Affiliate turnover poses an ongoing challenge to the sustainability of healthcare organizations and systems. High dropout rates are a serious problem and are directly correlated with loss of income and increase in customer acquisition cost (CAC), as adding a new affiliate is around five times more expensive than retaining an existing one.
The depth of the problem and the pressing need to stop client rotation means that organizations can no longer postpone the question: “Why do partners stop using our services?” Is this issue also a concern in your organization?
It’s certainly concerning, considering that "After just one negative experience, more than half (61%) of the customers say they would move to a competitor; this number rises to 76% in the case of multiple negative experiences," according to 2023 statistics from LawnStarter.
The good news is that digital transformation opens new opportunities to deal more efficiently with this issue. Today, the potential of artificial intelligence (AI), machine learning (ML) and data mining can be harnessed to predict affiliate attrition. How? Among other things, solutions based on these technologies make it possible to examine the past behaviors of affiliates (among other variables), using these observations to determine what their future actions may be and identify those who are at risk of emigrating. On the other hand, by creating mathematical models (algorithms) that support the solution, rapid proofs of concept can be carried out to put solutions into action, measuring and comparing their performance and predictive capacity.
Attrition anticipation
Indeed, AI-based solutions allow us to analyze data for insights that we otherwise may not notice. As such, health sector organizations can anticipate the loss of affiliates and design strategies to retain them.
By analyzing data from current and past members, hidden behavior patterns and correlations can be inferred, which may explain the factors that caused the flight of those who left the system or opted for the services of another organization. Achieving this kind of insight would be nearly impossible through manual analysis, as they would have to correlate thousands of data points by hand. But, by using computational power and ML algorithms, historical data makes it possible to accurately predict future attrition.
Before we go any further, we should define with more precision what machine learning or AI machine learning is. Essentially, it’s a subset of artificial intelligence techniques that give software the ability to learn without human intervention or assistance and without previously establishing the rules of the program. These are techniques that allow the design of predictive models. And in turn, predictive modeling provides vital information to manage, among other things, the pressing problem that eats up potential business results and profitability: the constant flight of members.
How is this accomplished? We have already seen that based on the historical data collected, it is possible to analyze the behavioral characteristics correlated or linked to affiliates. By obtaining key information in advance, that loss can be proactively prevented.
Churn analytics
Occasionally, churn analytics (also called customer/affiliate/client attrition) solutions offer real-time information about affiliate groups and their likelihood of leaving. They also help organizations analyze behavior trends to determine the “breaking point” at which an affiliate (or affiliate group) decides to leave. Then, based on this information, the institution will be in a better position to deploy retention strategies, which may include, for example, modernizing their marketing campaigns or deploying personalized efforts to reduce the probability of attrition.
These solutions not only help to identify those who aren’t happy with the response provided by the health organization, but also to learn more about weak points of service or disadvantages in the pricing scheme.
Obviously, in order to obtain this precious information and achieve powerful predictive capabilities about disgruntled affiliates, it will be necessary to have a generous and comprehensive data set in qualitative terms, which allows for deepening the interaction with the customer and their motivations. Afterward, it is important to develop a predictive model of ML that’s as comprehensive as possible.
Proof of concept
No “customer churn” model can be extrapolated to any scenario or company, as it’s closely related to organization-specific variables. Therefore, it’s necessary to carry out an analysis from the domain of each business and its unique context.
But this does not represent any drawback, because under this model, health organizations can quickly do a proof of concept to see which model best suits their needs for predictive detection of customer churn.
The prior proof of concept under “Lab as a Service” concept is carried out independently from the software architecture in which it could be deployed in the future. This is so that it can be integrated into each company’s specific landscape and systems.
Among other topics, the POC must discover which variables approximate the reasons for attrition by analyzing predictive capacities and the weight of each random variable, which support a strategy for decision-making on commercial actions of retention.
The method used is an iterative and recursive process to explore large volumes of structured and unstructured information, find hidden patterns, and determine correlations between them. The focus is to find valuable insights for business decision making. In this sense, data mining algorithms allow for the discovery of patterns—essentially local structures that provide information about a space restricted by variables. Thus, you iterate to find the right model by improving your understanding of input datasets.
The decision about the supervised ML models will be defined during the analysis stage, but based on experience with similar classification problems, the following may be considered:
- Decision trees
- Vector machines (SVM)
- Random forest
- Gradient enhancement
- Logistic regression
- Naïve Bayes
- Neural networks
For each of the algorithms used, a quantitative analysis is made based on test dataset predictions. What is sought is to evaluate its predictive effectiveness for the problem posed. For this, score visualization tools are used. Then, a comparison is made between all used algorithms.
Now, since there is an infinite ocean of data and combinations, the solution must be limited in objectives and time, considering iterative improvements over the actual measurements in the future. For this reason, we propose a service model associated with a target outcome (for example, reducing churn by 10% in the next quarter).
Implementation keys
Predicting member attrition through ML helps you identify risk cases and understand why they are willing to leave. Attaining evidence of the “why?” will help your organization to take actions that prevent the attrition of members most prone to leave. That is, you will be able to work on your retention rate and significantly improve overall performance.
But how does churn analytics work and how is it implemented? An automated process continuously collects information on member behavior (procedures, claims, activities, etc.), as well as voice recordings and telephone conversations. Through sentiment analysis from calls, complaints and conversations, key information can be added to the predictive model for affiliate flight.
Thus, when initiating contact in the service portal, the predictive model will indicate if the member has a “risk of leaving” alert for follow-up. Additionally, this solution can be integrated with messaging to perform automatic actions for the retention of the member.
Fundamental evolution
In the past, predicting whether a member would stop using prepaid services, for example, was based exclusively on commercial rules imbued with market knowledge. In turn, these rules were based on managers’ experiences—and not on exhaustive data analysis.
Now, we can take advantage of the vast amounts of dated collected and empower machine learning models to automatically learn those rules. ML techniques serve precisely to find patterns and relationships hidden within large amounts of data, and the rules discovered by their models are guaranteed through evidence—no longer through mere intuitions or hunches of management. In addition, ML algorithms can process and extract patterns from many variables, resulting in more complex and thorough rules.
The advantages of using churn analytics solutions with ML are multiple and powerful, but there are two that stand out above the rest:
- Help businesses grow by re-engaging and retaining members who would otherwise likely leave. ML allows you to apply the knowledge before it’s too late.
- With good quality historical data from members, ML algorithms can learn very precise rules in much shorter times. These rules can also be adapted to changes that may occur over time.
By obtaining early information about members at risk of leaving or switching providers, your organization may be able to provide some additional benefits or pay special attention to them.
In other words, by predicting which of them are likely to leave and why, you can take action and retain them. If retention measures are well thought out, they will have an opportune effect and your organization will continue to maintain health as the #1 priority, both for affiliates and for the income statement.
