How BrandGuard's Models Work

BrandGuard is a brand governance platform that helps brands ensure the content that they, their partners, and vendors create is on-brand. There are multiple key use cases for brands that address the growing needs of brands and brand governance teams. The engine that drives the BrandGuard technology is a suite of AI and machine learning models examining a piece of content from multiple angles. To shed some light on the process, we're detailing a simple explanation for how each of our checks works below. Most of these are in the application, but a few are in development.

Safety models ensure that a normal person would find the content acceptable. While all brands fall on a continuum for these checks, we have ways to modify our scoring system to account for what a brand finds acceptable.

• Image quality: Generative AI can create a lot of content that does not look realistic, such as misshapen people, or has visually artifacting that is unacceptable for professional marketing content. These models identify visual anomalies that can affect the overall quality of an image.
• Nudity: Determines the level of nudity in an image and if it is acceptable. For example, the acceptable threshold for nudity levels differs between a family restaurant and a lingerie company.
• Violence: Determines the level of violence in an image and if it is acceptable. For example, the acceptable threshold for violence will differ between an MMA promotion and a coffee shop.
• Profanity: This model identifies the amount and location of profanity in text and provides a score based on the acceptability of the brand. While some brands are edgier, most brands keep their profanity to a minimum.
• Hate speech: This model identifies any offensive speech that would be deemed as hate speech.

On-Brand models ensure that content has the same look, feel, and voice for how you've built your brand. In essence, these checks assess if your CMO would approve the content.

• Visual brand consistency: Every brand has an intentional look and feel to their content. These models examine historic content of the brand to provide that understanding. That understanding can then be used to assess new pieces of content.
• Textual brand consistency: Similar to visual consistency, each brand has an intentional voice for how it sounds in text communication. These models understand how a brand sounds and assess new content to determine if it sounds like it is coming from the brand.
• Video consistency: In addition to images and text, video is a widely used medium by brands. These models assess if video content adheres to the look and feel that the brand wants to portray.
• Tone and Voice: Each brand wants to speak in a specified tone and voice, with different words being emphasized and other words being removed from use. These models check to make sure tone and voice guidelines are followed.
• Complex rules: Most brands have complex requirements for how their content needs to be composed. These requirements are highly specific and do not generalize well. Our complex rule interpretation engine is able to judge content with specifics such as "if an adult is present, they must be eating a hamburger on a wooden table, and if children are present, they must be eating vegetables with a smile on their face."
• Textual sentiment analysis: Most brands and campaigns have certain emotional sentiments they wish to convey. These models assess if the content is evoking the right emotions.
• Visual emotional consistency: Similar to text sentiment analysis, BrandGuard has adapted emotions to visual content as a consistency check. This model assesses if the imagery is falling within the desired emotions.
• Consistency explainability: BrandGuard uses many advanced AI models that can appear to be black boxes. However, brand managers need an understanding of why brand content that has been flagged as correct or incorrect. These methods and models provide an explanation of what needs to be changed.
• Text and Image Coherency: Text on imagery should align with the imagery in the content. These models check for coherency between text and imagery. For instance, text proclaiming a sale on dresses should not be placed next to imagery of suits.

On-Style models verify that the rules within your style guide are being adhered to. Visual style and design correctness have measurable impacts on revenue and customer acceptance.

• Correct logo: This check determines if the piece of content is using the current and correct version of a logo. By providing older logos, we can also determine when a phased-out version of a logo has been accidentally inserted.
• Distorted logo: During the content creation process, logos can become accidentally rotated, resized, or contorted. This model checks to make sure all present logos are distortion-free.
• Logo clear space: For proper design and readability, logos have requirements about how margins or clear space around the logo. This model checks to make sure that no other edges or graphical elements encroach on the logo.
• Correct logo colors: Templates and color palettes can mess with inserted images. This model ensures that a logo has the right colors on it while still accounting for image compression distortions that may happen when flighted.
• Correct Font: Fonts provide a visual voice to brands. Thus, making sure that production teams are using the correct fonts is paramount to quality design and style. These checks look over all text in an image to ensure that each block uses only approved fonts.
• Background and graphical colors: Every brand has acceptable color palettes along with specifications of where and when those colors can be used. These models identify if the correct color has been used in the correct way.
• Grammar checks: As a brand manager, whether you are writing in the AP, Chicago, MLA, APA styles, or your own homegrown style, you want your text to look a certain way. This model ensures that your text and writing rules that you've established for your brand.
• Correct fine print: Some content contains fine print that is routinely updated. This model identifies if the correct text is being used for fine print.

Compliance models help ensure that content aligns with various rules, laws, and regulations.  These models are in development and we welcome any feedback or concerns around this area.

• Copyright check: This model checks against up-to-date copyright databases to determine if there are any violations.
• Trademark check: This model checks against up-to-date trademark databases to determine if there are any violations.
• Pharmaceutical rule enforcement: The marketing rules for pharmaceuticals are complex and highly costly if violated. These models check compliance and adherence to FDA regulations.
• Financial services rule enforcement: The marketing rules for financial services are complex and highly costly if violated. These models check compliance and adherence to laws and regulations.
• Healthcare rule enforcement: The marketing rules for healthcare are complex and highly costly if violated. These models check compliance and adherence to laws and regulations.

This provides an overview of the various models and checks that BrandGuard performs and what's coming in development. If you're interested in seeing how it works, please reach out to schedule a demo.

‍