Skip to main content
#3Case Study

Personalized Online Marketing: AI Technology Gone Astray

Magazin #3 | Autumn 2023

Personalized Online Marketing: AI Technology Gone Astray

Nine out of 10 internet users worldwide use the Google search engine. The social network Facebook has 3 billion users, Instagram has 2 billion. In Germany, more than 70 percent of the population have a Facebook account, and over 60 percent use Instagram. But if using these platforms is free of charge, how have their operators – Meta and Alphabet – grown to be two of the largest companies in the world, making billions of dollars in profit every year?

The short answer: Their business models are based on generating revenue from advertising and using personal data to optimize the targeting of ads. This enables them to provide “free” content or services to users. Google, for example, generates 78 percent of its revenue through advertising.

Yet even though users don’t have to pay money to use such platforms, ad-funded content and services come at a price: Users pay with their personal data and by exposing themselves to advertising. We encounter advertising primarily as paid hits in search results or as ads on websites and social media channels – for example in the form of the banners or videos, which are omnipresent on most of our screens. Their tremendous value does not stem solely from their presence online. Rather, advertisers spend substantial amounts of money on ads that are published exactly where the generation of clicks, views and sales is maximized – this is called personalized advertising. Digital targeting mechanisms have inflated online advertising markets. In Germany, the size of the market has doubled over the last six years. Furthermore, online marketing is making significant advances through Artificial Intelligence (AI) technologies. Marketing has become one of the most important fields for AI applications.

How Does Personalized Online Advertising Work?

The journey of a personalized ad can be quite complex: Starting from an advertiser, it passes through several processes managed by intermediary actors in the ad tech sector before it finally appears in a publisher’s advertising space on the screen of a targeted online user. The basis for the whole personalization cycle is personal data, which is most commonly collected, analyzed and managed by the intermediaries.

All steps are based increasingly on machine learning techniques (ML), which form the most common foundation for AI systems. To increase the likelihood that ads will lead to purchases, marketing strategies are usually designed to target specific online users as precisely as possible. Intermediaries use data mining to capture personal data. This data can include: online (and, increasingly, offline) behavior; geolocation data and movement trajectories; device or user identifiers (mobile IDs or advertising IDs, for example) or information from user profiles, which may include sensitive demographic data such as age, gender, ethnicity, sexual orientation, political or religious beliefs, education level, employment status or income. Even if this data is not directly available, ML models can infer such information with surprisingly high accuracy. Cookies – small data files containing browsing histories (website visits) and background connections – are widely used to feed AI systems.

Intermediaries then use this personal data to create user profiles with the help of digital identifiers. Using a method of online tracking known as fingerprinting, they capture identifiers by combining user attributes with data generally provided in network requests (such as IP address, web browser, operating system and hardware specifications) and compress them into a single “digital fingerprint.” Further sources of information include payment histories or cross-platform verifications using, for example, a unique phone number. It is also possible to obtain relevant personal data from third parties. By segmenting such digital fingerprints into various sorts of groups, advertising can now target individually configured profiles of users who are most likely to follow up with a purchase. The profiles are fed into ML models, which are trained and fine-tuned to “predict” the success of the particular ad as accurately as possible. These predictions are then used in real-time micro auctions that sell advertising space to the highest bidder. These real-time auctions take place at the moment a user accesses a website, opens an email or launches a mobile app – invisibly and in the blink of an eye.

In practice, advertisers are not just companies advertising their products but media agencies that plan, implement and evaluate advertising campaigns for companies (Figure 1). Intermediaries link demand-side platforms (operated by advertisers) with supply-side platforms (operated by publishers) to organize the advertising spaces that ultimately deliver ads to the end user. Those spaces can be on feeds, apps, search engines, websites, or other interfaces, and they are constantly monitored to maximize ad performance. The success of specific ads can be adjusted by varying ad content, the frequency with which they are served, their size and shape, and other factors. Buying and selling ad space is taken care of on ad exchange platforms. Data can be stored, enriched, analyzed and segmented on data management platforms. They are also used to create usage profiles.

Finally, media agencies are tasked by advertisers with measuring and predicting the success of placed ads. They use various criteria to do so: Engagement rates indicate how many people click or swipe a specific ad; impressions represent the number of times an ad is served; viewability shows whether an ad was actually seen; reach describes the number of unique users who view the ad; frequency indicates how many different times a user sees a specific ad; and conversion describes the extent to which an ad leads people to take a particular action (such as buying a product). These analyses incorporate ML not only to target users, but also to evaluate performance and adjust future ad campaign strategies. AI thus plays a significant role along the entire life cycle of personalized ads, supporting data acquisition, data analysis, user targeting and the continuous adaptation of all these processes.


Advertisers want to increase the visibility of products and increase revenues through online advertising. They develop advertisements and purchase advertising space.

Intermediaries are interposed between advertisers and publishers, forming a complex network of actors that offer technology services.

Publishers provide the online space that is considered to be most suitable (profitable) for a specific ad to reach a specific user. Publishers might be search engines, social media sites or media and video platforms. They provide ad space in the form of banners, video clips or search results. Meta and Alphabet act as publishers, for example, as do providers of various apps and website operators such as online news outlets.

Users interact with their devices, operating systems, application software or online services. Through these interactions, they see advertising, but they also directly or indirectly provide personal data, which fuels the further personalization of the ads they see.

The Dangers of Market Concentration

The online advertising market is dominated by two companies: Alphabet, better known for its subsidiaries Google and YouTube, and Meta, which owns Facebook and Instagram. Not only are they the most important ad publishers, they also employ some of the biggest intermediaries, which broker advertising space and use personal data to improve ad targeting. This places the three main activities of the industry – advertising, mediating and publishing – within one and the same company meaning that just a few players control the entire journey of an ad.

Alphabet and Meta provide the technical infrastructure in addition to much of the advertising space, and they are also in possession of vast amounts of user data. The collection of this data both within and outside of their platforms gives them a significant competitive advantage, making it almost impossible for other companies to compete. This domination of the online advertising market is frequently referred to as a “duopoly” – the monopoly of two. Many of their services (search engines, social networks, navigation and office apps, cloud storage, translation tools, entertainment platforms, development tools, news feeds, etc.) have become indispensable. With a lack of alternatives that provide a similar level of utility, users often have no choice but to opt for their services. This concentration is not new to the digital economy, where the five companies known as GAFAM (Google, Amazon, Facebook, Apple and Microsoft) are dominant. All of them were among the eight companies with the highest market value worldwide in 2022. In the advertising sector, this centralization of power is particularly extreme as the value and success of advertising is heavily dependent on the volume and accuracy of data.

Due to the lack of competition in the marketing ecosystem, Alphabet and Meta are far more influential – when it comes to pricing, practices and technical standards. This makes it difficult for advertisers, who are dependent on the technological infrastructure offered by, for example, Google when it comes to cloud systems and AI algorithms. Users, meanwhile, hardly have a choice regarding making their personal data available and who is involved in data collection when visiting a website – an information asymmetry that is intentionally reinforced by website design. And governments have a hard time regulating the activities of corporations and keeping pace with technological developments. The upshot is that the current advertising market caters largely to the needs of a tiny handful of companies, while other actors are unable to compete.

People and the Planet Are Paying the Price

The current situation poses a threat to individuals and society. Not only does the extensive collection and analysis of personal data undermine privacy and data protection, resulting in a surveillance economy. Power imbalances and information asymmetries between companies, individuals and countries are also being used to influence political processes. The Cambridge Analytica scandal has clearly demonstrated the extent to which micro-targeting and the dissemination of misinformation endanger the formation of public opinion and thus democracy as a whole.

On top of this, the marketing activities described above consume huge amounts of resources. About 15 percent of the network activity triggered by the loading of a news website comes from ad-related content. Ad exchange servers are running continuously to manage advertising all across the commercial web. The servers consume energy, produce carbon dioxide emissions and trigger additional purchases of resource-intensive consumer goods and services.

To make the online marketing industry sustainable, it might not be enough to simply open the market to new private players. Genuine competition alone would not solve the problem. Rather, the infrastructure of online business and communication networks must move away from the principle of personal advertising. Solutions may lie in the regulation of tracking and and the mass gathering of data. We need to look at alternatives so that public data infrastructures are no longer based on the exploitation of personal data.

Full Study:

Frick, V., Marken, G., Schmelzle, F. & Meyer, A. (2023). “The (Un-)Sustainability of Artificial Intelligence in Online Marketing. A Case Study on the Environmental, Social and Economic Impacts of Personalised Advertising .” IÖW Schriftenreihe 228/2023. ISBN 978-3-940920-33-1.

More about

Further reading:

Armitage et al (2023) Study on the impact of recent developments in digital advertising on privacy, publishers and advertisers.


Research Associates at the Institute for Ecological Economy Research (IÖW) in Berlin

They examine the role of digitalization in the socialecological transformation.