How LightFlow by Epic Labs optimize OTTs workflows

LightFlow by Epic Labs is a machine learning engine focused on optimizing OTTs workflows by improving QoE (quality of experience) while reducing cost, depending on each company business KPIs and rules.

The OTT Challenge

Actually, every time a user presses play to consume a video over the internet there’s a lot of complexity that starts with the nature of the content itself: is it an action movie, sports, a documentary? Is it VoD or Live? Which device is being used? Where is the viewer?

Then, business rules come into play as well: Is your TV service based on subscription or advertising?
All of these things make every play unique. The context changes for every single playback.

Human based solutions.

There are several ways to cope this complexity. In most cases OTT operators deal with this with a one-size-fits-all approach. This is a very common solution nowadays but it’s not optimal. It attempts to tackle each step of the workflow with the most standard manner to guarantee it fits into any requirement. This is one way of dealing with the complexity, but it is far from ideal in terms of efficiency and quality of experience, because this approach relies on legacy codecs, a large number of by-default renditions and conventional CDN choice to ensure that content is consumed.

The multi-expert approach.

This is theorical exercise. We could potentially solve all the problems for any workflow if we hosted a large group of experts to decide the most appropriate way to optimize the experience. This means they would decide which type of resolution and encoding parameters are optimal for each content and every playback, and they could also monitor the audience regarding the amount of devices, formats, delivery, screen sizes, players, network conditions and all the things that really matter in the OTT environment.

Then, for each playback, they would recommend specific settings that should be applied on real time according to business KPIs. Obviously, although the outcome would be good, this solution is not feasible.

Machine based solution: LightFlow by Epic Labs

LightFlow is an AI stream optimizer engine that can take decisions based on the content itself and the whole playback context.

LightFlow can be fed with any data or video analytics, network conditions, devices and also publisher’s business KPIs.

This trained algorithm factors the playback context as input data making your analytics actionable, in real time. Consequently,

LightFlow performs actions into the workflow optimizing what it’s more important in your business, and for every unique playback.

LightFlow uses a feedback loop to further improve results over time.

Content encoding and preparation

Let’s focus on content preparation. LightFlow has been trained with millions of hours of video and different encoding options. As a result, just by making a very fast analysis, the algorithm is capable of predicting the quality of the output videos simulating thousands of permutations of encoding parameters, and choosing the optimal combination among them.

This process of inferring the best encoding parameters occurs for every single video asset analyzed and is applied individually to produce the optimal bitrate ladder as well as other parameters.

As a result of that, LightFlow can reduce up to 40% of bitrate without any appreciable quality degradation.

The devices

We also analyze what is happening on the users’ devices. Once the content has been optimized as previously described, when it is delivered to users, LightFlow captures users’ behavior, including type of device and network conditions. The device itself triggers smart decision on the delivery side, such as maximum bitrate (avoiding wasting bandwidth for small screens) and most efficient codec supported.

Based on the user’s device, LightFlow may decide whether H264, HEVC or VP9 codecs are the most adequate. Finally, based on the overall users base, different copies of the original content may be created, using a hybrid multi-codec ladder only if quality and savings efficiencies outnumber additional computing and storage costs associated to render such hybrid bitrate ladder.
Delivery and load balance

The third factor, briefly mentioned before, is the network conditions. LightFlow can optimize it, by selecting different CDNs, balancing content delivery in real time.

LightFlow combines settings predefined by the customer that include location, and CDN contract structures with real time QoE metrics. LightFlow connects to customer’s analytics to capture information about start time, dropped frames, or rebuffering and weigh it in the algorithm. The result is a decision tree that adapts its outcome to real time changing conditions always selecting the optimal path for each video session.

While the engine is taking smart decisions with all that information, LightFlow is constantly running A/B testing to ensure that it is monitoring potential best paths to match customer’s businesses KPIs as well as trying other hypothesis for the better performance.

How does it work?

LightFlow is fully based on APIs, so it can be easily integrated into any media workflow. It also provides a high-level dashboard in which you can introduce your business KPIs depending on what are your priorities regarding visual quality, costs, quality of service, network, ads… All those factors will have an influence on the behavior of Lightflow.

While LightFlow technology, which can be deployed both in the cloud or on prem, can be used out of the box with immediate benefits, its AI engine learns as it goes, tuning itself to produce even further improvements as you onboard the platform with your specific content and context, be it either VOD or live.