When an organization experiences challenges related to diversity, it’s typically because subtle forms of bias are being transmitted every day in the basic business systems — impacting who gets hired and who rises through the ranks.

To make progress on your organizations’ diversity, equity and inclusion (DEI) goals, you need to de-bias the business systems by using metrics to diagnose where bias is creeping in, implementing evidence-based tweaks to existing systems and persevering in an iterative fashion until the metrics improve. Simultaneously, you need to train and empower managers at all levels to disrupt bias in the business systems they have control over.

Performance evaluations are a great place to start because they often serve as a gatekeeper for who receives promotions and raises. If you suspect that bias is impacting the performance evaluations at your organization, use metrics to find out. You can ask questions like:

• Do the performance evaluations show consistently higher ratings for white men than women and people of color?
• Do women’s ratings fall after they have children?
• Do the same performance ratings result in different promotion or compensation rates for different groups?

If you find disparities, the first step should be to arm evaluators with the knowledge to spot and correct for bias.

With performance evaluations, there are two common biases. The first is that the majority white men tend to be judged on their potential while “prove-it-again groups” (women, people of color, individuals with disabilities, members of the LGBTQIA+ community, older employees and first-generation professionals) are judged, or scrutinized, on their performance. Small biases can add up quickly. According to a recent study, women received significantly lower “potential” ratings despite higher job performance ratings, accounting for 30% to 50% of the gender promotion gap.

The second common pattern is that women and people of color are more likely to be evaluated on their personality (instead of their job-related performance). In a study of tech, 66% of women’s evaluations contained a negative personality criticism (i.e., “You come off as abrasive”) vs. only 1% of men’s evaluations.

Our research shows that training can go a long way. After conducting a performance evaluation audit of one company’s evaluations to identify bias, we helped them develop a one-hour workshop that equipped employees with knowledge on the common patterns of bias and then asked them to identify which patterns of bias were present in real excerpts from the previous cycle’s evaluations.

The results were dramatic. For example, the year before the workshop, 20% of white women’s evaluations contained comments that reflected that they didn’t want to make partner. The workshop explained how this reflected assumptions that women who are mothers are less committed to their work. The next year, only one women’s evaluation contained such a comment. In a separate lab study, we found that simply being read our two-page handout on “Identifying and Interrupting Bias in Performance Evaluations” led participants to give higher performance evaluations ratings and bonuses for white women and African Americans.

If there’s still distance to be covered, consider redesigning the performance evaluation form. Instead of open-ended evaluations, which are a petri dish for bias, provide evaluators with specific and well-defined performance criteria and require two to three pieces of evidence from the evaluation period to back up any numerical rating. In addition, have evaluators separately assess employees’ performance from their potential and their personal style from their skills that need to be developed. Keep experimenting with different “bias interrupters” until the disparities diminish.

Another high-potential business system to focus on is assignments. In industry after industry, our research shows that women and people of color report less access to career-enhancing assignments and more informal pressures to do the “office housework.” The right assignments can be pivotal to an employee’s career trajectory as they can provide opportunities to develop new skills and gain visibility with leadership. The process for interrupting bias in assignments is the same: metrics, evidence and persistence.

First, train managers to gather intel: Who is doing the office housework and how much of their time is it taking up? To do this, they can distribute our Office Housework Survey. With the data in hand, they can implement “bias interrupters,” such as setting up a rotation for recurring office housework tasks and requiring accountability. If administrative support is available, managers can make certain tasks — like sending the virtual meeting or cleaning up the conference room — part of their job duties. Periodically, managers should send the Office Housework Survey around to assess whether things have improved.

For equalizing access to the career-enhancing assignments, the first step is to identify what the “glamour work” opportunities are that lead to promotions. This can be done using our Assignment Typology Guide. Then train managers to keep track of who is getting those plum assignments — and to look for demographic patterns.

If managers notice that the pool is lop-sided, they need to zero in on what the reason is. If a diverse pool has the requisite skillset, they can simply set up a formal rotation to make sure everyone is being treated fairly. If the pool is not diverse, managers should examine how the pool was first assembled. Relying on self-promotion or volunteering will often put women and people of color at a disadvantage. If the pool needs to expand, managers should identify what the essential skills or competencies are for the assignment and develop a plan to help a wider range of employees develop those requisite skills. It may be as simple as having them shadow a more senior employee to see how it’s done.

Unfortunately, there is not a magic wand that will eliminate all the bias from your company. But that shouldn’t be the goal. It should be to take small steps consistently, and over time these small tweaks will amount to real and lasting inclusion.