Faultline Online Reporter recently ran an article questioning if recommendation software had lost its X factor and, after a quick search of the Faultline archive, we were still bemused to discover that market frontrunner ThinkAnalytics has yet to feature in our publication so far this year. In our defense, the developer of the world’s most widely deployed recommendation engine hasn’t issued a noteworthy press release since winning a deployment at Voot back in September, so meeting ThinkAnalytics at NAB 2019 to discuss the next big thing in recommendations was of high priority last week.
Just a fortnight ago, our coverage of a recommendations research project between Altice Labs and Utelly mentioned how the general slowdown in coverage might imply the technology has peaked. We echoed this to ThinkAnalytics CTO Peter Doherty in Las Vegas, while adding how we see content recommendations becoming an increasingly integral pillar in a broader connected home ecosystem.
“Making things work is where we are. The industry needs to understand that making people change their behavior isn’t easy; viewers aren’t making these decisions on a whim,” said Doherty. “I’m not going to tell you exactly what I think the next big thing is as you’ll just go and write about it,” he jested. Fair point.
In a nut shell, Doherty mostly agreed with the suggestion that recommendation algorithms at their most basic and fundamental have probably peaked in a sense, yet having the ability to adapt and change, while adding capabilities to act upon user behavior, are fundamental pieces in the progression of the recommendations and personalization industry.
This is all good and well, but in order to “make things work” you have to understand what is not working and this where ThinkAnalytics says its BigData platform comes in. For example, in some customer use cases the technology could unearth something shocking to an SVoD platform provider such as how some 80% of their content library is not being viewed nearly regularly enough. But then what is causing this? Such a dramatically inefficient content selection would surely require a substantial revamp and helping a customer adapt is therefore key. For example, ThinkAnalytics would start off with A/B tests on some 5% of a user base before finding the magic solution and rolling it out fully across a footprint.
It goes without saying that ThinkAnalytics has been incrementally increasing its machine learning capabilities, using a combination of in-house and off-the-shelf algorithms and last week we dug a bit deeper. “Traditional data mining algorithms don’t always work but, in our experience, we know they work as we have 250 million users worldwide. You don’t need a brand new algorithm to predict churn when you have perfectly good off the shelf algorithms for getting the job done. For segmentation and such you can just choose the best algorithm for the data set you’ve got,” added Doherty.
So, while the UK software outfit has been distinctly quiet on the customer announcement front over the past six months, developments behind the scenes have included adding more natural expressions to its voice search products, including improved predictive churn analysis.
Doherty estimates onboarding an additional 50 million users since this time last year to reach the quarter of a billion milestone, during which time its technology went live at BBC iPlayer about 6 months ago, kicking out ContentWise in the process, and ThinkAnalytics has also picked up big deployment deals at US sports streaming service Dazn, Deutsche Telekom and Tata Sky, along with a handful of major deployments the company can’t speak about just yet.