Introducing the AI-dience: The most accurate consumer model in advertising today

The digital advertising industry is beginning to embrace AI models so powerful they can accurately predict not only your interests but how likely someone is to make a purchase at any point in a browsing session.

This is the era of the AI-dience, the Web3 version of audience segmentation – the division of consumers into subgroups according to demographics and interests. Traditional audience segmentation is the process that delivers luxury brand advertising to over 30 executives living in Hong Kong’s Central and ads for tutoring services to parents in the Mid-Levels.

It is also flawed: serving cat food ads to dog owners and facilitating the reappearance of annoyingly persistent links to sneakers you searched for and bought, a week ago. The use of AI puts an end to this, transforming consumer relationships with online ads.

For brands, the price is high. Marketers in APAC face two perennial challenges. The first is the return on investment (ROI) – research shows that brands in the region waste 40% of their marketing budgets. The second is reaching new consumers, particularly those outside the tentacles of the big social media platforms.

Overcoming these will only get tougher with the imminent phase-out of third-party cookies, the technology that enables websites to share consumer browsing behaviour with each other. As this part of advertising’s arsenal disappears, the industry needs technology that will make its limited reserves of first-party data – information that brands collect with a user's permission – work even harder.

Fortunately, the disappearance of third-party cookies coincides with the increasing adoption of AI, a potential solution to the challenges of reach and ROI. Businesses throughout APAC, the world’s fastest growing AI market, have been keen to embrace the new technology.

AI’s contribution to advertising goes beyond its ability to analyse large data sets; it can model advertising campaign effectiveness before a single dollar of budget is spent.

One of the tools AI uses to do this is machine learning (ML). In very simple terms, this is when a computer iteratively trains itself to best predict (in the case of advertising) when a particular user is likely to carry out an action such as adding an object to their cart.

The resulting algorithms can consider thousands of variables, such as time of day, how many ads a consumer has already seen, previous browsing history, their preferred format etc. They can also process Big Data at a rate that would have been unthinkable even five years ago. Another advantage is that they improve over time to constantly optimise results.

ML can be used to create a more powerful version of lookalike audiences – groups of consumers that share similar characteristics to an existing audience that is known to be engaged with a particular brand or product. Advertisers that use lookalike audiences are betting on ‘more of the same’, which has proven to be a fairly safe wager. However, even better results can be obtained when combining lookalike audiences with AI-created audience segments.

Predictive audiences (AI-diences) are a group of consumers defined by the likelihood of them taking a specific action, such as downloading an app or making a purchase. AI creates these from mixed inputs, including historical performance, a user’s website interactions, and feedback from on-screen user surveys.

Predicative audience modelling is more precise than lookalike modelling as it examines individual behaviour rather than similarities to a static seed group. It solves the reach problem by spotlighting consumers who are most likely to be interested in a brand’s product.

AI also helps with the ROI problem. A Microsoft study shows that advertisers using predictive targeting achieved an average 46% higher conversion rate compared to ad groups created by other methods. The ability to optimise audience segments in real-time enables marketers (or rather, their chosen programmatic advertising software) to channel media spending to where it is most likely to make an impact.

For example, if a website ad placement costs an advertising agency $5, they could serve it to a lookalike audience, but can only guess at probable engagement rates. However, using a predictive model, they can serve it to a narrower audience that is highly likely to engage. This AI-assisted predictive method reduces the risk of wasting investment on ad space that won’t meet goals by improving performance.

Moreover, using a pool of adverts (perhaps generated by AI themselves - but that’s another article), AI can tailor the messaging and the format to each audience segment. For example, it might serve as a discount link to a group of consumers that it predicts are poised to buy or a brand awareness video to others who are new to a product or service.

Finally, AI-dience modelling is powerful enough to work from user-supplied, first-party data without dependence on third-party data. Consumers will retain maximum privacy while seeing ads they actually want to see (no more cat food ads for puppy parents!). While a lot of the narrative around AI’s seemingly sudden proliferation has concentrated on the division of humans and machines, the AI-dience has the potential to bring computers closer to understanding capricious human behaviour and delivering a personalised consumer journey.

The article was written by Bernard Fung, managing director, head of North APAC at LoopMe.