Churn reduction is one of the most common ways to make it happen. Churn is loosely defined as the percentage of users that leave a telco each month. Any ability to drive these churn percentages down improves a telco’s bottom line, so the concept of churn reduction itself is certainly nothing new to telcos.
Nevertheless, the ability to use real-time data as a potential resource for managing churn reduction is still a relatively new concept. Machine-learning-based churn models can now predict a potential churn customer with 65% to 85% accuracy. Yet, typically only 5% to 15% of these high-risk customers are likely to accept the telco’s retention offer—leaving millions of dollars on the table. So, why the discrepancy?
The Problem with Yesterday’s News
Technology is now available to help telcos identify and turn high-risk users into satisfied customers. As long as churn-reduction campaigns still rely on after-the-fact data, however, retention-success rates are likely to remain disappointingly low.
In most scenarios today, churn-reduction models depend on call centre logs, online interactions, complaints, and other historical data to identify and target customers to understand the drivers of potential churn.
By working with historical data events, however, at-risk customers have already had a bad experience. Once a customer is overtly dissatisfied, it’s far more challenging to resolve a problem than being spotted and addressed beforehand.
And it definitely can be.
Most customer-experience problems that result in churn are caused by three distinct issues: dropouts, connection, and speed. This data can be incredibly difficult to source because it typically resides deep within the network infrastructure and the customer-premise equipment.
Moreover, the size and velocity of this data are often prohibitive for traditional IT architectures to ingest in a timeframe that is useful for proactive churn-prediction models. Truly proactive models not only need to process such predictive data seamlessly, but also provide churn-probability scores in real-time.
Performance-based data collected from deep within the network and the CPE is exactly what is needed to inform and improve the overall customer experience.
The Main Event
Event-driven platforms capable of ingesting millions of messages per second are up to the task of extracting relevant events from network logs, customer devices, and other legacy systems. These platforms use the same technologies that giant tech firms, like Facebook, LinkedIn and Uber, rely on to manage the billions of ongoing messages between their users and applications.
Rather than focusing on what an individual customer has done in the past, telcos can use an event-driven model to monitor and measure how likely customers feel about their ongoing, day-to-day experiences. Answers to event-based questions that impact the likelihood of churn include:
- Are customers getting the upload and download speeds they expect?
- Are they experiencing dropouts?
- Has the line connecting their service deteriorated?
- Are customers overloading their network with too many connections?
- Is the modem routinely rebooting or resynching?
- Are there correlations between dropouts and a vendor firmware upgrade?
The questions can go on, but the common denominator with an event-driven model can answer them all, respond to them proactively, and improve the customer’s experience before, not after, a bad impression is made.
Telcos can’t do anything about involuntary churn events, such as customer relocations, deaths, and other non-manageable occurrences. Roughly 25% to 40% of all telco churn comes from these types of involuntary causes. The majority of churn-related causes, however, are both predictable and correctable. With an event-driven approach, telcos can improve their customer experience in real-time—and add millions to their bottom line—even in a saturated market.