Buzzword friendly technologies have a rich history of being co-opted by the advertising industry to add a veneer of legitimacy to otherwise commoditized products (AI, Algorithmic, Programmatic, Dynamic Optimization, Cross-Platform, etc.) In spite of most of these technologies being vastly overhyped, one technology has stood along in truly redefining the way we advertise: Machine Learning, says, Chris Graham, CPO/GM, TONIK+.
The (Near) Past
Before we continue, a definition of ML:
Machine learning is the scientific study of algorithms and statistical models that computer systems use to perform a specific task without using explicit instructions, relying on patterns and inference instead.
ML is a broad field with a number of approaches, but the two we’ll speak to in this article are supervised and unsupervised. Supervised learning algorithms build a mathematical model of a set of data that contains both the inputs and the desired outputs. In other words, we tell a machine what we want to happen and the machine gradually learns to generate that outcome. A common example of this is feeding a huge number of images of, say, broccoli to an algorithm, then sending random images and testing how often it identifies broccoli correctly. Unsupervised learning algorithms take a set of data that contains only inputs, and find structure in the data, like grouping or clustering of data points. This is useful for finding emergent behaviors that would be impossible to predict ahead of time.
Learn More: The State of Localized Programmatic: Why Marketers Need to Adopt a Personalized Digital Strategy
To date, ML has been used to optimize ad delivery without the need for human oversight. It would be remarkably inefficient (and costly) for Facebook, Google, or any other digital advertising platform to serve ads purely based on who bid the most, particularly when advertisers aren’t getting the results they want and users hate the ads they see. ML allows those companies to rapidly optimize campaigns in the context of inventory demand, bidding, user response, and literally hundreds of other variables, which means more relevant ads for users and turnkey campaigns for advertisers.
What this doesn’t do is tell us why. Why did video A result in 10x sales relative to video B? What about the themes, humor, duration, content, even color, worked? What didn’t? And just as importantly, what worked for each individual viewer, and how do we improve our media moving forward?
Supervised ML tools provide the ability to analyze every frame of a video for content and themes, such as specific actors, objects, action, humor, locations, and emotions. When combined with performance and retention data, we can start to understand not only what content resonates, but what about that content resonates. This allows us to optimize assets and preemptively create videos that drive results.
Large tech companies that process millions of hours of video already analyze those videos frame by frame to determine the content and what’s resonating with users, but this isn’t a technology that’s in wide adoption for content creators. The reasons for this are numerous:
- It’s a relatively new technology
- Most off the shelf tech isn’t aimed at advertisers
- A robust solution requires many ML libraries and tools used in concert
- Creative is typically delivered to media teams as a one-way street, with optimization limited to serving different variants to different audiences
- Most advertisers don’t even realize it’s possible
TONIK+ is solving these problems by building and aggregating a number of ML tools to preemptively process video into individual scenes and label all of those scenes with relevant labels. Once this is done, we can then overlay performance data and provide brands with their first intra-video intelligence report, including:
- What scenes resonate best across all users
- Scene resonance by the audience, including gender, age, and location
- Performance by interest, such as fans of different film genres or sports
- Efficacy across platforms, devices, and video durations
While this information is incredibly valuable in its own right, we’ve also built tools that allow us to automatically generate new, audience and platform-specific creative based on these insights. Videos are cut to appropriate durations and aspect ratios and re-published, typically resulting in average view lengths 20-50% greater than the original videos and clients getting nearly double the seconds of video viewed per advertising dollar spent.
Learn More: AI Analytics in Adtech: 3 Competitive Advantages of Proactive Incident Management
The applications of this technology are boundless, and will ultimately result in video being personalized across TV, OTT, and Social on a per-user basis. Unsupervised ML will be essential in optimizing across these platforms, creating video variants and identifying audiences at a staggering scale. It will also help to preemptively optimize content, aligning content to audiences before we run a single impression. Media plans will be shaped in real-time, with emergent behaviors leading to real-time changes in the content we distribute, our budget allocations, and the ad products we leverage.
In short, everything will change, slowly and then all at once. Make sure your team is ready.
Credit: Google News