Automating E-commerce SKUs Creation and Shipping Estimates with Amazon Bedrock

Managing a digital storefront in Nairobi while juggling a demanding professional schedule means time is your most expensive currency. Every single day, the WhatsApp groups I belong to—run by Nairobi downtown wholesalers—flood my phone with dozens of product photos and raw wholesale prices.

 

To list these consumer electronics and kitchen appliances on my store, I previously had to sit down, analyze each photo, and manually write SEO-optimized descriptions tailored to a Kenyan audience. I already had the idea of connecting AI directly to my e-commerce backend to templatize and automate this workflow. When I saw a recent LinkedIn post demonstrating how Generative AI could take a product photo and instantly write sales copy, it completely validated my concept. But as I started building this pipeline for my own store, I realized we could push the AI a step further to solve a massive logistics bottleneck.

In e-commerce, local delivery providers and freight forwarders need one crucial piece of data before they can calculate a shipping price: the physical dimensions of the product. Calculating volumetric weight requires length, width, and height. If we were already passing the product image to an AI to write the sales copy, why not ask the AI to estimate the dimensions of the box at the same time?

Think of Amazon Bedrock as a universal translator and creative assistant rolled into one fully managed service. Instead of building and training a massive neural network from scratch, Bedrock provides a single API to interact with industry-leading foundation models. For this use case, we need a multimodal model—an AI that can "see" images and "write" text simultaneously.

 

 

The process begins the moment a wholesaler drops a photo into the WhatsApp group. I uploaded it Amazon S3 bucket. S3 acts as our highly durable digital warehouse for media. When the image lands in the bucket, it acts as the trigger for the next phase of the backend pipeline.

 

We use the AWS SDK to securely call Amazon Bedrock. We pass the image and our carefully crafted prompt integrated in our backend to a multimodal model like Anthropic Claude Sonnet 4.5 (available natively through Bedrock), which excels at fast, highly accurate visual analysis.

 

The magic happens in the prompt engineering. We are not just asking for a description; we are setting strict constraints and operational personas. The prompt instructs the model to act as an expert Kenyan digital marketer. It is told to extract the visual details of the product, write an engaging, SEO-optimized description that resonates with the local consumer base, and crucially, analyze the contextual clues in the photo—like hands, tables, or standard USB ports—to estimate the packaging dimensions in centimeters.

 

We instruct Bedrock to return the output in a strict JSON format. This ensures that the generated sales copy and the estimated length, width, and height are neatly separated. This JSON payload can then be parsed and pushed directly into the e-commerce database, seamlessly updating the digital storefront and instantly calculating the estimated delivery fee for the customer at checkout based on those AI-guessed dimensions.

 

By chaining together Amazon S3 and the multimodal capabilities of Amazon Bedrock, what used to be a tedious, hours-long data entry shift is now an automated pipeline. It proves that cloud AI isn't just about generating text; it's about looking at your specific operational bottlenecks—like missing shipping dimensions for couriers—and engineering a single API call to solve them all at once.