Machine learning technology is optimising audience targeting across the Out of Home landscape. This has implications for everything, from future programmatic opportunities to automated tailoring of campaigns to meeting brand safety requirements. But perhaps the most exciting part for some is how it is affecting campaign ROI through the heightened accuracy of audience targeting it can provide.

At oOh!, we have spent the last few years building our own machine learning technology, amongst other things, to drive our propriety audience-led campaign planning tool. Here we will share some of our learnings and decisions that have come together to transform a tantalising opportunity into a marketable product.

Parting the sea of data

The rapid evolution of big data has opened commercial possibilities that would have seemed far-fetched just a few years ago. The technology required to capture huge amounts of information from a vast array of sources is now pervasive, and a dream for marketers seeking to reach target audiences efficiently and effectively.

But how to make sense of it all? And what are the technological building blocks that help turn a sea of data into something of use for advertisers and marketers?

We have always known that Out of Home is highly effective at reaching audiences at scale, but our challenge was to work out the smartest and most efficient way to deliver targeted audiences to marketers. With over 37,000 asset locations across Australia and New Zealand selecting and optimising for the most effective outcomes for campaigns can present challenges. Factor in data from numerous feeds such as our own audience data, industry audience measurement, public information such as traffic volumes, and billions of transactions and mobility data from our external data partners, and there are some real challenges in capturing, prioritising and deploying the information appropriately and in a way that adds value.  

oOh! needed a solution that would allow us to optimise campaigns to target audiences using real world behavioural data we have access to, and then identify the best combination of oOh! assets to reach different audience segments. This was a genuine technical challenge.

Tech build

To make this work in practice, our strategy focused on building a recommendation engine powered by our data science capabilities. The engine would be the backbone of audience-led buying, providing rapid predictions on the best asset mix to deliver maximum reach of a specific target audience. This would then allow clients to ensure they would be putting their message in front of as many prospective buyers as possible to drive the greatest potential ROI.

Our first step was to develop oOh!’s Smart Reach capabilities, a recommendation engine designed to optimise reach of audience segments by Out of Home channel. Smart Reach leverages other systems in our tech ecosystem, including those of our data partners, to operate at scales and volumes not seen before. 

This work encompassed the development of scalable algorithms that rely upon historic data and continuous information streams to recommend optimised site locations for maximum audience reach across channels aligned to specific campaign objectives. The development process involved utilising these algorithms in combination with advanced Machine Learning (ML) techniques.

We next trained the recommendation engine on vast amounts of data. This included running analyses on thousands of campaigns to understand how the predictive models could be finessed, on the basis that the more data the algorithms could process, the better their recommendations would be.

Various Machine Learning tools helped us achieve these goals, and we were able to develop a proprietary system that combines AI learning and decision science principles on top of the predictive models to increase the recommendation quality further.

Cloud support

The development of a recommendation engine led us to explore new approaches in the wider technology space, especially with support from cloud infrastructure providers.

We partnered with Amazon Web Services (AWS), leveraging some of their latest Infrastructure as a Service (IaaS) features to build out our analysis and machine learning capabilities. We also deepened our collaboration with other partners such as Databricks, using their data and AI features, which played a critical role in our ability to quickly, reliably and inexpensively generate quality datasets to train our predictive models on.

With both providers, we can scale up and down as required to meet the demand of any post-Covid market growth, and this adaptability is a critical design principle for us.

Future prospects

Today, these tools described above are already in use and delivering results. A recent campaign for Ingham’s saw the brand increase its share of the whole chicken category and attract a significant proportion of new buyers to the brand following an audience-led campaign that used the recommendation engine.

Where we are today is not the end game, but rather the building blocks for future innovation and market offerings as data, digitisation and audience targeting become ever more important and sophisticated. As more campaigns run across our national network, the more data is fed into the system and the better the predictive models become.