Efficient Product Sequencing Recommendation Model
Use Case Description
The Challenge
An e-commerce client was using a rule-driven product placement approach on its website to sequence products. In this approach, pre-defined rules were used to calculate a score for each product based on a formula that determined the sequence. The formula assigned weightage to different behavior metrics, including previous sales, product price, inventory on hand, etc. The e-commerce experts had to assign the weightage to these metrics based on their experience.
In this way, the whole process was dependent on human judgment. The entire process was stagnant and irrelevant because of changing customer behavior. The client wanted an intelligent and dynamic method based on data rather than human judgment to assign weights to different metrics dynamically and avoid human error.
The Solution
Developed 2 ML models based on the supervised ML regression model. These models predict the next day’s ‘product views’ and ‘sales dollars’ for each product by using the historical data. To rank the products, we used the predicted views as qualifiers.
These models use the attributes based on consumer buying patterns. As the trends change over time, the new patterns present in the recent data get accommodated in the auto-refreshed model. The current model’s refresh frequency is x days, that auto determines weights based on the historical data.
The Implementation
ML models are developed by data scientists in the team, after preliminary A/B testing they are hosted on UiPath Ai Center in a standalone environment to keep the proprietary data in-house.
The ML model has learned from 1.5+ million transaction data. System performance of both rule-based and new ML-based models were evaluated & compared against each other & 83% improvement is recorded with training data. The ML algorithms were applied to 58 of product categories to search and browse products.
Value Impact
• Out of 58 product categories, 45 of them saw improved revenue per session, add to cart, conversion rate, etc., which translates to a 78% success ratio.
• Some of the most prominent product categories that benefited from the model were Women’s jewelry, Baby Girls and Boys, Dresses, and Men’s Shoes.
• In search, 100% of product categories benefited using ML algorithms, while in browsing, 78% of product categories benefited using the same.
AS-IS WORKFLOW, TO-BE WORKFLOW
Other information about the use case
Industry categories for this use case: Customer Service, Information Technology and Services, Operations
Skill level required: Advanced
UiPath Products that were used: UiPath Studio, UiPath AI Center
Other applications that were used: Python, sci-kit, pandas, numpy, vscode
Other resources: [2102.04238] Amazon Product Recommender System
http://infolab.stanford.edu/~wangz/project/imsearch/climate/KBS19/fernandez.pdf
What is the top ROI driver for this use case?: Accelerate growth and operational efficiency
Don’t forget to share it on your social media followers to cast their vote for your use case! Stay connected 
.
Anil_G
Yoichi
Shiva_Nikhil
Sara_Piana
Palaniyappan
Morten_Rasmussen
pere
lrtetala
Anirudh_Tugawe
jlpath