BrandGuard: The Trust Layer Powering Hyper-Personalization

The Trust and Safety Layer Enabling New Marketing Capabilities

One of the new capabilities that AI is enabling is hyper-personalization. Hyper-personalization is the creation and delivery of unique content tailored to a specific individual. It is a marketing strategy that goes beyond traditional personalization techniques by leveraging artificial intelligence (AI) and real-time data to deliver a unique experience to each customer. In the era of information overload, hyper-personalization stands out as a beacon of relevance. It allows businesses to cut through the noise and deliver messages that resonate with the individual preferences and needs of each customer. This level of personalization can significantly enhance the customer experience, leading to increased engagement, loyalty, and ultimately, sales.

Why does personalization matter? It has been shown to improve uplift. This doesn’t take a strong mental leap, as you perform personalization everyday. You change the way you describe an event or situation differently to your peer vs your boss vs your spouse vs your kids. The core content is the same but the way it is expressed differs. In a similar fashion you’re much more likely to take a recommendation from a trusted friend who speaks your language rather than a giant billboard ad proclaiming a broad message.

AI Trust and Safety Enablers

Messaging has always been about two parts - the message you’re trying to convey and the actual content you are using whether words, images, videos, or actions. Generative AI takes care of the actual content to use while allowing you to focus on the message you are trying to convey. This is great because when performing hyper-personalization you wish to convey a similar message to many different people but the exact message needs to differ for each person.

However, in talking to a lot of large companies, they are hesitant to use generative AI at scale. The reasons fall into a few main categories:

• They are unsure of the legality of generative content

• They are not certain what the outputs will be

• They can’t verify the outputs at scale

• They don’t want to damage their brand and reputation from sending out poor quality or potential litigious content

These issues all point to an inability to trust AI or understand the ways to govern models. Companies work with vendors who operate in a similar blackbox and might have similar concerns, so why are these issues? Machine learning and AI are ultimately about automating a decision making process. A certain amount of rigor needs to be applied with additional processes in place to ensure outputs are sound. Without automated verification, hyper-personalization is not possible because the concerns above are left unchecked.

Think about it this way. You don’t outright trust what a young associate creates for a given task and thrust it directly into the public eye. You take whatever the associate has created and put it through a verification process. You review it. Your boss might review it. Perhaps your boss’s boss reviews it as well. Each touchpoint provides not only an opportunity for feedback but also a moment at which the output is prevented from moving forward. An automated system using generative AI, such as one to hyper-personalize content that is unique to each individual, requires similar verification steps. Enter BrandGuard.

BrandGuard: The Guardian of Your Brand

BrandGuard verifies that any content it sees is on-brand for your brand and conforms to the style of your brand. This is possible because we have taught machines how to understand brands. BrandGuard uses a range of different statistical, machine learning, and AI models to determine the underlying factors that cause a piece of content to be on or off brand. These models are built into an automated system that can be used across a variety of use cases. From checking content from vendors to allowing large scale automated verification.

How does BrandGuard enable hyper-personalization? BrandGuard provides the trust and safety layer of any generated content and ensures that the content remains on-brand and of quality. Here’s an example:

Let’s imagine you are Coca-Cola and you want to run a hyper-personalized ad campaign. You have a really good seed piece of content you like but you want to alter the picture so that the individual in the picture appears more like an individual with the unique person’s demographic characteristics. So you use a generative AI algorithm that takes your seed picture and combines it with demographic information for millions of individuals to generate millions of hyper-targeted images. Those images are too numerous to check with human labor in an economical way so you need to implement a trust and safety layer to provide a verification step to ensure that every piece of content created would be approved by your CMO. You use BrandGuard as this trust and safety layer to prevent any off-brand or poor quality content from reaching the end consumer. As high quality, on-brand pieces of content are approved automatically they reach consumers who ultimately drive higher uplift to your campaign from hyper-personalized content. A diagram of the process looks like this:

Without BrandGuard, hyper-personalization becomes either extremely costly or impossible. BrandGuard provides a way to create a new, maximizing capability by turning something that was valuable but too costly into an economically viable process. By providing trust, safety, and model governance into a generative AI system, BrandGuard can remove a lot of the fear and concern around using generative AI at scale. Let’s take a look at how it works.

A Short Technical Dive into BrandGuard

BrandGuard uses an ensemble of models to assess if a given piece of content is on or off brand. By combining a range of classical machine learning and advanced cutting edge techniques, such as multi-modal models, we have been able to greatly improve the performance of brand understanding. There are several classes of models we use to verify content:

• Safety - Would a normal person find the content acceptable?

• Quality - Does the image or text have errors or seem like it was generated?

• On-Brandness - Does the content conform to what we think our brand ideal is?

• Style Adherence - Does the content adhere to our style requirements?

• Campaign Feel - Is the content in line with the current campaign look and feel we are going for?

• Compliance - Does the content cause any legal concerns?

The outputs of all the models that fall into each of these categories culminates in a single BrandGuard Score that determines if the content is on or off brand. We provide a way for you to agree or disagree with any classification that is made which helps strengthen the tailoring for your brand. In order to create the best models we can, we’ve collected over 5TB of content and that number is growing everyday.

For the best experience with our system what do we require from you? Historical content that you believe is on-brand and your associated style guide. Having content that used to be on-brand but no longer is or other content you want to negate helps too but is not required.

If you’d like to use BrandGuard in your hyper-personalization engine or if you’re interested in a demo, please contact us.