Thomas Z. Ramsøy

Eating pizza weirdly and the need for consumer neuroscience AI

In the realm of advertising, the emergence of generative AI methods has ushered in a new era of ad production. Although this technology enables faster ad production, it also poses a new set of challenges for marketers. Previously, ad campaigns had a longer production life cycle, allowing marketers to focus on a few channels and rely on a more unified audience.

Today, however, the advertising industry is fragmented, and marketers must determine the most effective means of reaching their audience through variations in ad creatives, platforms, and segmentation.

With the recent advances in generative AI, the speed in which we are making variations and trying things out is about to explode! But this explosion is potentially at the risk of commercial failure on your behalf.

In this post, I will dive a bit more into the problems that are currently obviously wrong with generative AI, and what this provides as a broader picture of the state of generative AI.

As an example, let’s look at some pizza!

The Pizza Problem: Why Neuroscience Matters for Generative AI in Advertising

While generative AI can create impressive and almost realistic images, there are still some limitations and oddities that must be addressed. The example of people eating pizza highlights how generative AI can produce something that looks almost correct, but still has something off.

In the below examples, I simply gave Midjourney the prompt “person eating pizza”, and it gave me this:

The generative AI solution Modjourney thinks people are eating pizza in a very particular way…

Sure…they’re eating pizza…kind of. But it’s freakish how people are actually eating the pizza…

Recently, we also saw the result of a generative AI that made a novel commercial at the Pizza Later’s YouTube channel, and you can see it below:

Clearly, there’s something off with how AI expects people to eat pizza. The really interesting aspect here is how you respond to the images and video: many people find the images disturbing, uncanny, and disgusting.

It’s not quite how you’d like people to respond to your ad!

Neuroscience-Based AI for Effective Advertising

As AI technology continues to evolve, we can expect these issues to be resolved through more advanced AI models. However, the challenge lies in calibrating generative AI output based on how people actually respond to it beyond stated preferences. This requires the development of AI models that accurately predict how consumers will respond to advertising stimuli. Here, the advances in neuroscience and neuromarketing have proven to be the by far best methods for measuring and predicting consumer responses to ads, packaging, and many other consumer touchpoints.

In this sense, the worst part is that the Pizza Problem is the easy problem. That is, we can easily spot the mistakes that the AI has made, and we can read and hear how people respond to these images and videos. But this all points to a bigger problem: generative AI is not automatically able to produce content that hits the bull’s eye in consumer responses.

On other words, relying on generative AI alone can exacerbate your problems, instead of solving them.

Even if you don’t produce another Pizza Problem, the winner of a thousand versions of an ad you make with generative AI should never be chosen purely based on your liking! As I’ve written earlier, you need predictive AI to correct generative AI.

Countermeasures to the shotgun approach

As marketing increasingly adopts a shotgun approach, trying out various ad creatives at a rapid pace to see what sticks, the risk of creating more commercial noise for the average consumer also increases. This is where predictive AI solutions such as the product Predict from Neurons come into play.

Predict provides feedback to creatives and other assets in a matter of seconds to minutes, allowing marketers to quickly test and refine their ad creatives. Here’s a practical solution for testing generative AI output with Predict:

  1. Generate a set of ad creatives using your preferred generative AI tool.
  2. Use a plugin to import your ad creatives into Predict (available for platforms such as Figma, Adobe, and Chrome). Or do it directly in the Predict platform.
  3. Test each ad creative using Predict’s suite of performance metrics, including attention to branding, call to action, product, text, and more, as well as predictions of how customers will feel in terms of clarity and engagement.
  4. Analyze the results to determine which ad creatives performed best.
  5. Iterate on the best-performing ad creatives, making changes as needed and re-testing with Predict.

By using Predict, marketers can test and refine their ad creatives much more quickly than before, allowing them to stay ahead of the curve and avoid creating more commercial noise for consumers. With the rise of generative AI methods in advertising, it’s essential that marketers have a tool like Predict to help them navigate this new landscape.