How BrandGuard Understands Complex Styles

Comprehension of Tone and Voice

Brands are about communication. Communicating to the world about what they are about, stand for, and value. However, the tricky thing about brands is that they are influenced by their interactions with consumers and the public. At the same time, just like brands, language is simultaneously exact and fluid. Because of this, we at BrandGuard are routinely asked how we can determine if a piece of copy adheres to the Tone and Voice located in the brand style guide.

Let's look at an example. Two brands might use the same words to define their brand but use subtly different definitions of those words. For instance, two brands might both define themselves as "Personable" and "Simplistic". Let's call them Brand A and Brand B and examine their differences in definitions.

Brand A
Personable: We want customers to feel like they are talking to an old friend. Language should be familiar and casual.
Simplistic: We use down to Earth language that could be understood by a five year old. Anything more complex should not be used.

Brand B
Personable: When speaking to customers, they should feel like their needs are being taken care of. We want to relate to people as to ease their stress and let them know that someone is watching out for them and their concerns.
Simplistic: We break down complex topics to their root representation. We convey large topics in easy to understand ways. We want to inform our customers and make them more knowledgeable.

Clearly, there is a difference in how these brands will communicate despite both using the same words for their brand values. Using different or expanded definitions of existing words is an area that can cause confusion. Here’s how the language might differ in addressing a customer concern:

• On-brand for Brand A but not Brand B: "Oops! Let's sort that out for ya."

• On-brand for Brand B but not Brand A: "We've heard your concerns and are dedicated to resolving them."

• On-Brand for both Brand A and B: "We're here to help. What's bugging you?"

• Not on-brand for Brand A and B: "Addressing the recent inquiries, we have implemented a multi-tiered solution."

While each of those statements is a way to address a customer's concern, they differ greatly in the Tone and Voice that are conveyed. Detecting and determining this type of nuance is a tough problem that many brands spend a lot of time working on and thinking about. Most people can discern the difference after the fact, and that’s the goal of just one BrandGuard’s models.

There’s a lot that goes into creating quality models that can automatically check the tone and voice of these two different brands. Naively, one might think, "Oh, I'll just use ChatGPT to understand the tone and voice of my brand after providing some definitions." Well, we ran that test across 500 samples for 10 brands, through various models, and the results are below.

GPT3.5: 54% Precision*
GPT4: 74% Precision*
Google PaLM2-Bison: 51% Precision*
BrandGuard System: 96% Precision*

How is this possible? First, we take the definition and put it into the context of our larger language understanding. This provides a compass for our other models and decision-making engines. Next, we use text samples that are in the style guide coupled with other samples that a brand has supplied. This part is core because, by definition, existing content that brands have released has passed the guideline requirements. These approved assets are encoded with information that demonstrates the correct application of the rules.

The question you might be asking is, well that's great, but what happens if we want to change our brand or tweak our tone and voice? How does your system work when there is no prior content to assist with? It is true that performance will improve with more samples, but we find that the best brands provide some samples in their style guide about what is on-brand and what is off-brand. The BrandGuard system extracts these samples, and you can provide more if you wish. The same process above still applies. It may take some work to tune a new brand direction, but we believe that BrandGuard assists brands in their decision-making of different "golden samples" by allowing them to iterate with real-time feedback.

Consistency in communication is the hallmark of successful brands. With BrandGuard at the helm, brands can be assured of a quality control system that meticulously checks their content, ensuring it captures their essence with unparalleled precision. Leveraging approved brand assets and understanding the intricacies of brand language, BrandGuard reviews messages to ensure they are unmistakably on-brand.

*Note: Precision is a variant of accuracy that represents the likelihood that when our system predicts something is on-brand, it actually is on-brand. Precision is a metric we use instead of accuracy because we believe that false positives are worse for a brand than false negatives. It is more harmful for our system to say something is on-brand when it is, in fact, off-brand than to say something is off-brand when it is actually on-brand.