CHALLENGE
Writing a product description that highlights relevant features can be a laborious task to undertake
RESULTS
Turnaround time has reduced significantly, enabling them to take their products to market faster
SOLUTION
Automate product descriptions
Creating engaging product descriptions and highlighting distinctive features of a garment
Writing a product description that is well-crafted, succinct, and highlights relevant features can be a laborious task to undertake. So, we decided to automate it.
Product descriptions have a direct impact on a customer’s buying behavior, making them critical touchpoints. A well-written product description should address every FAQ, highlight relevant details, and be compelling enough to engage customers.
This industry demands skilled teams who not only understand fashion but have creative writing skills. This is not easy as descriptions cannot sound repetitive. To do this for 100s or 1000s of products is incredibly time-consuming and labor-intensive.
CURRENT SOLUTION
Companies struggle to find the right people, and they find it challenging to retain these people. This is because it's an operations task that has to be performed under tight deadlines. Otherwise, it delays the listing of products, resulting in a loss of sales.
HOW AI STUDIO HAS SOLVED IT
With the help of AI Studio, we used AutoML Models to train Category classifiers to recognize the distinctive features of a garment.
At first, we decided to train the model in a way that it could identify the most prominent features of the garment. For instance, hemline style. To do this, we curated a list of values for every Top-level category classifier- straight, round, asymmetric, tulip handkerchief, high-low. We made sure to exclude similar values to prevent any confusion in the ML model.
The training was further enhanced by gathering a collection of images. 80% of these were eCommerce images, some with plain backgrounds and a few that were crowdsourced. The bias in this data collection was intentional as it was congruent to the final use case.
After running the data under a rapid QC, the images and descriptions generated had an accuracy of over 86%, based on expert evaluation.
With the creation of this robust model, we can now keep adding to the bank of description templates and fine-tune them based on Brands' requirements and maintaining tonality at once.
BUSINESS IMPACT
We have converted Content writers into Content editors with full control of the end content. This has reduced their turnaround time significantly, enabling them to take their products to market faster. This process can now be managed by anyone who has a proficient command of English.
Contact us to learn more.
MODEL BUILDER: Shruthi – Fashion Analyst and Classifiers Team.