Product Scoring Algorithm

Data Science & Insights - 21 October 2022

Data Science & Insights

Product Analytics & Data

You are curious about performance of your products. And also, you are worried about how you can manage your product scope to increase revenue. Here are the answers for all of your questions, check out our blog and learn about our product scoring algorithm.

Common problems in product management

There are a number of questions that professionals might have in their mind while evaluating their companies’ performance. And some of these questions are more beneficial when thinking about important actions to increase efficiency of products that are sold on websites, apps and marketing channels.

Here are some “general” questions that each team might be curious about the answers to. Which e-commerce metrics should we focus on while considering inventory management? Which products are more valuable? How can we increase our revenue by having the same product scope? What should be our key metrics for optimising marketing campaigns of our products? How should we decide our product listing in our website or app? Yes, you are right! AnalyticaHouse Data Science & Insights Team have another magical algorithm that would be helpful for all mysterious questions above and more, which is Product Scoring Algorithm. In this article, we will be talking about the structure of this algorithm and applications of it in the digital marketing sector.

How is the algorithm built?

There are a lot of variants for product groups in each website and app differentiating based on various dimensions such as colours, sizes, product types, season/offseason, target genders, target age groups and so on. We would be able to analyse every detail of products and product variants thanks to the Product Scoring Algorithm depending on the depth of the data we have.The minimum data requirement for this analysis are transactional data of products and current inventory information of them. We are focusing on these metrics while analysing transactional data; number of purchases, total quantity, revenue, average price and so on. Of course, this analysis would be more interesting when we have user/session-based data. We would be able to add other important metrics such as product detail views, add to cart, # of buyers, # of users who look at the product, revenue/product detail views, transaction/product detail views and so on. These metrics can be grouped into parts according to results they show;
Revenue supporting metrics : number of purchases, total quantity, revenue, average price, revenue/product detail views, transaction/product detail viewsMetrics that show customer interest : product detail views, add to cart, # of buyers, # of users who look at the product, revenue/product detail views

Metric definitions

• number of purchases: Number of transactions
• Total quantity:  Number of products that are purchased
• Revenue: Total amount of revenue
• Average Price: Average price of products
• revenue/product detail views: Revenue per product detail views for each product
• transaction/product detail views: Conversion rate (based on product detail views)
• add to cart: Total number of products that are added to cart
• # of buyers: Total number of customers who purchased related product
• # of users who look at the product: Total number of users who viewed related product

Application of the model

The algorithm can be applied to each product level and can bring insights for bigger picture and tiny details.

The example below shows a category/product level for an e-commerce company;

• Level 1: Shoes/Bags
• Level 2: Shoes
• Level 3: Sneakers
• Level 4: X Brand’s Sneakers
• Level 5: X Brand’s White Sneakers
• Level 6: X Brand’s White Sneakers with Size:38

How is product score calculated?

is basically formed from two parts. The first part is based on calculating ratio of lower levels (eg. X Brand’s White Sneakers with Size:38) in totals of higher levels (eg. X Brand’s Sneakers) by using metrics that we discussed above, and we will explain this part by showing an example. The second part is dependent on current inventory information of products, and this part is very crucial when we take inventory management into account for short term and long term. 

Example above shows a product scoring calculation by using metrics add to cart and quantity sold in addition to instant inventory score ( if there is inventory it is 1 and if not it is 0). In this example distribution of level 6 in total of level 5 for add to cart and quantity sold metrics and inventory score are used as three multipliers of product score of level 6. 


After calculating the score of each row (each breakdown, for this example it is level 6), we calculate the score of the higher level ( in this example it is level 5 ). We take the sum of the score of all breakdowns for level 5 and as a result we have the total score of level 5. For example; for X brand’s White Sneakers the total score can be calculated as :0,0747 + 0,0245 + 0,0036 + 0,0028 + 0,0251 + 0,0000 = 0,131In this example X brand’s Black Sneakers has the highest score (0,193) and X brand’s Red Sneakers has the smallest score (0,071).

How can we use the results?

This analysis can be applied to all levels from bottom to top. And of course, this analysis can be enlarged with additional metrics that we talked about before and results can be calculated in more detail. In addition to that, we can change the importance of each metric in the calculation step of the product score by using Variable Importance Analysis which is another outstanding algorithm that our team improved. Last but not least, results of the analysis can be used for updating product feed automatically for marketing campaigns.