Customer churn accounts for one of the most expensive business losses. Attracting new customers involves substantial marketing costs, compared to retaining existing ones. Although the churn rate varies significantly by industry, from 5% for SaaS companies up to 38% for telecom organizations, all companies aim to reduce these rates as much as possible by identifying the underlying causes while offering compensations and incentives.
Each customer has various motivations for not using a specific product or service anymore. These could include the price, getting a better offer, forgetting to renew the subscription, moving into a different stage in their life where they no longer need it, and much more. Dissatisfaction is also a major trigger for high churn rates. Yet, it’s possible to curb this negative with the use of artificial neural networks. Let’s take a closer look at how it applies in the marketing realm.
How Does It Work?
First, the company makes a list with all the possible factors which could impact churn rates. These include demographics (age, gender, location) and sociographic data like preferences, behavioral patterns, risk measures, estimated earnings, etc.
Service usage data is also extremely relevant, as some could drop out because they are not using the service and thus can get no value from it. Another scenario is when they are using it too frequently, as in the case of sports betting, as we can see in this case.
Next, the company gathers data on these criteria, or, more likely, just retrieves the information from the company’s logs. Once it has a large enough volume of records, it splits the data into a training set and a calibration set. There is no generally accepted rule, but the proportion has to be considerably higher for the training part, about 4 to 1.
The results will show the probabilities of churning for each customer. With this information, the marketing department can create relevant strategies correlated with the respective customer’s stage in the life cycle. They could offer promotions, additional value-added services, or discounts in recognition of the years spent with the company.
Inside the Artificial Neural Network
The most significant advantage of using neural networks is that these learn in non-linear ways. This translates to the fact that these systems spot trends and associations that are not obvious, but might be crucial for solving the problem. In the case of predicting churn rates, characteristics such as the frequency of the communication between the company and the customer or the number of friends which have disconnected from the service might not be the apparent causes of churn, but a closer analysis might uncover the opposite.
There are several ways of building a churn prediction model, which have been synthesized in different studies. One possibility is to use logistic regression, to predicts if the churn occurs. This is also called survival analysis, and the result is the probability for each of the states.
Another option is to use decision trees, which divide the total universe into disjunct sets. These, in turn, can be sub-divided into disjunct sub-sets, for example, churn vs. non-churn. In the churn set, we can see churn due to a high price, an unfriendly interface, or other reasons.
More specifically, the best neural networks for predicting customer churn are recurrent neural networks (RNN). These are recommended for data where the points are not independent of each other. In the case of customers, it is evident that other participants may heavily influence individual decisions.
RNNs are also useful for natural language processing. When it comes to listening to customer voice, it makes sense apply RNNs to look at reviews, comments and emails, where each word needs to be considered in context.
Possible Challenges of Artificial Neural Networks for Customer Churn
There are a few notable obstacles in training neural networks which need to be taken into consideration when using this solution. Data quality is out of the question, as there are extensive studies on that and it boils down to garbage in, garbage out. More factors should be counted in.
One noteworthy problem is the depth of the network. The more layers a system has, the higher the possibility of the generalization error. This happens when the network doesn’t learn any more from new data, although it hasn’t yet understood every aspect.
This technology is quite new, and although it can come at a price, it’s one of the best potential solutions to such non-trivial task as customer churn prediction.
First, it can highlight problems a company didn’t even know they had. For example, you could see that all customers who terminate their contracts spoke with a specific call center agent, or they have similar issues with certain equipment, product or service.
Next, it offers the opportunity to save business-critical costs. This is possible as the AI system provides information on two different levels. It identifies those customers who have a clear intention of terminating the contract, and further divides that set into sub-sets based on profitability. This gives a clear picture of which customer relationships are worth investing in further and which are just costing the organization more money than they bring.
Artificial neural networks can easily identify clusters of customers. Adding information retrieved from social media can uncover additional relationships which could be useful in reducing the churn rate. This happens because social connections are critical in the buying decision. The need for social validation can be the main trigger behind a purchasing decision. If this is the case, as soon as a customer’s social group moves on toward the next cool thing, the company can expect high churn rates. Even simple discussions between friends, neighbors or colleagues constitute a base for canceling a subscription, for example.
Is This the Future?
The advantage of artificial neural networks is that it can help organizations save money by focusing on those customers who offer a higher ROI. It is also a valuable tool during budget and marketing planning. If you know which customers are likely to churn, you can create anti-churn campaigns and factor them in the budget.
Artificial neural networks are more than a new tool; it’s a change of paradigm, as it makes companies move from firefighting customer loss to planning and having the right strategies in place for each situation, complete with cost estimation.