New research from the Hebrew University and New York University utilizes Artificial Intelligence and Machine Learning tools to study the collective perception of brands. “It is possible to map all brands on a unified space of associations. We can understand what differentiates one brand from another or to whom they are similar, even if they don’t look like they are close to one another.”
Even before it was published, the new research from Professor Renana Peres from Hebrew University and Prof. Daria Dzyabura from NYU has been making waves. “We are flooded with questions from Israel and around the world, from giants in the field of advertising”, shares Peres. "The quest for a tool that will help pinpoint the collective image that each brand evokes creates great interest outside of academia."
Brand perception can be defined as the set of associations that arises among consumers when they think of a particular brand. Understanding this set assists brand managers apply effective advertising techniques and better manage brand image. Peres and Dzyabura think that their new research, recently published in the prestigious Journal of Marketing, is one of those rare moments when the academia is not dealing with theories alone, but rather proposing a methodology for mapping a field that can be practically applied, and therefore extremely helpful.
Moving backwards, while using new tools
It began back in 2016 when Peres, a professor in Marketing at the School of Business Administration at Hebrew University spent her sabbatical year at NYU. “One day, a researcher whom I did not know came into my office holding a book from the 60’s, which described how they used to check consumer’s perception of a brand during that time period,” she explains.
“Back then, we would give consumers a brand name, and ask them to describe the brand by creating a collage from newspaper clippings. Then, we had psychologists analyze and gather conclusions based on the world of associations the pictures found in the collage. It was a qualitative and tedious approach, that did not allow for the application and analysis on a multitude of brands or on a wide range of customers. Research has advanced since then, to measure brand perceptions using text and image analysis from social media, but this was still far from a perfect method.
“People don’t really post about everyday brands that they use, like their toothpaste. When they do post pictures of themselves with a brand like Nike, the goal is to represent themselves and not necessarily the shoes. Because understanding brand perception is a major challenge for marketing executives, we wondered if it is possible to develop a tool that would maximize the benefits of brand communications by understanding the collective brand image on a deeper level.”
As it turns out, the method was to turn to the past, while using modern tools. Peres holds both undergraduate and graduate degrees in Physics from the Hebrew University, and Dzyabura studied Mathematics at MIT. The researchers used their technological skills to reshape the traditional collage making methods and move it to the digital space, using technological tools that included advanced options for gathering data and image processing.
“We applied image processing and machine learning tools in order to solve managerial questions from the Social Science world, and we developed a smart platform which allows for creating online collages,” explains Peres. “The participants created a collage by choosing from hundreds of thousands of pictures, and described in a visual and creative way what the brand meant to them. We neither ask nor assume, that way the deepest metaphors are drawn. Afterwards, we take all of the collages from a specific brand, and using artificial intelligence-based methods each brand is characterized by a common denominator, while being differentiated from other brands. This is qualitative research that can survey a particularly large public, and can be done on any brand.”
The association of Lamborghini: Lingerie
Over the past four years, the two researchers conducted experiments that gathered 5,000 collages from 1,851 people, who were requested to each edit three brands that they are most familiar with. They were allotted twenty minutes, and for every collage they were paid $2.5 through crowdsourcing platforms such as Amazon Mechanical Turk. The participants were all American, 56% were women, and there were fewer younger participants than older. “It was not a bad sampling in terms of demographic balance,” says Peres.
In total, more than 300 leading American brands were mapped. The results, as described in the paper, are very interesting. “After we did this, we got a multi-dimensional space that included 150 associations, on which we were able to map every brand in the world,” says Peres. “You can position a brand within this association space, and understand what differentiates it from other brands in the same sector. Additionally, you can do cross-branding, and understand which brands have a similar sequence of associations, even if the brands themselves do not seem similar.
Give an example of new things you learned about a brand:
Peres: “We learned that people not only associate Starbucks with coffee and food, but also with pastries, and, surprisingly, different types of energy, from electricity to wind energy. Another strong association of Starbucks is computers and technology. Why does a Coffee chain elicit such strong associations with energy? It is not entirely clear, but it is worthwhile information to know.
“When we studied the Band-Aid brand, we expected collages which show images of blood, cuts, and tears, but the associations were completely different: mothers, homes, love, warmth, childhood, an American staple. Cuts and bruises were almost not present altogether. For brand managers, this information is gold.”
By studying the collages that participants made for prestigious brands Lamborghini or Porsche, there were clear results that differentiated the two brands. “For Lamborghinis, the strongest association was with lingerie,” Peres says, “more so than speed or movement. Additional associations were houses, money, estates, yachts, old cities—essentially anything that could be defined as ‘old money’.
“In contrast, the strongest associations that came up for Porsche were urban landscapes, followed by finances, planes, suits, parties, and jewelry. These are two elite brands of cars, but have completely different worlds of customer associations. By the way, the collages that were collected for Ferrari were very similar to those of Porsche.”
“One of the advantages of our technique is the ability to compare and understand which brands share associations. It became clear to us that Kitchen Aid and Cheesecake Factory share many associations, and they are not the only ones. It turns out that it is worthwhile for brands to cooperate, something that they would have never thought of doing before.”
“Because we are dealing with a system that has a large quantity of information, it is possible for the study to be applied in different disciplines. For example, possible visual recommendations that could be beneficial in campaigns. We know that brands which are perceived as Charming, appear to be more flowery and less industrial. It can be another application of the system, that you are able to recommend one type of visuals over another.”
Who will actually be using this tool, if it is indeed commercialized?
“Consulting or research companies, to whom their consumers will turn both before and after the position processes. It is important to understand that in the worlds of design and advertising there are many assumptions that have never been tested—for example, the logos of the banks are usually blue because the traditional perception is that it is a color that conveys solidity. This is a kind of historical assumption that has not been thoroughly examined.
“With the help of the system that we are proposing, it will be possible to run a data-based process on a qualitative field. The advantage is not just in advertising, rather it is possible to process and apply this tool in any field where we want to understand more intelligently what people think. "