Marketing Thought

Predicting What Image Works For An Individual

One of the central challenges of marketing is that people differ. Marketers would love to create marketing communications that work for everyone. While some communications do pretty well nothing has universal appeal. Some images work fantastically for some people but turn others off. How can we predict who will like what? What helps us in predicting what image works for whom?

ML Can Help Us In Predicting What Image Works

Sandra Matz and her colleagues delved into this problem using machine learning (ML) techniques. The idea is that marketers can extract detailed information about images using algorithms. They can then compare this information about the image to what we know about the viewer. In essence, they want to see what personal characteristics are associated with liking what elements of an image.

As they wanted to get beyond the obvious — e.g., demographics — they get participants to answer a standard battery of tests that assesses them on five dimensions. They use the OCEAN model; openness, conscientiousness, extroversion, agreeableness, and neuroticism. These are pretty standard in psychology and each measure pretty much tries to capture what you think they do. (Looking at this I also found out that people like to argue about the spelling of extraversion versus extroversion. I’ve used the version the authors use here — but it feels wrong to me).

Lack Of Theory

An interesting thing about this approach is the lack of theory. We don’t know why neuroticism is associated with liking cat pictures but it is. And “[o]penness appeal was positively correlated with the colors blue and black” (Matz et al, 2019, page 378). I don’t know why but the idea is to have good prediction and they show the ability to predict what people would like.

Are You Prepared To Admit To Liking Cat Pictures? Photo By Pixabay From Pexels.com

Such a prediction is appealing to marketers but they note that they had access to participants who happily filled out a personality questionnaire. They see this being used more practically in a world where we can assess personality from “..digital footprints such as the contents of personal websites” (Matz et al, 2019, page 385/6). We aren’t far from that though.

Modest Effects

They also note an issue with their study is that the effects are pretty small. This procedure is a lot of effort for a pretty modest change in liking. This is a worry with a lot of academic research. You find an effect which (even assuming it is truly there) doesn’t really matter. To be fair I think they have a pretty good response. Online appeals can involve communications to a very large number of people. “…when implemented at a scale that is as large as that of most multinational’s marketing campaigns even small improvements over existing approaches could lead to meaningful gains. (Matz et al, 2019, page 386).

For more on ML see here

Read: Sandra C. Matz, Cristina Segalin, David Stillwell, Sandrine R. Müller, and Maarten W. Bos. “Predicting the personal appeal of marketing images using computational methods.” Journal of Consumer Psychology 29, no. 3 (2019): 370-390.

Exit mobile version