Leveraging large language models for automated performance appraisals: Opportunities and challenges

One major issue with traditional performance appraisals is inefficiency, bias and subjectivity. Oftentimes large language models (LLMs) like GPT-4 offer a promising approach to standardize performance evaluations which leverage structured and unstructured feedback for data-driven assessments. In this study, a data set with structured and unstructured data is taken and fed into GPT-4 to analyze self-evaluations and mid-year performance reviews to automate the appraisal process and compare it to human evaluations. Although GPT-4 is generally accurate and is similar to human assessment, the main challenge lies in the non-quantifiable factors such as workplace dynamics and lack of emotional intelligence. Although AI models have a much more accurate prediction rate than manual performance appraisals, there is always a need for a human-in-the-loop (HITL) approach to help AI perform better. This study focuses on how human-in-the-loop (HITL) can help AI-based performance appraisals by bringing in non-quantifiable factors such as workplace dynamics and conflict resolution within the employee data.