What are effective ways to remove bias from promotions and salary increases?

We have a bunch of new tools that help us reduce bias in the recruiting process. This includes changing the language of our job postings, blind reviews, and scoring models.

However, we’d like to take the same approach for our leveling - promotions and salary increases. At this stage it’s hard to not know a lot about the employee in question. Are there any new tools or techniques that can help us understand and eliminate bias for known employees?

2 Likes

Great question! What does your performance assessment process look like? Do you have opportunities for colleagues besides direct managers to provide performance feedback? Are you able to do any tracking of promotions/salary increases and various demographic data to dive into identifying any possible examples of bias? (ie: age correlating with promotions). You may have already seen this but I love buffer (they also use culture amp!) and how they approached this topic: https://open.buffer.com/bias-performance-reviews/

Hey Jon, to help address this we built out a set of competencies attached to each role within the company and roles that could exist. Each Manager is responsible for building the competencies for their teams and is strongly encouraged to get their teams input and feedback on those competencies and how they are written. This helps create a level of ownership from all levels of the organization and provides clear paths for advancement and promotion within each team. It is also a project that will continually evolve and expand as new positions open up and competencies are created and expanded upon. It is also open for anyone in the organization to add to and edit. We open sourced our current competencies list on Github, https://github.com/sendwithus/competencies for anyone to use. Hope that helps.

I think the precursor is to define/set a reference framework or model - otherwise there can be no consistency or comparison against desired qualities, characteristics, or skills. Otherwise any subsequent tool or process is flawed to start with (and the lack + internal publication of such modeling is one major contributor to bias and expectation mismatch).

Recently Circle CI released their engineering competency matrix -> https://circleci.com/blog/why-we-re-designed-our-engineering-career-paths-at-circleci/ and there’s always the Radford approach https://radford.aon.com/insights/articles/2015/radford-global-job-leveling but essentially some foundational codifying has to be done IMHO.

On salary increases - Providing past increase and/or incentive outcomes will influence managers’ current decision making so any past biases will be carried on. So removing historical data from salary review & incentive review files will help.

Harvard researcher, Paola Cecchi-Dimeglio found that performance appraisals were more fair when raters:

  • Crowdsourced feedback from multiple sources (peers, customers, etc.) rather than relying on direct supervisors
  • Give raters specific criteria to rate people on rather than just asking them to assess overall performance (e.g., team impact, customer impact. company impact, etc.)
  • Read more here

Three researchers from Stanford’s VMware Women’s Leadership Lab boiled bias down to a simple equation: Open Boxes = Open for Bias.

What they meant by this is that, when left to their own devices, leaders will interpret others’ performance based on arbitrary criteria unless they are given some specificity. Without that specificity, they use their own biases.

To counteract this tendency, they recommend the following:

  • Create better prompts for managers. Give them a nudge in the right direction
  • Ask reviewers to identify measurable outcomes
  • Read more here

The Dimeglio results are consistent with what I’ve seen.

I’d be on alert for qualitative data and open responses ESPECIALLY racially/gender coded language.

Some examples that raise red flags:
“Executive presence”
“Tone”
Language of fear “intimidating” “not fitting in”

Anything dealing with feeling/emotion or culturally “fitting in” is often a trigger for identity as opposed to performance or behavior.

Failing to account for this provides significant turnover risk, legal risk and can quickly destroy a company’s culture, as it allows bias to determine promotions/comp vs contribution.

In addition: women, people of color and marginalized employees are often forced to do emotional labor with employees that’s essential for company performance/retention but ignored (perhaps deliberately) in assessments of cultural contribution or productivity. Such work MUST be factored into evaluations of performance, as it depresses performance sometimes and often times elevates/sustains the performance of other employees.

Example: mentoring/coaching, being asked to do marketing work for brochures, asked (or voluntold) to participate on ERGs, D&I initiatives, etc.

Failure to account for these factors predicts massive turnover for diverse employees who get tired of doing free labor with little or no respect or such work.

Culture Amp just published an article on their blog that outline 10 of the most common biases and how to mitigate them: