If everyone uses AI, who stands out?

When all businesses use the same playbook, they just step on each other’s toes and miss out on potential customers

Featured faculty: Jesse Yao
Written by Putro Harnowo

For the first time, Meta will eclipse Google as the largest advertising platform on earth. The social media company is poised to claim more than US$243 billion in revenue this year, edging out Google’s US$240 billion. Artificial intelligence (AI) has optimised Meta’s algorithms to better match ads to its 3.56 billion daily users while engaging them with short videos across Instagram and Facebook.

Both giants have invested heavily in AI, yet neither can sit on its laurels since other big techs have also ramped up their AI-driven algorithms to refine their social media ad targeting. Advertisers pay digital platforms to ensure their message reaches the right users and receive a payoff if those users make a purchase. Without being targeted with an ad, a potential consumer would not be aware of the product and may never buy it.

Ad targeting creates value for businesses and consumers when advertisers accurately reach their target audience. AI technologies like machine learning and large language models come in handy to help platforms collect information about users’ behaviours and demographics, such as age and location, and crunch the data to predict their likelihood of purchase.

At this point, it may be scary to see how social media uses AI-fueled algorithms to match their user profiles with the diverse needs of advertisers. However, Jesse Yao, an Associate Professor at the Department of Marketing at the Chinese University of Hong Kong (CUHK) Business School, argues that perfect targeting is impossible.

“When competition is strong, companies have a high chance of targeting the same pool of individuals, especially if their algorithms use similar mechanisms,” he says. “Companies will definitely develop more sophisticated algorithms to improve their targeting ability, but data privacy regulations will continue to limit their ability to perfect their targeting.”

Data privacy laws restrict businesses from collecting personal data that could identify an individual, so businesses must implement measures to decouple users from their real identities. Consequently, algorithms cannot achieve 100 per cent accurate targeting or certainty that someone is interested, and even if they did, the targets would no longer be high-quality.

How do businesses deal with flawed targeting

Since perfect targeting is arduous, advertisers often ponder whether to focus on “precision” to carefully target only a small number of very interested groups or on “recall” to cast a wider net to reach almost everyone who might be interested. The drawback is that prioritising high precision may miss out on other individuals who are also interested but were not identified, while prioritising high recall may waste resources.

A paper titled Algorithmic targeting and the precision-recall tradeoff, tries to navigate this dilemma. In the study, Professor Yao collaborates with Ganesh Iyer at the University of California, Berkeley and Zachary Zhong Zemin at the University of Toronto to use game theory, a mathematical study of strategic decision-making in which the outcome for each party depends on the choices of all involved.

“When competition is strong, companies have a high chance of targeting the same pool of individuals, especially if their algorithms use similar mechanisms.”

Professor Jesse Yao

The study highlights that the algorithms used by many platforms may be highly similar. While the algorithms used by the platforms are proprietary, their underlying machine learning techniques are similar, especially when they use public data and the same data analytics tools or AI models. People also typically search for products on multiple channels, signalling their interest in different platforms.

When algorithms identify individuals with a high probability of purchase, there is a big chance that other competitors are also eyeing the same targets. Even across distinct apps like TikTok, Facebook, and Amazon, there would still be significant overlap, and advertisers end up targeting similar groups with their competitors. This is often why, when you Google a brand, you might see an ad for the product you searched for on social media, and then see more ads from other brands.

“Companies can benefit a lot from being the only one that targets interested people, but will benefit less from competing for the same targets,” says Professor Yao. “To soften competition, they strategically target fewer people who are moderately interested, or lower both recall and precision. This way, the targets still have a reasonable chance of making a purchase when seeing ads, but without as much costly head-to-head competition.”

What if businesses tailor their algorithms differently?

Undoubtedly, targeting fewer people with moderate interest is not ideal, as it could also result in a pool of duds. Hence, advertisers are motivated to develop unique algorithms and analytics tools to differentiate their predictions. “The more unique, proprietary data a company uses, the less likely they are to target the same people as their competitors,” Professor Yao adds.

Advertisers nowadays can either work with digital platforms or build their own custom algorithms, but this approach is costly. It may also indirectly help their competitors by making the markets less crowded with ads, without the competitors even needing to spend a penny, so this strategy must be executed carefully.

Some may consider combining public and proprietary data to improve predictions and reduce costs. However, data privacy regulations in many jurisdictions require companies to choose either public or proprietary data for a given consumer, but not both at once.

Combining data sources also creates new personal data profiles, which require new consent or a re-evaluation under most data privacy regulations. Some digital platforms even explicitly forbid data scraping or combining user profiles for commercial targeting, especially if the data becomes personally identifiable.

Modern targeting in ever-competitive markets

Given that there is no easy way to avoid targeting overlap, Professor Yao suggests that advertisers strategically adjust their precision and recall while keeping tabs on the rivals’ digital campaigns. “Companies should not only think about the potential customers but also consider the strategic response from their competitors.”

When ad costs are low, they may consider targeting a larger pool of people to reach as wide an audience as possible or to achieve high recall. When ad costs are high, they should be more selective and target only those genuinely interested. To minimise overlap, advertisers should continue seeking unique segments they can own, rather than competing for the most obvious targets all the time.

For Meta, perhaps its unique segments are its short videos and cross-platform ecosystem, but for advertisers, its vast user base may increase overlap with their competitors. While the study does not specifically analyse Meta’s success, its framework provides a useful lens on how developing proprietary data analytics and strategic differentiation can gain a competitive edge.

There are plenty of fish in the sea, but if all the boats have trawlers, a wise fisher is those who know where to cast their line.