Will the reborn “Data Fusion” technology become the savior of ad technology?
Targeted advertisement has become more important than ever for any businesses. However, the use of third-party cookies has become challenging and it almost becomes endless game.
One potential avenue for this is through “data fusion.” Data fusion is a technique that combines two or more separately acquired datasets into a new analyzable dataset. By integrating data acquired from separate samples based on the similarity of the samples, it enhances the integrity and accuracy of the data.
For example, let’s say we have data (A) that examines awareness and purchase experiences of a product, and data (B) that investigates whether individuals have seen advertisements for that product and how much they like those ads. By combining data (A) and data (B) based on attributes such as respondents’ age and gender, we can create a fused dataset. This allows us to analyze the relationship between ad viewing experiences, evaluations, and the purchase of the product.
Data Fusion
The concept of data fusion is not new and dates back to the 1990s. It began primarily in the field of marketing, where data was used to analyze the effectiveness of advertisements and promotions, as well as to classify consumers and examine their purchasing behaviors. For analysis, single-source data that obtained multifaceted information such as purchasing behavior, ad exposure, and lifestyle from the same subjects was necessary. However, there was a challenge: conducting surveys to gather a lot of information imposed a significant burden on subjects, and obtaining a sufficient number of analyzable samples was costly. While single-source data, acquired through panel surveys of tens of thousands of people, was available in the past, it was not easily accessible due to its high cost.
To address this, research was conducted on a technique that could utilize data gathered from different subjects through different survey items, combining them at an individual level based on the similarity of samples. This technique, known as data fusion at the time, proposed statistical methods for combining data.
Third-Party Cookies as Central to Linking Individuals and Behaviors
The situation changed with internet advertising and cookies. With third-party cookies, if browsing history, actions, and website visitation history were collected across sites and linked to browser IDs, it could be analyzed as single-source data without the need for data fusion. Ad technology developed on this data foundation, enabling highly accurate targeting of individuals with machine learning models.
Even after the 2000s, the use of data fusion continued to advance mainly in the field of database enhancement, with sophisticated fusion techniques such as automatic data mapping using machine learning becoming possible. However, the term “data fusion” became less heard of in the fields of advertising and marketing.
Reverting to the Past
With the regulation of third-party cookies, however, obtaining single-source data freely on the internet became difficult, leading to the search for methods to determine how to advertise to whom. Given that direct linking of ads to individuals as was done before is no longer possible, there is a need to rethink targeting strategies.
In addition, what’s different now is the quantity and quality of data available for identifying groups. In the pre-internet era, usable data was limited to demographic attributes such as gender, age, family composition, and income, which also required direct inquiries from the subjects. Today, however, the types of data available have exponentially increased. With everyone carrying smartphones and conducting information gathering and transactions on them, behavioral information such as location, web access, and service usage status is shared across various services as fragmented data.
By combining this disparate data, we can improve the accuracy of identifying target groups. For example, in the case of “educational services for toddlers,” by combining location information with YouTube usage data, we can infer that the group using YouTube during the day in the “Manhattan” is likely to be “a relatively high-income household with toddlers.”
By fusing separately obtained data such as location information and YouTube usage data, we can acquire new information. Data fusion, which once exited the world of marketing, has made a return.
Bringing Data Fusion to Ad Technology with Generative AI and Machine Learning
The connection between “Manhattan,” “daytime,” and “YouTube” leading to “high-income households with toddlers” is largely due to marketers’ insights. To correctly identify what insights can be gained by fusing which types of data, a vast knowledge base on lifestyles is required, in addition to an understanding of market and technological trends, resembling a game of associations.
This is where generative AI comes in, showcasing new possibilities. Generative AI, trained on knowledge bases on the internet, can suggest combinations of several attributes representing the target audience for advertising. Machine learning-based pattern extraction and matching can be used for actual data fusion.
With the unavailability of third-party cookies in the future, ad technology will find it challenging to deliver optimal ads based on IDs. However, by incorporating data fusion through generative AI and machine learning, it seems possible to achieve the original goal of delivering “meaningful ads that reach the audience” even with a larger granularity of targets. After all, what advertisers truly seek is “maximizing the effectiveness of their ad spending,” and whether or not individuals can be identified might not be as crucial. Data fusion could become a key technology to reconcile advertisers’ demands with consumers’ privacy protection.
~Other information source~
https://blog.trocco.io/glossary/data-fusion
