In earlier fieldwork, Castilla studied a large service company that used a two-stage system, in which one manager rated an employee's performance and a second manager decided pay based on that rating. Women there earned smaller raises than men who held the same job title, reported to the
same manager, and received the same performance scores. The same shortfall applied to ethnic minorities and to employees born outside the country. The disparity was largest in bonuses. Coworkers can see promotions, hires, and firings, so unfair treatment there is easy to spot. Bonus amounts stay private. Castilla argues that managers feel less accountable for a decision no one else sees, which is why the bias appears mostly in bonuses.
Managers also disagree about
what merit is. A second study by Castilla and colleague Aruna Ranganathan found that managers inside the same company define merit differently, often by the qualities that led to their own promotions. When managers measure merit by different standards, a company can't reward it consistently, whatever it claims about its process.
Together, these two problems can make a merit-based system produce the unfairness it was meant to prevent. A company can follow all of its
own rules and still pay employees according to their gender or race as much as their performance, because the belief that the system is fair stops anyone from checking whether it is.
Common fixes such as diversity workshops, blind résumé screening, and rewritten job ads have limited effect, he writes, because companies apply them before identifying their specific problem. Castilla proposes instead that a company audit the decisions it already makes, tracking who
applies for each job, who advances, who's promoted, and what each person is paid. The company should then check whether gender or race still predicts those outcomes after accounting for the skills and experience the job requires. If gender or race does, the cause is bias, and it can be traced to the specific decision behind it. | 
