In Short
Online advertising promises to better match adverts with consumers’ desires than offline advertising ever could. Its hunger for personal data, ability to spot vulnerabilities in consumers’ behaviour, and tendency to reproduce discrimination in society has led to ample criticism. To bolster current policy efforts, we searched for a way to reduce the risks of online advertising while maintaining its benefits. We created a new tool to govern personalisation in advertising. It balances risks with benefits and produces an optimal degree of intervention in the distribution of online adverts. This is done by introducing a carefully considered amount of untargeted or ‘noisy’ adverts amongst the personalised ones.
The Age of Personalised Advertising
Today, it is impossible to imagine an internet without advertisements. We search online for flights to a beach destination and ads for hotels and rental cars keep on following us around the web. Such personalised advertising is truly a creation of the internet age: algorithms track our online behaviour, infer our preferences, and deliver (more or less) relevant ads. As much as such personalisation leads to a better matching of ads with our true preferences, both consumers and advertisers benefit. Consumers receive more useful information and advertisers’ messages reach their intended target. Unfortunately, this is not a guaranteed outcome.
When Personalisation Goes Wrong
Personalisation can exclude consumers from offers other consumers receive. Some may not be served with ads because of a personal characteristic. Historical biases and discrimination can play a huge role here. At the same time, advertisers may target people precisely because of a vulnerable characteristic. For example, they may serve consumers who are predictably irrational in their spending habits with exploitative ads. Personalisation can further lead to some consumers being charged higher prices than others. Considering these problems, we started looking for a way to keep the benefits of online advertising while defusing some of its risks.
A Smart Tool for Governance…
Our suggested solution is to introduce noise to personalised adverts. Adding noise means that we steer personalised ads off target. Imagine a seller targeting only a particular group of consumers, for example, single mothers. Through adding noise some of the seller’s ads will now be distributed to consumers who are not single mothers. We suggest noise should be added randomly. This means that we do not decide which particular consumer groups will receive the seller’s noisy ads. Instead, for a proportion of the seller’s adverts, we are metaphorically speaking “flipping a coin” as to who will receive them. In our paper, we show that the optimal degree of noise can be calculated: Just enough to reduce excessive personalisation while not intervening too much and keeping the benefits of advertising.
To implement this calculation, we first developed a measure for the concentration of personalised ads: The Concentration-after-Personalisation Index (CAPI). We assumed that the more heavily a person is served with personalised ads, the higher the risk is for them to experience negative effects of personalisation. With the CAPI alone, regulators can already detect areas of concern in the distribution of personalised advertising.
…in a Changing Policy Landscape
It is good timing for a novel tool such as noisy targeting based on the CAPI. Regulators are introducing new laws, for example, the European Union’s Digital Services Act which aims to better control the harms of personalised advertising. Our tool could be an urgently needed yardstick for regulators to balance the risks and the benefits of personalised advertising, as we show that adding the right level of noise to the ad market is an effective mechanism for a balanced governance towards fairer online advertising.
References
Johann Laux, Fabian Stephany, Chris Russell, Sandra Wachter, and Brent Mittelstadt, “The Concentration-after-Personalisation Index (CAPI): Governing effects of personalisation using the example of targeted online advertising”, (2022) Big Data & Society, Volume 9, Issue 2
