Your browser is not supported. Please update it.

5 July 2018

Spideo, NPAW dive into churn prediction pointing to psychographics

French recommendation engine developer Spideo came out of hibernation recently, hosting a joint webinar with QoE expert Nice People at Work (NPAW) in which the two companies proposed the idea of psychographic data usurping demographic data in the battle for churn prediction products.

Recommendation software firms as we once knew them are rapidly rebranding to metadata companies, most claiming to use machine learning techniques to power various processes worlds away from purely recommending content – yet the underlying algorithms themselves seemingly remain much the same.

Psychographic data is about building up a personality for each individual user, based on aspects such as lifestyle, interests and habits, rather than relying on demographics and past viewing behaviors. Spideo and NPAW suggest psychographic data has more inherent value, allowing service providers to predict tastes and refine user segments using microtargeting techniques.

“Demographic data is still being used in ad campaigns but it’s bad at predicting users’ interests,” noted the Counter Churn by Understanding Users’ Internal Motivations webinar, in which 55% of attendees said their companies do not collect enough user data.

Demographic data is not being replaced by psychographic data in our view, rather entire systems are being enhanced by pumping in additional metadata, as well as geographical data and more. This, in turn, is feeding machine learning algorithms, a term Spideo and NPAW did not drop until the very end of the webinar.

It seems Faultline Online Reporter is right on trend, reporting just two weeks ago that psychographic data is an old term being revitalized by content data companies by unlocking new subtle differences within audience data and therefore more opportunities in advertising. Cost reductions in the video viewer data sector have triggered a resurgence of sorts for the term psychographic. Crucially, the worth of psychographic to operators and OTT video service providers is unproven at best, laughable at worse.

That has no bearing on the analytics data that NPAW delivers, only on the judgement and testing criteria of Spideo.

The two companies reeled off an interesting line up of stats and QoE figures in areas of video streaming which the two companies are focused on improving. The likelihood for churn is significantly increased when play failure rate exceeds 3%, buffer ratio exceeds 8%, join time takes over 4 seconds, and EBVS (exit before video starts) rate is above 8%, according to data from the two firms. On the other hand, optimum customer retention rate is achieved at the following – play failures 0-1%, buffer ratio 0-2%, join time of 4 seconds, and EBVS 0-3%.

Interestingly, adding just one interruption increases churn by 2%, highlighting the difficulty of maintaining viewer loyalty while relying on ad-supported business models. Spideo and NPAW therefore advise investing in getting buffer ratio below 1% to increase play time by about 2.3 times, in turn meaning more leeway for placing ads without frustrating viewers. The webinar warned of avoiding the danger zone where join time creeps over 20 seconds, causing some 40% of audiences to fall away.

Another poll carried out during the webinar asked: “Do you feel your user insights are actionable?” – to which a confident 0% answered “No”, 44% “Yes” and 56% “Somewhat”. You would need a score above 95% to make an impression on a pay TV operator churn, we suspect.

At this point during the webinar, Spideo flaunted a new personalization feature the company has branded the Spideo Battle Module, for enabling user onboarding by introducing content and products. A screenshot provided of the Battle Module during the webinar looked something akin to dating app Tinder, in terms of suggesting two titles to viewers from which they can either select which they prefer, select both or select none. Presumably these are previously viewed titles, unless the system is providing a brief synopsis of each and users then choose a preference based on this. However getting customers to score content is notoriously difficult.

It is this type of method that forms the foundations for creating semantic user profiles based on natural language and psychographic data. In this manner, past viewing habits can establish emotional contexts within various titles to build a profile, as opposed to creating a profile based on genres and categories – for example a viewer might enjoy content where “Hope” is a key emotional undertone.

The Battle Module is designed to be a “playful” tool but most users probably won’t want to put in groundwork to receive recommendations. Yet the feature is a neat idea for quickly personalizing an experience for new users to avoid interest immediately tailing off, as Spideo confessed machine learning algorithms for churn prediction typically need user data stretching back 60 days. Spideo recently won a deployment for its Battle Module at VoD streaming platform Anime Digital Network (ADN).

“We are right 90% of the time for churn prediction, the other 10% is related to content which is where we are focused,” said Spideo.