Angela is a call-center assistant in an insurance company, who has been working from home since the beginning of the COVID-19 pandemic. In the first couple of months, she struggled with her internet connection, as her tariff plan didn’t cover all her needs. She tried to reach the technical support of her current provider, but the lines were too busy due to the increased demand as everyone in Angela’s area moved to home offices. After two or three unsuccessful calls and unanswered e-mails, she gave up on the idea to improve her situation and switched to another vendor, that offered a readily available plan that satisfied her needs. There are thousands of households that underwent the same issues as Angela, according to the Capgemini research institute. There are dozens of companies that lost their customers due to a million tiny issues or a couple of big ones.
The situation might have been different if one of the customer support representatives (CSRs) knew about their clients’ struggles and could act on it in time. Of course, it is impossible for personnel to keep track of every single end-user and their current relationship with the company. It can be done, however, using the right technology.
In this article, I would like to share the recent case from my company, MaxBill, of developing a churn prediction model for one of our partners. Its success rate turned out to have 97% accuracy and it has two major points of usage that will benefit customers.
Single Model Combining Several Algorithms
There are multiple methods of data collection and subsequent assessment. We combined a number of different algorithms in an ensemble methodology for the particular client. Such a model includes:
- DecisionTreeClassifier,
- KNeighborsClassifier,
- LogisticRegression,
- RandomForest Classifier,
- VotingClassifier.
The results of a control run on historical data show 97% accuracy of churn prediction. We plan to enrich the model with the additional data artefacts and use it to determine the customers, which are at risk. Consequently, they can further be included in the retention process.
Customer Retention Use Cases
Creating a model would be useless without a clear understanding of the particular scenarios it would be beneficial for.
Per-Customer Basis
When connected to the customer summary in our CRM the churn prediction engine will give the customer service representative insight into the particular client’s probability to end their relationship with the company. In such a case, any CSR can assert the pro-active role and cater to the client in a way that would lower their churn risk.
Bulk Report
A report will be generated on specified dates highlighting the list of customers that are in the risk group. This data can be used for customer retention marketing campaigns that will further lower the risk of customer churn, engage customers with the service provider and increase loyalty.
The model will not only expose the customers that need to be paid more attention to — it is going to be interactive, allowing the management to apply new tariffs, update existing product pricing, offer discounts, special promotions and see the projected consequences of their actions based on new variables in the model that are added. This tool will be very handy for creating specialized offers for particular customer groups and finding the best balance between marketing ROI and customer churn while maintaining a low false-positive ratio.
Needless to say that all of these processes can be automated. As soon as the system predicts that a customer or a customer group is under the risk of churn — the preferred churn prevention actions can be taken without the need for human intervention.
The Future of Customer Care
As shown by the Deloitte analysis, this year, we’re seeing an unprecedented opportunity for multi-play providers to grow their customer base, area of influence and service packages. However, many companies have found that their back-office lacks scalability. One of the many pain points of the majority of providers is the heightened pressure on the customer service representatives. Due to the COVID-19 pandemic, increasing the number of personnel is rarely possible. At the same time, another side effect of the pandemic causes the demand for their services to grow. For example, some tenants experience hardship with getting good connectivity in rural areas, a problem that also has to be answered by the providers.
To keep up with the fast-growing opportunities, companies have to both improve the customer care experience for their client base and be able to predict and avoid churn.
The UX can be enhanced by encapsulating omnichannel communication (Facebook Messenger, Viber, WhatsApp, Telegram) in service flows, automating the customer handling processes through self-service in the SelfCare Portal, deploying personal assistants with conversational AI, possessing NLP capabilities and integrating RPA bots, that can perform both front-end and back-end tasks for the user.
To fulfil the majority of the tasks, that lay before the company, machine learning can be successfully utilized. For each specific case, a particular combination of algorithms can be chosen, trained, tested and implemented in different processes. With its help, businesses can enhance their retention abilities, and expand their client base much faster in a more cost-effective way.
Thanks to Konstantin Dolgushin, Vladimir Penkov, Val Morozov, and Rost Bitterlikh for their valuable insights and help with this article.
