A large enterprise organization of approximately 20,000 employees engaged Lynx Analytics to better understand the employee experience across every stage of the lifecycle — from onboarding through exit. To gather meaningful insights, the organization conducted multiple employee surveys, including a biannual engagement survey, an onboarding survey for new hires, and an exit survey for departing employees. In addition to these internal feedback mechanisms, the client sought to incorporate external sentiment by analyzing Glassdoor reviews alongside survey responses. Each of these data sources generated high volumes of open-text feedback alongside quantitative survey scores. While quantitative ratings were relatively easy to analyze and interpret, the verbatim comments contained deeper context and nuance—and were significantly harder to review at scale without a structured approach to summarizing responses and grouping them into consistent themes.
Before working with Lynx Analytics, the client relied primarily on a manual process to make sense of survey feedback. In one engagement survey cycle alone, the organization received approximately 14,000 responses — far more than a team could reasonably review. As a result, the HR function typically selected a smaller sample, sometimes around 1,000 comments, and asked human reviewers to manually tag each response using an existing theme taxonomy.
This approach was time-consuming and limited in coverage, but the larger issue was that it could not keep pace with evolving employee concerns. Because the taxonomy was predefined, it didn’t naturally adapt when new themes emerged due to shifting business priorities or major external events. The process was also prone to inconsistency. Different reviewers could interpret the same comment in different ways, and as the volume of coding increased, fatigue and human error could impact accuracy. Ultimately, the client needed a way to extract insights from the full set of employee comments while improving consistency and making the taxonomy more dynamic over time.
Lynx Analytics designed and implemented an AI-driven solution to automatically categorize, summarize, and analyze employee survey verbatims using Large Language Models. The objective was to enable the client to process every response (not just a sample) while organizing results into themes that HR teams could easily explore and communicate. A key element of the solution was the creation of a structured, scalable theme taxonomy built from the data itself. Lynx Analytics used a bottom-up approach in which employee comments were first processed through an LLM in order to identify meaningful themes emerging directly from the responses. Once validated, the taxonomy could then be applied top-down to new survey cycles, with the flexibility to revisit and refresh the structure whenever new themes appeared.
During development, Lynx Analytics also refined the taxonomy hierarchy to make it more usable for the client. Rather than relying only on macro and micro themes, Lynx Analytics introduced an additional middle layer — “meso themes” — to provide the right level of granularity for interpretation and decision-making. In the final structure, users interacted with a set of macro themes and meso themes, while more detailed micro themes were retained behind the scenes to support the taxonomy logic.
To make the solution practical for day-to-day HR usage, Lynx Analytics delivered two main outputs. The first was a Databricks-based dashboard that allowed HR leaders to explore and compare themes across time, including year-over-year deltas. The second was a “chat-with-your-data” chatbot that enabled HR professionals to ask questions such as what the key themes were in a given engagement cycle, and to quickly retrieve supporting examples of negative or positive comments under specific topics.
The overall delivery required close collaboration with the client to align the taxonomy with their internal expectations and reporting approach. While early model outputs were generated quickly, the project went through approximately 15 to 16 iterations to incorporate client feedback and ensure that the theme structure matched how the organization wanted to interpret employee sentiment. The solution was developed over approximately five months by a team consisting of three data scientists and one data engineer, and access was intentionally limited to a small HR audience due to the sensitive nature of the data.
With Lynx Analytics’ solution in place, the client was able to shift from manually coding a small sample of survey comments to analyzing the full set of employee feedback at scale. This automation delivered significant time savings by reducing the need for human tagging, while also improving consistency by minimizing the subjective bias that can affect manual categorization. Most importantly, the solution enabled HR stakeholders to capture a more complete and accurate picture of employee sentiment across multiple feedback channels, including engagement surveys, onboarding surveys, exit surveys, and external Glassdoor reviews. By applying a consistent taxonomy across these survey touchpoints, the organization was able to connect feedback across the full employee lifecycle rather than viewing each survey in isolation. This made it possible to identify issues early during onboarding, assess whether they persisted over time through engagement feedback, and determine whether they ultimately surfaced again in exit responses. Combined with the dashboard and chatbot experience, HR teams could explore patterns, drill into themes, and support reporting during the post-survey analysis period when teams meet to review results and define actions. The result was deeper insight into employee sentiment, delivered faster and at scale, with a solution designed to evolve over time.