This article reports on research co-authored by Iris Bohnet, Albert Pratt Professor of Business and Government at Harvard Kennedy School, examining how gender and race influence performance ratings at a multinational financial services company. The central argument is that sharing employee self-appraisals with managers prior to managerial review creates an 'anchoring' effect, where manager scores closely mirror self-ratings. The research finds that women and workers of color consistently rate themselves lower, with women of color assigning themselves the lowest self-ratings of any group. Managers, in turn, rated people of color more harshly than white employees, while showing a partial 'gender boost' that reduced — but did not eliminate — disadvantages for women. A system glitch that prevented managers from viewing self-appraisals in one year showed reduced correlation between self- and manager ratings, though racial and gender dynamics persisted, partly because managers appeared to reference prior-year self-evaluations. The authors conclude that demographic patterns in ratings are sufficiently correlated with social identity to raise concerns, even absent definitive proof of bias, and the company subsequently ceased sharing self-evaluations with managers prior to review. Key insights: Women and workers of color systematically assign themselves lower self-ratings, with women of color rating themselves the lowest of any demographic group studied. Manager ratings exhibit a strong 'anchoring' effect when managers view employee self-appraisals beforehand — scores closely correlate with those self-ratings, embedding pre-existing self-evaluation gaps into official performance records. Even when the anchoring mechanism was disrupted by a system glitch, racial and gender disparities in manager ratings persisted, partly because managers reverted to prior-year self-evaluations as a reference point. Practical takeaways: Organizations that allow managers to view employee self-appraisals before submitting their own ratings may be inadvertently encoding demographic disparities into official performance outcomes. Data analysis of internal performance rating patterns — disaggregated by gender, race, and employee tenure — can serve as a diagnostic tool for identifying potential inequities in appraisal systems.