Ozon will help sellers predict demand for goods using machine learning

2023

Ability to forecast demand for goods using machine learning

Ozon sellers who work under the Sales from Warehouse (FBO) scheme have access to the Demand for Goods report. The company announced this on July 12, 2023. He predicts the potential demand of buyers for the seller's goods and shows in which regions he will be more in demand and in what volume. This data will help sellers schedule deliveries with the greatest accuracy, taking into account the needs of buyers. In addition, the new report will help avoid a situation where the product in demand is over and not available for purchase.

To develop the report, Ozon engineers used machine learning models that make a forecast based on a large number of parameters: customer demand, seasonality, the price of the product and its discount within the promotion, balances in warehouses, goods on the way and in planned deliveries. Separately, the forecast takes into account sales on holidays and during sales, when demand for certain goods increases. In order to more correctly predict the increase in demand and not overestimate it, the model takes into account historical sales and dynamics that were for similar periods in the past.

The seller will be able to receive a report on any product that has had at least one sale within 6 months, and for seasonal new products - even without sales (data on analogue products will be taken into account). Forecasts for 10 weeks ahead are available for July 2023: this should be enough to adjust its work and schedule future deliveries to the warehouse. The report is available in your personal account, and you can find out more about it in the Knowledge Base.

In addition, Ozon launched the Turnover of Goods report for sellers. It reflects detailed analytics on the speed of sale of both a specific product and the store as a whole. The report shows: how many goods are left and in which regions, how many are sold daily and after how many days the current goods will end in warehouses. The seller's store turnover indicator also appeared - this is the average forecast for the timing of sales of goods. The higher the number, the longer the items will be in the warehouse. With this report, you can simply adjust the assortment matrix and it is easier to build a business development strategy. Sellers can combine data from reports to pay attention to the demand of the most wrapped goods.

Creating a Moderation System

Ozon on March 20, 2023 announced that it had developed its own system for moderating product cards based on machine learning methods. The system automates the verification of new items and reduces the processing time of the card by 10 times. Thanks to this, sellers will be able to place goods on the site as quickly as possible - the speed of automated verification will average three minutes.

The system automatically checks the product cards when loading and analyzes potential violations in the description and photo. If a problem is identified, the product is not allowed to the marketplace or sent for manual verification. This moderation system uses machine learning (ML) methods: the algorithm is trained on the accumulated array of solutions of Ozon moderators and blocked cards that previously revealed violations.

The system allows you to concentrate the efforts of the manual moderation team on non-standard situations that require human participation.

The moderation technology was created by Ozon Tech's own IT team and is being phased into the marketplace's business processes. Sellers place more than one million new product names on the marketplace every day, and more than 20% of them are already undergoing automated moderation. The algorithm decides to allow the card to the site if the rules are not violated, or send it to the seller for revision if problems are found. In difficult cases, the card can go to double-check for manual moderation. In the future, most cards will be processed automatically using ML moderation.

Ozon carefully monitors the quality and legality of goods that sellers place on the site. It is important for us to exclude goods with violations from entering the site, while placing new positions on the online showcase as quickly and efficiently as possible. Therefore, we have trained our own algorithm on an array of millions of moderator solutions. It allows you to move from manual to automated moderation of cards, which sharply reduces the time for placing goods and excludes the human factor, - said Anton Stepanenko, CTO Ozon.

In addition to pre-checking the cards, Ozon Quality Control also postmoderates goods already placed on the site using ML models. This provides additional buyer protection.