Conviva reveals more AI flesh in customer seduction bid

A whitepaper from OTT analytics company Conviva has unveiled a few more details behind the company’s recent foray into AI following the launch of its AI Alerts product in September – bucking the usual trend whereby companies are reluctant to reveal specific ingredients in their AI secret sauces. This extra marketing push also hints that converting its existing customers over to the fledgling artificial intelligence side of things might not be proving so easy, given the convoluted minefield of AI buzzwords, although we feel Conviva has done another decent job of explaining its technologies.

It told us this week that this development is now in production for most of its customers, but at present can only reference HBO which is using AI Alerts for HBO Go and HBO Now.

Applying AI to OTT video analytics aims to end the moot point that analytics companies are stuck between a rock and a hard place, in that simply identifying a problem is no longer satisfactory on its own. The technology must take the next step to predicting an issue before it occurs and then rectify it before video quality is affected. Achieving this poses problems, however, largely due to the proliferation of publishers’ video assets, spanning content management systems, origin servers, CDNs, a multitude of consumer devices, and more – amounting to more than 100 million different combinations of possible issues.

“Defining thresholds and manual configuration of alerts for all possible business-critical conditions is physically impossible due to the sheer number of possibilities”, according to the Conviva whitepaper. This statement suggests a conflict of interests, but also shows a display of acceptance that making the next leap in analytics is a major ask, instead of making wild claims that analytics can solve all the world’s video problems in one wave of its wand.

Conviva explains how its proprietary machine learning algorithm can discover patterns within individual metrics such as rebuffering video and video start failures (VSF)– to pinpoint anomalies and similarities in a publisher’s video streams. It then applies root cause analysis once anomalies have been detected, claiming to reduce the time and manpower required to solve video stream quality issues; not quite predictive detection and eradication, but baby steps towards that ever-desirable future.

While Conviva and similar companies do not provide the specific tools to repair all buffering, VSF and wider network issues themselves, the diagnostic reports do provide valuable insights that can lead to an eventual fix, but perhaps one day detection and repair could merge as one product thanks to further development of AI.

Conviva says it developed this method of detection by capturing millions of QoE and engagement events from millions of devices, using this to develop technology that could model all the possible QoE dimensions that make up the viewing experience.

Many vendors in this space have access to millions of metrics in millions of devices, so what is stopping other analytics firms from making these advancements too? Receiving $40 million in funding this summer will certainly have helped Conviva get the ball rolling in that department, and the company told us back in June that the algorithms behind its new AI product have actually been in development for more than four years – so maybe others are just playing catch up for now.

Verimatrix is one company chasing down Conviva’s dominance in QoE OTT analytics, although the security expert is yet to push the idea of machine learning algorithms in its analytics products.

Conviva says after a pattern or anomaly has been identified in video metrics, it can drill down into each event and unpack all of the dimensional factors affecting quality, as well as provide data intelligence to take it a step further and proactively scan the data and uncover issues, including those not currently on a customer’s roadmap.

Customer wins for AI Alerts other than HBO have not been publicized, but it has provided a case study outlining how a small OTT provider used the Conviva Video AI Alert tool to pinpoint a video start time issue to a faulty router that was acting as a proxy server for iPhone users.

Conviva’s CMO Ed Haslam told Faultline Online Reporter earlier this year that the company’s R&D is focusing on transfer learning, a method of machine learning with an additional source of information apart from the standard training data, whereby it removes the isolated tasks which come with traditional machine learning algorithms by developing methods to transfer learning on one or more source tasks, therefore improving learning in a related target task. Essentially this works in a similar way to how humans can apply learned knowledge to a variety of different tasks.

Alerts is powered by Conviva’s Video AI platform and Video Graph, claiming to be generated through the analysis of billions of video streams by over 2.5 billion unique sensors embedded within video players across its publisher network. The report adds that it measures more than 25 billion video streams a year, using its machine learning algorithm to process engagement and experience data for video applications in more than 180 countries.