Our client is a multibillion-dollar retailer with thousands of locations across North America. The client has an equally substantial online store, featuring tens of thousands of products, including a wide range of home décor.
The client wanted to improve the relevance of eCommerce search results, product recommendations, and labeling across the entire home décor category. To achieve this at the necessary scale, an AI-powered process is ultimately required.
But as ever, AI is only as good as the data it’s trained on. Enabling AI requires a sufficient volume of appropriate – and appropriately labeled – data. That’s where our team came in.
When an eCommerce site categorizes products, displays a related product, or delivers a search result, the quality of the experience is determined by the relevance of product the consumer sees. Relevance isn’t easy to achieve, however.
Human judgment isn’t scalable. Yet without accurate training data, AI can’t understand that tables and chairs go together, that a honey strainer isn’t beekeeping equipment, or that this cushion looks atrocious with that couch. So how do you bridge the gap between automation and human intuition?
To escape this dilemma, we recruited and trained a team of highly qualified resources to label data correctly and create a quality foundation for AI implementation.
We quickly realized we needed a foundation of deep expertise – not just people with an eye for home design and decor, but true subject-matter experts.
Relying on our proprietary OneForma platform, we recruited specialists in their fields – including a design consultant and a TV set designer – to help us ensure that product recommendations made sense together.
With the help of our Global Learning team, we created a self-paced online learning module to train the labeling resources on the various home design and décor tasks.
We set objective and repeatable guidelines for labeling, recruiting, training and certifying an initial team through OneForma. We’re continuing to ramp up, adding dozens of resources over time.
The resources rigorously checked and labeled the data, carefully reviewing over 700 products and sharpening the quality and accuracy of search results, sample matching, parent-child categories, complementary product suggestions, and more.
The client emerged from the engagement with a vastly improved dataset and the foundation on which to train AI to suggest more rigorous, relevant product selections in future.
Not only that, we set the metrics for what a team can be expected to achieve, how to troubleshoot, and how to train – in other words, the groundwork that will allow us to grow this effort and generate quality data at scale.