Uncovering search abandonment
Overview
Conducted foundational research to help understand why employees were abandoning searches on our intranet and enabled the team to make foundational improvements to the content architecture before committing to a new search engine.
Timeline
Sep'20 - Oct '20
Partners
Product, Eng
Methods
1:1 interviews, Topic modeling

Challenge
As Walmart onboarded large cohorts during the pandemic and increased cross-team co-location, effective knowledge management became critical. While the Enterprise Search team needed a significant investment in improving its intranet, executive leadership was cautious due to persistently high abandonment rates in the existing enterprise search. While the team knew from past research that our search was lacking, they did not understand what improvements to make to the content architecture to help people find what they need.
Learning objectives
What are the patterns of abandonment in industry standard search engines?
What are searchers’ self-reported patterns and attitudes around their search experience?
What were they really hoping to find that was missing?
Research approach
Secondary research
Understand abandonment paaterns in industry standard search engines
Hypothesis generation
Learn self-reported patterns for search abandonment and generate hypotheses for specific problem area (matching/retrieval etc.)
True intent matching
Generate an example dataset of queries labeled with intents (informational, navigational etc.)
Key Learnings
Rate of gain = Value of info/ Cost of obtaining info
Through contextual inquiry, we quickly learned that the core issue was not search retrieval or relevance matching. Search results were often technically correct yet abandoned because titles and descriptions relied on vague language or tribal knowledge, making them appear unhelpful or misleading. As a result, poor or inconsistent metadata drove abandonment even when the underlying content was relevant.
Informational queries have higher abandonment than navigational queries
Mapping search queries to user intent revealed a clear pattern in performance. The search engine worked well for navigational queries, where employees knew the exact platform they wanted to reach. However, abandonment was significantly higher for learning-oriented searches, particularly when users could not recall the name of the destination platform. Topic modeling further showed that many searches were action-driven (e.g., changing an address or marital status), yet inconsistent metadata and incomplete indexing of key task pages prevented users from finding the right entry points, leading to frequent abandonment.
Impact
This research shifted the team’s focus from search relevance to foundational metadata and indexing gaps, preventing costly reinvestment in a new search engine. By aligning improvements to user intent and high-abandonment query types, the team was able to prioritize changes that directly reduced friction in everyday employee tasks.
(01)
Abandonment reduced significantly
Given the many foundational issues, even low-hanging fixes, such as resolving broken links, indexing additional pages, and ensuring search results were consistent across secure and non-secure networks, led to meaningful improvements in retrieval. These changes alone resulted in a 20% decrease in query abandonment.
(02)
New research methods intoduced
This research introduced new approaches for evaluating retrieval systems and informed a six-month, insight-driven roadmap for the product team.