Solving First-Party Audience Expansion Using AI

As computer scientists, we are always interested to see how AI is being used to address challenges and solve real-world problems in different lives and industries.

The advertising industry faces particular challenges and opportunities at the moment with third-party cookies about to expire.

Adtech has some of the smartest technical minds in the world – from global giants such as Google and Meta down to small-scale start-ups which are breaking new ground. But without cookies, and the ability to use personal data in advertising, the industry is now facing significant disruption.

Proposed workarounds which consider identity may still encounter privacy issues in the future, and there remains a question mark over how easily these audiences can scale. Yet without this data, marketers face advertising to anonymous users. This is especially being felt since the pandemic, as large audiences have moved rapidly towards Connected-TV, which is still a nascent environment for advertising and data consistency.

However, cookie-based targeting has always had its limitations, regardless of the current challenge. While it’s been useful in assessing whether a person is relevant for a campaign, it is based on a set of assumptions and often only looks at past behaviour. By the time the data is actionable, the browsing moment has long passed, and what is more, cookie-based targeting has always been difficult to scale.

AI and machine learning are now making a difference by enabling marketers to process super-human amounts of data at speed. This new capability is opening up smart, privacy-friendly ways to understand audiences on the fly, during live campaigns. These learnings can then be expanded to find relevant new users, at scale.

Marketing Technology News: MarTech Interview with Mary Gilbert (Kerford), CMO at Folloze

Pre-processing and inputs 

While the processing of large amounts of data at scale is a classic use case for machine learning, the real art of AI solutions is often in the input system constructed around the algorithms, which cleans and organises data from different sources. This is where the true value of AI can be realised.

Illuma, for example, is combining data and systems to provide effective, scalable cookieless targeting. This creates a smart AI system to expand first-party audiences, which is proving particularly useful in scaling sparse but meaningful data to deliver qualified reach and relevance simultaneously.

The Illuma team is using a deep-learning-based AI technique which processes signals and contexts, and then makes targeted recommendations based on a high probability of delivering new and relevant audiences.

At its heart is an advanced recommendation system which uses proprietary feature scores to capture and model the available contextual information. Other audience behavioural signals feed into it, such as brand awareness, ad engagement and performance. These inputs power the recommendation engine, which expands campaigns to find new audiences with similar interests, on the fly.

Rapid response

Common problems with recommendation engines can be speed and cost. It’s one thing to be able to analyse and process the contextual information you’re seeing in a campaign, but to respond in time for these insights to bring value is a challenge, because of the way most recommendation systems work.

Illuma addresses this by organising the system infrastructure as different layers and by adopting latest technology such as serverless architecture, containerisation, and specialised databases. The system pre-computes large-scale tasks in advance and recalls and ranks results to enable agile targeting at speed.

To bring this back into the context of solving industry problems; when you combine valuable first-party data with a system that can learn and recommend, you create a proprietary AI infrastructure that can scale small but valuable known audiences at speed, in a way that is both useful and unique every time.

Without doubt the advertising world is evolving at pace, and AI and machine learning are supporting smart new thinking which should help the open web to thrive in the privacy-first era.

Marketing Technology News: Customer Intelligence 2.0 – Natural Language Understanding Opens New Horizons for Marketeers

ABOUT THE AUTHORS

Alex Rogers is a Professor of Computer Science at the University of Oxford, specialising in computer architecture, AI and machine learning. He is passionate about start-ups who are deploying these technologies to address real-world problems.

Yu Liu holds a PhD from Imperial College, London, and specialises in applying data engineering to advertising. He is head of data and technology at Illuma.