Back to Blog

Machine Learning-based Scoring comes to life: the Gradient Score

Norman Shi

Machine Learning-based Scoring comes to life: the Gradient Score

Just in time for the Q4 e-tail season, Gradient.io recently introduced a first-of-its-kind machine learning-based scoring service for eCommerce brands. The Gradient Score tells a brand how it stacks up against its category peers on the digital shelf. But more importantly, the Gradient Score highlights a marketers areas of strength and weakness, so that the brand can improve its performance and compete more effectively.

Generating a Gradient Score is a data-intensive process that combines thousands of measurements into a single result. We continually collect raw data pertaining to the products’ attributes as well as consumer response to those products in the marketplace. Then we pass those data through trained machine learning models to predict the impact of these measurements to the success of a brand in the marketplace.

We created the Gradient Score to be a leading indicator of marketplace success. A strong Gradient Score signals sales growth. Conversely, a low Gradient Score signals sales headwinds and work to be done. A guiding principle we use to ensure Gradient Score is a leading indicator is baking in plenty of antecedents, or inputs to success, as opposed to outputs of success. Inputs are the things the brand can affect to drive results. An example of an antecedent input is ratings. A ratings bump can often cause a bump in listing rank which can in turn lead to a bump in sales. Whereas outputs are like observations in the rear view mirror. For example, a bump in best-seller ranking reveals the downstream results of a job already well done in other impact areas.

We also designed the Gradient Score to be a tool to uncover opportunities. This works by allowing brand leaders to unpack the ingredients, or subcomponent to their brand’s overall score. Each Gradient Score subcomponent relies on a different dimension of the model to score the attributes of a brand’s products against the attributes of hundreds of other products in the category. Drilling into the subcomponents allows a brand leader to understand and target their relative strengths and weaknesses in relation to category leaders.

The foundation of our models are universal and can be applied broadly. The three classes of inputs used by our model apply to any marketplace. Product Positioning which is how well products are merchandised using descriptions, images, and pricing. Brand Presence is how well-positioned products are in the flow of interested shoppers. And, Customer Response measures aspects of customer enthusiasm and satisfaction. Initially, we’ve applied our models to Amazon, but we anticipate expanding to other marketplaces soon. 

One of the most exciting possibilities created by machine learning for brands is the ability to make connections that might have been otherwise invisible to the naked eye. For example, early identification of a surging product segment could provide a timely signal for spotlighting that segment in other channels. Or, early detection of a competitor’s entry into a niche segment that could help avoid a price war for paid traffic. Or, early identification of sagging organic volume could be swiftly shored up with low-cost awareness efforts outside the channel. Each of which could be the subject of future blog posts.

We are happy to be able to provide any brand with a Gradient Score to check their brand’s health. But it is really just the beginning of what Gradient can do for eCommerce brands. We look forward to sharing future innovations to help brands succeed on Amazon and beyond.

Learn more about the Gradient Score at https://gradient.io/score.

Share on social media: 

More from the Feed