The people who make a company’s products often know about quality problems long before customers or investors do. A new working paper turns that insight into a measurement, using millions of Glassdoor employee reviews to build an index that predicts product quality, recalls and profitability before they become visible from outside.
The study is by Wei Cai, Dennis Campbell, Yaxuan Chen, Yufei Chen and Andrea Prat, and was released as NBER Working Paper No. 35348 in June 2026.
Turning employee reviews into a quality score
Quality is central to a firm’s value but hard to observe in advance. The authors note that existing measures have limited coverage, are hard to compare across companies, and mostly capture failures only after they occur.
Employees, by contrast, watch internal operations firsthand. The researchers analysed more than 4.3 million employee reviews on Glassdoor, training machine learning models on a subset of firms that also had third-party customer satisfaction data. They then produced yearly quality indices for S&P 1500 companies from 2008 to 2023.
Better than the usual predictors
In out-of-sample tests, the indices strongly predicted future quality. They emerged as the single most important predictor when compared with firm fundamentals and with Glassdoor’s own star ratings.
The scores varied meaningfully between companies and within the same company over time. They also lined up with alternative quality measures and helped predict outcomes tied to quality, including product recalls, brand value and profitability.
The team also built versions of the measure using GPT models, both with zero-shot prompts and with supervised fine-tuning, to test how well large language models can capture firm-level quality.
The authors argue that employee voices offer a forward-looking and scalable signal for anyone trying to spot quality risks or opportunities before they become externally visible.
Study Details:
- Title: The Informed Insider: A Leading Measure of Quality
- Authors: Wei Cai, Dennis Campbell, Yaxuan Chen, Yufei Chen, Andrea Prat
- Journal: NBER Working Paper No. 35348
- Publication Date: June 2026
- DOI: 10.3386/w35348