For best experience please turn on javascript and use a modern browser!
You are using a browser that is no longer supported by Microsoft. Please upgrade your browser. The site may not present itself correctly if you continue browsing.
What makes a photo perfect? Usually was a question of experience and gut feeling for marketers, but PhD student Gijs Overgoor went looking for the answer with the help of machine learning, focusing among other things on the visual complexity of images on Instagram.

Overgoor divides his time between the Netherlands, where he is conducting research for his PhD at UvA’s Amsterdam Business School, and in the United States. He is married to an American and teaches at Poole College of Management at North Carolina State University. In his work, he focuses on marketing issues, applying techniques from AI and econometrics. One of the things that has fascinated him for some time now is the enormous amount of unstructured online data. ‘On a standard booking website, such as booking.com, consumers are overloaded with data. A considerable portion is unstructured in the form of images. People simply tend to respond strongly to visual cues and these images convey a lot of information.’

Neither too simple nor too complex

In one of his studies, Overgoor targets the visual complexity of photos on social media. ‘Here, we demonstrate that a picture should not be too simple or too complex if it is meant to attract the attention of Instagram users.’ The study examines the design (how many objects are there in the photo, where are they, how are they presented) as well as the number of details and colours. ‘The latter is possible by means of a fairly simple algorithm for object recognition. For this, we used the MAK R-CNN method.’

Overgoor measured the visual complexity of more than 150,000 photos. ‘This enabled us to structure the information from these images and to put it into tables. Together with other information such as the number of likes, the number of followers and the timestamp of the postings, we put it all in a statistical model.’ Using this, Overgoor demonstrates that it does not matter all that much what exactly can be seen on a photo.  ‘Your behaviour has already been influenced before you have had a good look at what is on it. Based on its complexity, you decide whether or not to stop scrolling as you go through your Instagram timeline.’

 

Copyright: ABS
Your behaviour has already been influenced before you have had a good look at what is on it.

Can a hotel picture predict clicking behaviour?

Other research into images by Overgoor concentrates on pictures of hotels on the website of online booking agencies such as Booking.com and Expedia. 'Here, we are trying to establish a link between the thumbnail photos and consumers’ clicking behaviour. Bases on the images, we can predict which hotel consumers will click on more often. We also know that they can play a role in the evaluation of other details about a hotel, like the price.'

E-commerce parties can use a image to convey much information in a short time. 'This information could be correlated with your assessment of the price, the type of hotel room you are looking for or the facilities (e.g. swimming pool, fitness room) that attract your attention.' What stood out is that location has a substantial effect on the types of images that are successful. 'We compared hotels in Miami, Boston and New York. For Miami, consumers responded strongly to images of the surroundings and the facilities such as a swimming pool or hotel bar, whereas this mattered far less for Boston. This in itself can be explained when you realise that Miami is more of a holiday destination where guests stay longer. Photos, after all, allow you to more or less experience the place you are going to.'

Data science in marketing

In the case of Boston as well as New York, hotel guests actually seemed more interested in pictures of the hotel room itself. 'In New York especially, consumers responded strongly to photos of the front of the hotel.’ Overgoor’s research can be applied more widely to marketing issues where visual and textual components are combined. ‘In the study, we show how it works for hotels specifically but a clothing brand or online retailer can also use our method.'

However, the tools for this are not ready to hand. 'You need a data scientist and you must be able to use tools like Python. I see very little of this in marketing teams, but I do expect an increase the coming years. Knowledge about unstructured visual data is growing. As an example, we can already use images to predict how likely it is that a product will be returned. For Airbnb, a study has shown that a high-quality photo can increase the demand for a particular flat in the short run. But in the long run, it can also lead to a lower demand if there is a mismatch with the quality of the accommodation. Developments within data science are going fast and photos will become an important tool for marketers in the next few years. And I do not so much mean pictures to manipulate consumers but to make their entire online shopping experience better and nicer.'