: A bank uses customer commitment modelling to identify customers that are most likely to have higher levels of satisfaction with its products and services. Marketers reach out to these customers with requests for testimonials, using these positive social proofs on their website to convert new visitors.
Whether you're discovering new customers, retaining old ones or capitalising on loyalty, there's a lot that statistical modelling can do to drive a successful marketing campaign. But this is just the beginning of what predictive analytics could mean for your business - for more information, download the Datamine Guide to Predictive and AI Modelling below, or contact us here to chat through your modelling needs with one of our experts.