Lamoda implements ML model to predict product scrap

2024: Implementation of a model to determine product scrap

Lamoda has implemented and is developing a machine learning model that predicts marriage among goods. This model has increased the speed of processing returns by 2 times after fitting by buyers. The company announced this on June 24, 2024.

All items in the warehouse that customers returned are first checked through the ML model. It performs daily scoring of received items and assigns one of three statuses:

• Everything is fine with the thing, it is sent for storage, gets to the showcase and is sold further.
• There may be problems with the thing, a quick check by a warehouse specialist is required.
• There are definitely problems with the thing, it is sent to the quality service for careful inspection by a specialist. If the condition of the product does not allow it to be sold anymore, then it will be sent for charitable purposes.

The machine learning model of marriage prediction is the company's own development. To create and integrate the model, 20 million lines of historical data were taken and more than 60 characteristics about the product and order were taken into account - including characteristics received from other ML models.

At Lamoda, we pay great attention to the quality of the product and how customers get experience, interacting with us at any stage of purchase. The introduction of such an ML model allows us to more closely monitor the marriage of goods and influence whether a person will buy this thing next time. In addition, we save a huge amount of human resources in the warehouse, "said Tatyana Umryaeva, Managing Director for Product Lamoda. - In our plans to continue to improve the model and work to increase the accuracy of marriage forecasts.

As a result of the preliminary audit, the quality and accuracy of the audit has significantly increased. Prior to the implementation of the ML model, defects in clothing and footwear were determined by employees after manual inspection of each item.