A churn model makes it possible to identify customer’s intentions to leave and thus intervene promptly with appropriate actions, thereby improving the company’s capacity to retain loyalty and limiting potential losses.
But how exactly do the mathematical algorithms and analytical software underlying the customer retention process work?
Case studies: when are churn models needed?
As an important premise, it should be noted that churn analyses are typically used by companies that sell services based on subscriptions (for example, Netflix), delivered over time (such as electricity supplies or telephone contracts) or linked to donations (in the case of non-profit organisations). They are used less in one-off product sales, when the relationship between company and customer ends with the act of purchasing. However, the phenomenon of servitisation (which includes the offer of digital services in addition to intelligent objects) is changing the rules of the game and opening up new scenarios of use.
The pioneers of churn analysis are telephone companies, which have a large quantity of customer data at their disposal, starting from the information provided for concluding the contract or regarding the use of the service (duration of calls, preferential times and so on). This data may not appear particularly significant when examined individually, but when aggregated, they provide the possibility of detecting behaviour patterns for various types of users. For example, it is possible to recognise and identify: customers who are always attentive to the latest news on the market and are inclined to switch to the competition when cheaper rates or additional services are offered; less impulsive users, who prefer to avoid the effort that a change of service provider entails; and dissatisfied people, who have repeatedly expressed their disappointment to customer care.
When developing a churn model, a very important aspect to note is the time period beyond which the customer intends to give up the service. The time horizon for stemming the risk of switching by implementing commercial and retention strategies varies according to the sector and type of service offered. For example, in the case of utilities, weeks are needed to convince the customer not to switch to competitors (discovering the intention only a few days beforehand does not allow intervention with appropriate actions), whereas telephone companies can act with less notice (they do not need algorithms sophisticated enough to counter the risk of switching months in advance).
Data origin and analytical purposes
In a churn model development project, the information on the customer base that contributes to the analysis is provided directly by the company and comes from multiple sources: contractual conditions, personal data, consumption, frequency of donations, billing (for example, payment methods and regularity), interactions with customer service, changes to the competitors’ offer and so on.
The goal is to outline the customer’s identikit and try to counter any intentions to switch, prepare and implement retention strategies, and identify the quality profiles that the company has an interest in retaining or acquiring (for example, because they are more loyal and pay on time).
Prescriptive analysis (which not only makes it possible to distinguish types of customers, but also to suggest appropriate retention strategies and automate the initiation of certain actions) is the ultimate goal of churn models and what many companies hope for. However, it is not possible to make fully automatic prescriptive models unless there is a long history. In fact, an extended data history is needed for statistical algorithms to work with sufficient reliability.
Moreover, automation of retention activities is suitable for initiatives that involve a sizable portion of the customer base: for example, sending newsletters, which even if performed on a large scale has low costs.
For more targeted and incisive strategies, such as offering discount coupons and rate changes, companies prefer using algorithms for providing targeted insights, together with the business experience of their staff.
From algorithm to software
When a business consults a company specialised in developing churn models, it must evaluate its skills on two fronts: mathematics (for the development of algorithms aimed at investigating and solving specific needs) and information technology (so that the algorithms are available to users in software with dashboard and advanced analytics functions).
In most cases, churn models are included among the components of larger analytical platforms that cover a broad spectrum of needs and use cases.
These solutions usually offer a number of summary dashboards, which provide a view on macro trends to the people in the business, and the possibility for more advanced users to examine the details and take advantage of advanced analysis functions based on customised mathematical models. An advantage to be considered when choosing the platform is the ability to connect to dashboarding and data visualisation systems already existing in the company, which are then powered by new algorithms and mathematical models.
In summary, thanks to the development of ad hoc IT tools and analytical functions, it has become possible to develop effective churn models, capable of predicting and minimising the risk of switching by activating strategies for customer retention