Embeddings: Why I Believe they are key for Retail Media & Marketing 3.0

Embeddings: The Key to Unlocking Customer Insights in Retail Media
To truly know your customers, Retail Media and Marketing experts must go beyond surface-level interactions. Much of the valuable information is hidden within unstructured data like social media comments and product reviews. Embeddings offer a way to unlock these crucial insights.
Unstructured data encompasses a wide range of formats that lack a predefined data model. This includes text, image, audio, and video data.
Okay, but what is Embeddings?
Embeddings are a way to represent text, images, or videos as numerical data. This allows machine learning models, particularly in the field of generative AI, to better understand and process the relationships within these inputs. By identifying patterns in large datasets, these models create embeddings that capture complex meanings and semantic relationships specific to the content.
Technically, embeddings transform text, images, and videos into arrays of floating-point numbers, also known as vectors. These vectors, with their dimensionality determining the length of the array, encapsulate the essence of the input data. For instance, a text passage could be represented by a vector with hundreds of dimensions. By calculating the numerical distance between these vector representations, applications can gauge the similarity between different pieces of content.
Real-World Applications of Embeddings
You encounter embeddings daily without even realizing it. Google Search uses embeddings to understand your queries and deliver relevant results.


Music streaming services use them to recommend songs based on your listening history.
Let’s look at couple of scenarios for Retail Media
Let's consider a grocery retailer with a customer, Customer A, who frequently purchases baking products and leaves reviews like:
- "Love this organic flour! Makes the best bread."
- "This gluten-free cake mix is amazing! So moist and flavorful."
- "Brand X always delivers on quality. Their baking products are consistently good."

These reviews, expressed through text, images, and even video UGC, provide valuable insights into Customer A's preferences. Embeddings go beyond simple keyword analysis, capturing the positive sentiment, brand affinity, and preference for quality ingredients.

Now, let's introduce Customer B, a fashion enthusiast who leaves reviews like:
- "This dress is so cute! Love the floral print and the flowy fit."
- "Obsessed with this pastel pink sweater! Perfect for spring."
- "This vintage-inspired blouse is gorgeous! The lace details are beautiful."

Multimodal embeddings, generated from product images and reviews, capture Customer B's visual style preferences, such as "floral prints," "pastel colors," and "vintage aesthetic."
Personalized Recommendations with Embeddings
With embeddings, the retailer can provide highly personalized recommendations.
- For Customer A, the system might suggest a new organic, gluten-free pancake mix with rave reviews, as its embedding aligns closely with her preferences.
- For Customer B, a new dress with a similar floral print but in a different color could be recommended, based on the nuanced visual connection captured by the embeddings.
Embeddings enable retailers to move beyond basic demographic data and purchase history, delving into the nuances of customer preferences. This deep understanding allows for more targeted marketing campaigns, personalized product recommendations, and ultimately, increased customer satisfaction and loyalty.
Google Cloud Embedding APIs
Google Cloud offers various Embedding APIs, providing powerful tools for businesses to leverage the power of embeddings. By incorporating these APIs into their retail media strategies, companies can unlock valuable customer insights and drive business growth.
Embeddings are transforming the way retailers understand and interact with their customers. By capturing the true meaning behind customer interactions, embeddings enable personalized experiences and unlock valuable insights that drive business success.

For marketers and retail media experts, embracing embeddings would be a very interesting way to derive even more interesting insights beyond traditional Machine Learning.
Let me know if you would like me to write more about this topic!
Member discussion