Improving Literary Marketplace Product Recommendations With AI

Introduction 

Rekhta, a leading online bookstore specializing in South Asian literature, boasts a vast collection spanning English, Hindi, Urdu, and several other regional Indian languages. The literary marketplace is part of a non-profit organization dedicated to the preservation and promotion of Indian languages. While their commitment to linguistic diversity attracted a dedicated customer base, it also presented a challenge: effectively navigating a vast and multifaceted inventory to offer recommendations and maximize customer satisfaction.

Challenges Faced

1. Discoverability of Products: With thousands of titles across diverse languages and genres, Rekhta was looking for a better way to make their products easily discoverable. Customers often found it difficult to navigate the vast inventory and find books relevant to their specific interests and needs. 
2. Limited Data Insights and Awareness: While data generated from several million visitors on the site was available, the team had limited expertise and tools to extract meaningful insights and translate them into actionable recommendations. This limited understanding of customer behavior and preferences across different languages made it difficult to tailor recommendations and marketing strategies effectively.
3. Underutilized Indian Languages Collection: With its rich literature collection across Indian languages, Rekhta aimed to create a specific strategy to better engage readers and promote lesser-known gems within these genres which were often not discovered by users. 

Using AI and ML, Rekhta aimed to not only improve customer experience but also foster a deeper engagement with its diverse Indian language collection.

Solution Implemented

Rekhta partnered with RapidCanvas to implement market basket analysis and product recommendation using AI. The solution involved the following steps:

Data preparation

Rekhta has access to data of different types including customer interactions, on the website, purchase history, book ratings and reviews, and browsing behavior. The Customer data was cleaned in a few simple steps using the RapidCanvas platform and by identifying and removing duplicates, errors, and missing values. Additional metadata about books, such as genre, author, and publication date, and language, was incorporated to enrich the customer data and create a more comprehensive understanding of reading preferences.

Feature engineering

Features or attributes were extracted from the processed data, including the frequency and recency of purchases, categories of purchased books, and co-purchased items. Content-based features like genre, author names, and thematic keywords were also created.

Model training

1. Market Basket Analysis: A market basket analysis algorithm was trained on the enriched customer data to identify patterns and hidden correlations between book purchases using the sales data. This model uncovered patterns in buying behavior revealing what kind of books readers tended to buy together.
2. Recommendation Engine: An ML model was trained using the catalog-based information and feature-engineered data. This model learned to predict which books would be most relevant to customers browsing a specific book’s product page, based on the data points available including title, author, genre, and description.

These two models generated suggestions for each customer at different stages of the buyer journey, showcasing relevant books from across the diverse inventory, including hidden gems and lesser-known titles.

Implementation

To maximize visibility and impact, the recommendations were implemented in two key stages of the customer journey:

Product Pages: On individual book pages, content-based recommendations were displayed based on the specific book being viewed. These recommendations highlighted books with shared themes, similar authors, or complementary content, encouraging exploration within and across languages.

Checkout Page: At the checkout page, after adding a book to their cart, customers were presented with additional recommendations related to their chosen item, based on the market basket analysis carried out. This final nudge further increased the chances of discovering relevant titles before completing the purchase. 

Results and Benefits

Increased website engagement

Product recommendations led to a month-on-month increase in website metrics including sessions and sessions with clicks as well as add-to-carts.

Increased customer satisfaction

Customers experience a more enjoyable shopping journey, with relevant recommendations leading to quicker discovery of desired books and hidden gems.

Data-driven insights

Rekhta gained valuable insights into customer preferences and reading trends within each language community, allowing them to tailor their marketing strategies and curate their inventory accordingly.

Conclusion

Rekhta's successful implementation of AI demonstrates the power of technology to bridge the gap in literature access. This comprehensive approach followed by RapidCanvas helped unlock the hidden connections within Rekhta’s vast collection and deliver powerful recommendations that ultimately enhanced customer experience and boosted engagement across all Indian languages. By integrating recommendations on both product and checkout pages, Rekhta ensured their suggestions were visible at key moments in the customer journey, leading to a more enjoyable and efficient shopping experience. This demonstrates how AI can empower businesses to cater to the unique needs of multilingual audiences and unlock the full potential of their literary treasures.