Creative is the soul of effective advertising. A well-thought-out creative piques consumers’ interests and is bound to increase quality conversion. According to digital marketing experts, Americans are exposed to around 4000 to 6000 ads each day. And with digital marketing showing no signs of slowing down, this figure is only likely to increase.
According to research conducted by The CMO’s club, 74% of CMOs consider brand loyalty as most negatively affected by bad ads. The same research also suggested that 78% of CMO’s considered it inexcusable’ to serve the same ads to consumers.
And it’s almost a no-brainer why serving the same ad creative is inexcusable. Most of these ads don’t make a connection with the consumers simply because they weren’t compelling enough. One major reason for the underperforming creatives is creative fatigue’.
A creative is very unlikely to evoke interest if it is shown repeatedly to the same audience. It’s like scrolling through a loan creative that you have seen dozens of times – humdrum is an understatement, it becomes annoying.
Fighting creative fatigue needs agility and a strong data-backed strategy. Generally, as the campaign progresses, design teams gain insights into the performance of all the creatives that are working well. Based on the designs of these well-performing creatives, design teams come up with creatives that share the same design signatures.
But this becomes a challenge for a limited number of designers working on hundreds of campaigns. Besides, it’s a time-consuming process that slows down the overall campaign effectiveness.
This problem needs an agile and scalable approach where the best-performing creatives can be used as inputs to improve designs. Creative teams should focus on improving design communication and not spend time on repetitive tasks. Currently, designers spend a large chunk of their time in resizing, personalization, and other repetitive tasks. Generative Adversarial Networks (GAN) are artificial intelligence models that can help increase design team efficiency.
The GAN model architecture involves two sub-models:
Generator – An AI model that is used to generate new images.
Discriminator – An AI model that is used to classify whether generated creatives are real or fake and from the domain based on the input data.
Let’s consider that we are utilizing a GAN model to create images. A discriminator network is exposed to thousands of real images. These images are training set data. Once the discriminator learns the image patterns, it can distinguish between an image that is directly taken from the data itself and an image that is newly generated.
A generator network, on the other hand, generates new images based on the noise of the training images. Noises are the secondary features such as skin color, hair color, face symmetry, eyebrows, and more. Based on these features, the generator creates an image. The created image can be biased; which means it can be strikingly similar to one of the real images. At the same time, this image could be different from the original image, which is the desired outcome.
This ground truth detection by the discriminator is done using a binary classification logic running a sigmoid function under the hood. The interesting thing to note here is that the discriminator is just estimating the probability that the sample is coming from real data or generated using a threshold probability cut-off which could be tuned based upon the business costs estimation of false positives and false negatives.
How does GAN Overcome Creative Fatigue?
Creative performance to a great extent affects the overall campaign performance. Creative feedback can be taken from the performance management system. It highlights the creatives that are working well for different cohorts.
Creative AI – the dedicated AI engine for generating creatives utilizes this feedback from the performance management system. The AI understands the layout, color palette, design, objects, and other features of the ad creative. The GAN model then generates images that are similar in terms of the above features. For example, if a red background featuring a black-colored object is the creative feedback, the model might generate a maroon background featuring a grey-colored object.
This can be achieved across thousands of features that the human eye tends to miss out on. In a human face, these extremely minute features could be eyebrows, wrinkles, face symmetry, skin tone, and more. For AI, these features can be exploited to create a totally different image. This enables the marketers to reduce creative fatigue in real-time across campaigns.
GAN – The Future of Creatives
There has been an increasing emphasis on getting the targeting and optimization part right in the modern digital marketing scenario. And few brands are leveraging AI to achieve that level of optimization and efficiency. But a razor-sharp accurate targeting along with an underperforming creative only decreases the ad effectiveness.