Azumo Software Development

The Challenge

The primary challenge Meta encountered was the inability of their traditional search engine to effectively retrieve suppliers based on specific criteria. This resulted in time-consuming manual searches and often led to missed opportunities with potential suppliers who were hidden in the massive supplier database.

Additionally, the company's supplier database was continually growing, making it increasingly challenging to manage and maintain an up-to-date repository of supplier information. They needed a solution that could automatically adapt to the evolving supplier landscape and deliver timely and accurate search results.

Facebook approached Azumo with a need to improve its search capability for diverse suppliers within the Facebook supplier ecosystem.
They have over 3.5 million suppliers in their database, making it difficult for their Supplier Diversity team to find the right supplier for Facebook's needs.
This project was an extension of the application we built for them to track and manage Supplier information at their global conferences.

‍

The Solution

Recognizing the potential of AI-based technologies, Meta collaborated with our team to implement a cutting-edge enterprise search solution. We decided to build upon the approach outlined in our blog post "Enhancing Search Engine Results Using Domain-Based Similarity" to address Meta's specific requirements.

Azumo built a custom service called AIML. This included NER (named entity recognition) to extract supplier capabilities, products using a Natural Language Understanding (NLU) library.
AIML Service seeks to process both capabilities and products /services that each Supplier offers.
The service will label the outputs where possible and an entry will show up for each instance in which capabilities or products / services appear.
The output provides information at the company level and at the product/ service level that Facebook can use to improve their search capability.

‍

Leveraging Domain-Based Similarity

By using domain-based similarity, we created a word representation model that learned from the relationships between different suppliers and their attributes. This enabled the search engine to recognize similarities based on the characteristics of suppliers, rather than relying solely on traditional keyword matching.

‍

FastText and Unsupervised Learning

We chose FastText, an open-source, lightweight library that excels in handling rare and out-of-vocabulary words. Unsupervised learning allowed us to generate vector representations for each supplier and their attributes, even if specific terms were not explicitly present in the training dataset.

Training the AI Model

To train the AI model, we used Meta's extensive supplier database, which consisted of a diverse set of supplier profiles and attributes. We cleaned the data, removing irrelevant information and creating a training dataset of supplier attributes with domain-based permutations.

Implementing the AI-Based Enterprise Search

With the AI model trained, we integrated it into Meta's existing enterprise search platform. The AI-based search solution would now leverage domain-based similarity and FastText word embeddings to enhance search results dramatically.

Performance Improvement

To measure the performance improvement of the new search solution, we conducted a thorough evaluation. We defined specific queries to identify various types of suppliers within Meta's universe. Our analysis showed a substantial increase in the accuracy of search results compared to their previous search engine.

‍

Getting Results

With the AI-based enterprise search solution in place, Meta's procurement team has undergone a significant transformation in their supplier search capabilities.

Precision and Relevance: The new system consistently delivered more precise and relevant supplier matches, saving time and resources in finding the right suppliers for specific needs.
Efficiency: Employees across the organization experienced a significant reduction in the time required to search for suppliers. The enhanced search capability streamlined supplier discovery and selection processes.
Data Management: The AI-based solution efficiently handled Meta's ever-expanding supplier database, adapting to changes and ensuring up-to-date information was readily available.
Maximized Opportunities: The enhanced search capabilities uncovered hidden opportunities with suppliers that were previously overlooked due to suboptimal search results.
AI-Based Search: Meta has improved its ability to find the right supplier in-real time and has optimized their supplier search process.

Through a strategic partnership with Meta, our team successfully transformed their enterprise search capabilities. By leveraging domain-based similarity and AI-powered technologies, we enabled Meta to find suppliers with unprecedented accuracy and efficiency. The AI-based enterprise search solution not only enhanced their supplier discovery process.

With this successful implementation, Meta now reaps the benefits of advanced AI-based search capabilities, maximizing opportunities and gaining a competitive edge in their industry. The case of Meta serves as a compelling example of how AI can revolutionize enterprise search, making it an indispensable tool for large organizations with vast and diverse data repositories.

‍

Generative AI Enterprise Search