Ozon Search Working Principles

Ozon search is a product selection and sorting system. The search selects promising seller products and distributes customer attention between them.

Ozon determines if a product is promising depending on the product price, quality, and delivery speed. Sellers can influence their product positions, making these parameters more favorable to customers. Products in high positions get more views and sales.

How to raise a product in search results

The main principles of the search engine:

• Focus on orders. Our goal is to help customers choose a product and place an order.
• Speed. We make search results respond to seller and customer actions as quickly as possible.
• Transparency. We openly explain how the search works.
• Adaptability. Search results take into account the customer query, region, and behavior.

Ozon search workflow

When a customer submits a search query on Ozon, the search proceeds in several stages.

1st stage: query generation and selection of products

Before selecting products, the search engine transforms the query: adds synonyms, reduces the query to a single form, and excludes conjunctions and prepositions. For example, the “builder Johnson’s emerald ship” and “emerald ships by builder Johnson” queries result in the same output.

After that, the system searches for words from the query in the product attributes:

• name;
• category;
• brands;
• description;
• color, size, material, type, and other characteristics.

The system doesn’t search for query words in the store name, reviews, or questions.

Fill out product attributes: the more detailed the description, the higher the chance customers see a product in search results.

2nd stage: basic ranking layer

The system selects the most promising products based on:

• the similarity of the query words to the product attributes;
• the number of orders and delivery speed.

All products receive a base relevance score. Several thousand products with the highest scores move on to the next stage.

3rd stage: middle ranking layer

The search system uses a machine learning model to assess the probability that customers will purchase the products. The model is trained on past customer behavior and identifies patterns among factors. A factor is a number that characterizes a product, such as delivery speed or price.

The system calculates the factors for each product, compares them with past customer behavior, and estimates the purchase probability. For example, if customers more often choose products with many reviews, then the greater the number of reviews, the higher the purchase probability.

The same product can have different purchase probabilities depending on the query, region, and customers.

Factors considered in product ranking

Many factors are built on customer behavior. For example, order placement, viewing products, or adding them to the cart. Sellers can also influence product rankings in search results by changing the values of certain factors.

At the same time, different factors have different predictive power and a different effect on the purchase probability.

Main groups:

The factor influence depends dynamically on the context. For example, for broad queries like “smartphone”, the search selects smartphones from different price segments: price-related factors have more weight. If a query includes a specific smartphone model, then reviews are more important. If the query is from a remote region, the factors related to delivery are more important.

The system assigns a score from 0 to 1 to each product. The closer the score is to 1, the higher the product position in search results and the purchase probability.

4th stage: boosters

At this stage, increasing coefficients are applied to products that are ranked from 0 to 1. They’re called boosters.

Booster types on Ozon:

• Booster for participation in Ozon global sales, for example, the discount marathon, “11.11” and “Black Friday”. The coefficient depends on the specific promotion.
• Booster for products with a favorable and moderate price index.

Example

After passing through three stages of search, a product named “blue briefcase” gets a score of 0.811. If this is a product with a favorable price index, a booster is applied. For example, 1.1. In this case, the final product score is 0.811 × 1.1 = 0.892. If a product meets the conditions of several boosters, the booster coefficients are summed up.

As a result, we can evaluate the organic relevance of the product.

5th stage: paid promotion

Paid promotion also affects search results position. We assign a final score to each product using the formula:

organic weight × organic score + promotion weight × promotion score.

Variables in the formula:

• Organic weight and promotion weight: the regulatory coefficients that support competition and maintain the search results quality.
• Organic score: an evaluation of the product organic relevance calculated at the previous stage.
• Promotion score: an evaluation of the impression cost in all paid promotion types. The metric depends on the rate per click or rate per order and the probability of the customer interaction with the product. It’s calculated by the formula:
  Promotion score = product click probability × rate per click + product order probability × rate per order.
  If the product doesn’t participate in a paid promotion, the promotion score is zero.

After this, we sort the products in descending order of the final score.

Raise products in search results

• Use promotion tools. Select the tool in the Promotion section.

  Learn more about product promotion tools

• Reduce delivery times for your products. The average delivery time depends on the shipping cluster: the closer the product is to the customer, the faster the delivery. This makes the product more attractive to customers and improves its promotion in search results.

• Collect feedback from customers. The higher the product rating and the more reviews it has, the higher it appears in search results. You can offer customers points for submitting reviews of your product. In your account, open the Promotion → Reviews for points section to enable the tool.

  Learn more about reviews for points

• Fill out product attributes completely. The more information about a product, the easier it is for customers to find it. To edit product information, open the Products section or change information about several products at once using the XLS template.

  Learn more about editing PDPs

Problem solving

Can’t see new product in search results

New products appear in search results within minutes.

If a product doesn’t appear in search results 20 minutes later:

• check if the product price, name, and description are filled out;
• check the indexing error report;
• contact support.

Previously created product is missing in search results

A product may be excluded from search results if it has low views and sales over an extended period. This way, the system maintains high relevance of search results for customers. The product remains available via a direct link and is displayed in recommendations.

To increase product visibility in search:

• use promotion tools;
• optimize the price and improve your PDP: add images, clarify the description, and include relevant characteristics and keywords.

Product is missing in search results for expected queries

Search results can contain hundreds of products, and your product can be anywhere in this list. You can check which search queries customers use to find a product using the “Where the product is in the search” tool.

Learn more about the “Where the product is in the search” tool

You can understand why a product ranks the way it does in search results using the “What affects your search ranking” tool.

Learn more about the “What affects your search ranking” tool

Consequences of boosting performance using bots

Bots view products but don’t purchase them, thereby ruining behavioral metrics. The system may consider a product with many views and no orders to be uninteresting for customers. Then the PDP appears lower in search results. Use product promotion tools to increase the reach of PDPs.