Dynamic Ad Creation
Key Takeaway: We patented a system that made every ad smarter than the one before it.
Dynamic Ad Creation Vungle Creative Labs, 2016. US Patent No. 10846735, granted November 2020.
In 2016, most mobile ads were still fixed assets. The same creative, served to every user, on every device, regardless of context. We set out to change that.
​
The problem was not that advertisers were producing bad creative. It was that the delivery system had no intelligence. A high-end ad unit built for a flagship iPhone was being served identically to a user on a low-bandwidth Android device in an emerging market. The format didn't adapt. The layout didn't respond. The content didn't change based on what the user was likely to engage with.
At Vungle Creative Labs, we built the system that fixed that.
​
The product
Launched in November 2016, Dynamic Templates was the first public expression of a system we had been building for some time: a way to serve contextually appropriate ad creative to each user without requiring advertisers to manually produce hundreds of variants.
​
The initial template library covered the core format types: interstitial, carousel, overlays, and gradients. Each template was responsive and modular. A single creative could be scaled and optimised in real-time across multiple layouts. The advertiser provided assets once. The system handled the rest.
​
But Dynamic Templates was a starting point, not a destination. Over the following two years, the system evolved significantly. Adaptive Creative, the framework that succeeded it, went deeper: a front-end web framework built in Javascript, HTML, and CSS that separated Vungle's ad logic from the creative itself. Rather than serving a fixed ad unit, Adaptive Creative could listen to commands from the SDK in real time, pausing an experience if a user left the app, resuming it in the same state when they returned, and adjusting the ad dynamically based on publisher settings, device conditions, and user behaviour as it unfolded.
​
The template library grew from the original four formats to fifteen by early 2017, and continued expanding. Each new template was specified to the component level: token requirements, device compatibility across breakpoints, portrait and landscape orientations, iOS and Android support. The system that had started as a way to serve better ads became a platform for defining what a better ad could be.
Three zones, assembled dynamically: video preview at the top, app information in the middle, download prompt at the bottom.
The same content, a different template: full-screen animation, brand prominent, single action.
An interactive configuration where the user experiences app screenshots.
Three zones, assembled dynamically: video preview at the top, app information in the middle, download prompt at the bottom.
The patent: formalising the system
The technical architecture underlying this work became US Patent No. 10846735, filed in November 2017 and granted in November 2020.
​
My primary contribution was leading the front-end template design: the visual architecture of the ad units, the layout logic of the token positions within each template, and the design specification that defined how content items mapped to each zone across device types, orientations, and breakpoints. While the back-end selection system, machine learning layer, and serving infrastructure were built by the wider engineering and product team, I was an active participant in the broader system design discussions across all three layers. The creative and technical decisions were deeply interdependent, and the template design could not be separated from the logic that selected and assembled it.
​
The patent covers a token-based template system in which each template contains named positions at specific locations within its layout. The same token can appear in multiple templates at different positions. Advertisers associate content items with tokens once. At serving time, the ad server selects the appropriate template and assembles the ad dynamically, based on the user's profile, their device capabilities, the application context, and environmental conditions including time of day, location, and connectivity.
​
A machine learning layer generates predicted engagement values for each candidate template and content combination, serves the highest-probability configuration, and feeds actual engagement back to continuously improve the model.
​
The line to today
The work that started in Vungle Creative Labs in 2016 established the architectural principles that continue to define how Liftoff approaches creative technology. LXA, Liftoff's current creative automation and AI platform, is the direct descendant of those principles: a system that separates creative assets from delivery logic, selects and assembles ad experiences dynamically, and improves through continuous feedback from real-world performance.
​
The technology has changed substantially. The underlying idea has not. Build a system that gets better at serving the right creative to the right person at the right moment, without requiring a human to manually produce every variant. That was the goal in 2016. It remains the goal now.