Schedule a demo with Ad-Lib.io today
Have questions about how we make creative production easier? Fill out the form and an expert will be in touch shortly.
Join top global advertisers already using Ad-Lib.io
Between product, promotions and people, advertisers have many choices for images to include in their ads. Which of these has significant impact and why? We analyzed data to observe how people - particularly faces - impact ad performance, and what’s going on to make that happen.
In this series, Engaging Creative, we’ll dig into questions we get from clients about key levers of creative effectiveness and apply analytics to test hypotheses about the drivers of performance.
First, let’s consider the impact of faces.
What’s the impact of a person’s face on the ad viewer?... empathy?... self-recognition?... admiration? Whatever the emotional connection, our data analysis has shown that ads with faces outperform ads without faces.
However, throwing in a face or two is not the answer - it’s the face on its own that makes the difference. Why? Once again, it’s the influence of entropy at work.
What exactly is entropy, again?
As we discussed in our Engaging Creative #1, entropy is a statistical measure of information transmitted by an image. An image with lower entropy typically has fewer colors and less variation. Higher entropy means the reverse – lots of colors and elements coming together in a single image to convey lots of visual information.
So how do entropy and faces work together?
We analyzed ads with faces and different levels of entropy to measure their performance impact.
The result?
For simple ads with low entropy, adding a face leads to higher predicted CTR.
After all, adding a face to a busy ad (high entropy) only increases busy-ness and will decrease the ad performance because the face cannot stand out and capture attention.
What’s so special about faces to make this happen?
Faces, in short, are very memorable.
Research shows that faces make an image more memorable. 1In the graph below, based on industry-leading research with neural network modeling, the impact of a face detected by these models demonstrates significantly more memorability.
So, we took our face-related analysis one step further by applying state-of-the-art, open source memorability models to advertisements with and without faces. Memorability models produce a type of heat map to capture the locations that are getting the most attention, where a red color indicates a higher driver of attention.2
For every ad, there are different regions of the image that correspond to different sets of objects. These regions have different probabilities of being forgotten, and some regions have a probability of being imagined or hallucinated. The model uses a neural network to take into account clusters of gradients, textures and color features to highlight regions more memorable than others. For example…
For this ad,
But for this ad with a clearly visual face,
As measured by the model and visible by the strong, almost singular red area, this second ad has higher concentration on the face, causing a stronger memorability score.
Circling back around to the interaction of Faces and Entropy from a Memorability angle, our analysis further shows that fewer, larger faces in ads are consistently more memorable that no faces or crowds of faces, but when the ad is less busy, memorability is improved even more.
Memorability - one more piece of the puzzle
Memorability is just one of the many creative and media factors that come into play to impact ad performance, but with recent advances in data and modeling to measure visual impact, it is another important aspect to consider, especially in pre-campaign analysis. Heat maps can be used as an indicator of the key attention-drawing elements within the ad. Too many red, attention-grabbing areas are a warning flag that the ad may be overly busy, and too few may indicate that there is a lack of anything really noticeable. For improved ad performance, best practice is to keep faces, entropy and memorability at the top of the checklist for your pre-campaign evaluation of creatives.
Footnote for the Analytics Techies
Our multivariate regression model was fitted to variant-level data from April 2020 - December 2021 and filtered to exclude extreme CTRs (>2%) and variants with low serving (IMPRESSIONS < 1000).
The significant main effect of facePresence, and the significant interaction effects of facePresence and Entropy had a statistical significance at alpha level 0.05 (all p < 0.01).
%201.png)
%201.png)
