« Back to Posts

Reducing Churn with TV Data Analytics

We knew our client had a churn problem with their OTT service when only 35% of new customers used the service three months after they signed up. They asked our data science consulting team to look at the TV data and quality-of-service information they were collecting to drive a strategy to improve service retention.

Most OTT services struggle with retention when they launch, and the metric of 35% retained users is similar to a lot of our customers. However, it was apparent that the operator’s issue was partly a problem of its own making: they had fallen into the trap of applying old Pay TV logic to a new OTT service.

Rather than starting from scratch, they’d taken their existing retention process and had only loosely adapted for OTT. It’s easy to take into account a few differences between the two, such as cancellation windows, but we’ve seen time and again that this approach doesn’t work. OTT is fundamentally different to OTT and operators need an entirely new playbook to retain their customers.

Ripping up the rules and starting again

Our data science consulting team took the customer through our standard procedure for retention projects. We ingested six months’ worth of OTT TV data (consumption and quality of service) and mapped it to CRM data, and some first-party data that the client had access to. We excluded data which we knew was subject to data quality issues.

Our data science consulting team carried out a series of reporting and modeling of customers that had churned and customers that had not, and analyzed the essential characteristics and differences between the two. They found critical identifiers of churn amongst the OTT base including content, billing, and quality factors.

The team then built a predictive model that helped to identify customers who were most likely to churn, and worked on an overall strategy to reduce churn within the business, including the integration of our model into existing operational systems, and training the customer’s analytics team to run and iterate the model themselves.

Finally, we created a series of dashboards to demonstrate and monitor some of the churn triggers we wanted to show the customer.

Getting to the nub of the problem

The reporting and modeling revealed some easy wins that the customer could quickly implement. One was changing the time frame around which the business looked at churn. With its core TV service, the customer’s retention strategy worked by identifying potential churners 12 weeks before they disconnected the service. It was quickly evident that OTT customers were more impulsive in deciding to churn, so we built the model around this shorter time frame.

We also added credit card expiration to the factors that exhibited churn in the model, as this was a cause of far more service disconnections than was the case for the customer’s core TV service.

We undertook programme-level analysis to understand which content indexed particularly well on the OTT service. This uncovered niche programming that performed well on the platform, but whose strong performance was masked by hit programming that inevitably had more views. One such example was wrestling, which, while having a small audience, had a very dedicated one for OTT, and indexed highly on the service

Genres from TV Data

Some factors in the model were down-weighted. The operator had thought that customers who watched a lot of live content were less likely to churn. In fact, this data was biased by viewers watching a few major seasonal events, and some customers who took the service to view these events were more likely to churn, rather than less.

We also integrated quality of service data into the model. The company had used call center records and faults as a proxy for quality of service. We introduced real QoS data and analyzed the point at which a customer was more likely to churn if a particular element – buffering, start-up time, hit a certain level

Giving subscribers some love

Our work contributed to a 10-percentage point improvement in retention rates for the OTT service. OTT TV data from the model we built has been integrated into the customer’s outbound communications systems, allowing the customer to react more quickly to signs that a subscriber may be leaving.
Whenever a subscriber trials the OTT service, they are placed in one of several segments and marketed to accordingly.

We can’t take full credit for the improvement in the retention rate, but the client is convinced that data science is a critical factor in their success and is now working to personalize further what each subscriber receives and to send them even more relevant content and offer triggers.