A complete guide to vector search in 2023

Alan Yiu


In today's ever-evolving digital landscape, the demand for efficient and precise search capabilities within the enterprise is at an all-time high. Senior decision-makers and executives understand the critical role that effective information retrieval plays in driving productivity, decision-making, and innovation. One revolutionary approach that's transforming the way businesses navigate their vast repositories of data is vector search.

Vector search: A paradigm shift in enterprise information retrieval

The emergence of Large Language Models (LLMs) and generative AI has ushered in a new era of intelligent search experiences. Gone are the days of sifting through endless search results. Now, users can interact with AI chatbots, like ChatGPT, to obtain instant and contextually relevant responses. 

However, despite being powerful, LLMs often generate responses based on incomplete or biased knowledge. To bridge this gap, enterprises are turning to vector search systems. These systems can provide LLMs with accurate and trustworthy information.

Vector search leverages numerical representations called embeddings to capture the semantic essence of the text, enabling infrastructures to grasp intricate relationships between concepts. Unlike traditional keyword-based search, vector search enhances precision and context-awareness in particular scenarios.

Benchmarking vector search performance

We conducted a rigorous experiment comparing various text embedding models to gauge the effectiveness of vector search in an enterprise context. Our evaluation included embeddings from leading LLM providers and top-performing open-source models. We employed two key metrics to assess the quality of search results and retrieval performance: NDCG@10 and R@10.

Surprisingly, open-source embeddings like E5-large, Instructor-XL, and MPNet outperformed commercial API providers such as OpenAI and Cohere in this specific case. It highlights the ongoing superiority of open-source alternatives, but it's important to note that AI is rapidly evolving.

Vector search use cases

Vector search is a versatile technology with applications across various domains within the enterprise. Some notable use cases include:

  • Document retrieval: Quickly locate specific documents within vast document repositories.
  • Customer support: Provide agents instant access to pertinent information and solutions.
  • E-commerce: Improve product recommendations and search results for customers. It will increase conversion rates.
  • Legal research: Streamline investigations by retrieving relevant case law, statutes, and precedents.
  • Healthcare: Facilitate medical research by extracting relevant studies and patient records.

Vector search challenges

While vector search offers tremendous potential, it also presents particular challenges, including:

  • Data quality: The effectiveness of vector search relies on the quality of training data. Noisy or biased data can lead to inaccurate results.
  • Resource intensity: Fine-tuning embeddings for enterprise-specific language requires computational resources and expertise.
  • Privacy and security: Handling sensitive enterprise data raises privacy and security concerns.
  • Speed and scale: As enterprises accumulate vast amounts of data, the ability of vector search systems to operate at high speed and scope becomes increasingly critical. Efficiently handling large datasets and delivering real-time responses is a significant technical challenge.
  • Interoperability: Integrating vector search seamlessly with existing enterprise systems and workflows can be complex. It requires compatibility and integration efforts.
  • User adoption: Convincing users to adapt to new search methodologies and technologies can be challenging. Providing user-friendly interfaces and productive training is essential to ensure widespread adoption.
  • Maintenance and updates: Vector search models need regular updates to remain effective. Staying up-to-date with the latest advancements in the field is an ongoing commitment.

Addressing these challenges is essential for organizations looking to harness the full potential of vector search.


Customizing vector search for your enterprise

At Glean, we recognize that each enterprise possesses unique language and domain-specific terminology. This distinct vocabulary, which includes acronyms, project codes, and technical concepts, often eludes generic text embeddings, leading to suboptimal search results.

To address this challenge, we've developed a fine-tuning method. It customizes embeddings to your enterprise's language. This tailored approach ensures that your vector search understands and retrieves contextually relevant information. Our experiments have consistently demonstrated the superiority of in-domain fine-tuned embeddings over off-the-shelf models, whether from commercial API providers or open-source options.

Moreover, our research has unveiled a compelling trend: the longer your enterprise utilizes our services, the more refined and accurate your language model becomes. Adaptation and persistent fine-tuning result in an increasingly enhanced user experience and more precise search outcomes.

Vector search as a pillar of modern enterprise search

While vector search represents a fundamental shift in semantic understanding, it is just one piece of the puzzle in delivering high-quality results for enterprise search. Glean adopts a multidimensional approach by combining vector search with traditional keyword-based search and advanced personalization. This holistic hybrid search system provides a comprehensive solution that caters to the diverse needs of your enterprise.


The paramount importance of efficient information retrieval in fostering innovation and driving growth is undeniable. Vector search has the ability to bridge the gap between LLMs and reliable data. It is a transformative technology that can elevate your enterprise's search capabilities.

If you're eager to witness the power of vector search in revolutionizing the search function within your enterprise, we invite you to schedule a personalized demo with us. Experience firsthand how Glean is redefining the landscape of enterprise search and knowledge discovery.

Related articles

No items found.
How Glean search works

How Glean search works

Discover how Glean enables users to search across all their company’s apps and discover what they need to know through innovative machine learning and indexing.

The AI-powered work assistant. Across all your company's data.
Get a Demo
CTA Section Background Shape