Chapter 1
The Power of Data and AI in DEI and Well-being

Have you heard of the glass ceiling? How about the bamboo ceiling? These are references to barriers for women and Asians to be promoted into senior leadership roles in organizations, respectively. Have you ever wondered if such a "ceiling" existed in your organization? How would you measure that? Let's start by going through the measurement of diversity in an organization.

Diversity Measurement

Measuring diversity doesn't have to be complicated. As the former head of People Analytics, I've created the measurement strategy for diversity and DEI (diversity, equity, and inclusion) reporting at Fortune 500 companies. At times, I started from scratch with no previous measures, and at others I revamped the outdated measurements. Getting started doesn't have to be daunting.

Representation

How do you get started in measuring diversity? Representation is a great place to start, since it provides a snapshot of the organization's current state. This enables you to quickly answer questions such as, "How many female leaders do we have?" and "What percentage of our engineering organization is Black?"

If you are starting to measure representation, there are two considerations to keep in mind: dimension and segment.

The first consideration is around the dimension of diversity your organization wants to measure. Gender, race, ethnicity, age, LGBTQ+ status, veteran status, and disability status are some of the most common ones. Your leadership team and the board of directors may also focus on different dimensions.

The second question considers for which segment these metrics are important. Common segments may include levels, functions, business units, and locations.

Job levels can include categories such as vice president, director, manager, and so on. Functions may be as simple as tech versus non-tech or can include granular descriptions such as sales, marketing, engineering, finance, IT, and HR. Locations can take on regions, countries, and cities.

The combination of dimensions and segments will allow you to answer detailed questions such as, "What percentage of managers, senior managers, directors, vice presidents, and executive leadership team are women?" to help inform whether there is a glass ceiling in your organization.

A common question I answered while working in the tech industry is, "What percentage of tech and non-tech positions consist of women?" This may not be as relevant if your organization isn't in the tech industry, but there can be other job function segments that are relevant. If you look at the diversity reports from large global tech companies, such as Google, these are common dimensions and segmentations.1

While gender is a dimension that is globally consistent, race and ethnicity may require a more regional or country specific approach. There may be locations where race and ethnicity are more homogeneous or where ethnicity takes on a different meaning. This is why many organizations I have worked with choose to limit the race and ethnicity measurement to their US locations.

As I review the diversity data for various other global organizations, the decision may have to do with the context of race and ethnicity rather than not having diverse employee population to warrant such measurement globally. Some of the largest global organizations, Microsoft and Walmart, for example, disclose race and ethnicity for US-based employees. Google, on the other hand, publishes race and ethnicity data for each region, including Americas, Asia-Pacific, and EMEA (Europe, Middle East, and Africa).

In summary, to get started with diversity representation you need the minimum of gender, race, ethnicity, age or generation, and veteran status to paint a picture of the workforce. The additional data on LGBTQ+ status and disability status are also helpful if available in the organization.

How often should you refresh the diversity data? I recommend monitoring and updating the metrics of interest quarterly. For large organizations or those going through significant change, monthly would be a more appropriate frequency. Of course, the most ideal scenario is to refresh daily so metrics are available in near real time. However, that may not be realistic for analytics teams that are just getting started nor should they be considered a deal breaker if daily updates aren't available. In global organizations where such data are not stored in the HR system but instead have to be obtained via anonymous surveys due to legal reasons, an annual refresh may be the most realistic option given the time commitment.

Now that you know how to measure representation, what are some ways to improve it?

There are three key levers that can move the needle on diversity representation: hiring, promotion, and retention.

Hiring Diversity

Let's dive into the first lever, hiring. How do you measure the diversity in hiring? Hiring diversity is the diversity of your organization's new hires over a period. For gender diversity in hiring, for instance, you'll want to know what percentage of the new hires consist of women, men, and non-binary. If nothing else changes, hiring more women will improve the overall female representation number. If that seems too easy, it's because we're not done yet. You may receive questions from hiring managers or the HR leadership team. Why are we hiring so few women in engineering? It is useful to have measurement throughout the entire recruitment funnel.

Before diving into the data, you want to start with a focused and concrete problem to solve. For example, here are some questions I've seen organizations ask frequently.

• Are there any bottlenecks in the recruitment process that require immediate attention?
• Where do applicants who identify as woman drop off in the process?
• Which recruiting source brings in the highest acceptance rates among women?

A recruitment funnel, or hiring funnel, outlines each stage of the recruitment process, from sourcing candidates to onboarding. As you overlay the diversity dimension and source information onto the recruitment funnel, you'll be able to answer the second and third questions we just discussed. Here's why these questions are helpful: Imagine if you found out from the gender recruitment funnel analysis that women tend to drop off between qualified and interview stages. You could further examine if there are biases in an interview selection process. Alternatively, you might discover that women were less likely to make it to the final hired stage. This data might prompt follow-up questions such as, "Are women less likely to accept a job offer, or are they less likely to be offered?"

Identifying bias in a specific stage can be challenging, given the number of factors involved. You can review interview feedback or interview assessment scores by gender if such data are available. However, it is often challenging to get a clear reason why an offer is rejected because many candidates don't want to burn the bridge. There may also be multiple reasons, making it difficult to pinpoint a single factor.

Let's imagine your data shows that female applicants who apply via LinkedIn are the most likely to become a successful hire, whereas referrals are the least effective. Is it time to reconsider the sourcing strategy?

One underrated data source in the recruitment process is the Candidate Experience Survey data. Think back to your own experience as a candidate. Would you accept a job offer when you were treated poorly?

This data can shed light on whether applicants from different backgrounds experience the recruiting process differently. When you compare the survey results by dimension of diversity such as gender, you can also look at the results within each stage of the recruitment process.

You can put the candidate feedback data by stage on a single chart and compare across demographic groups, as shown in Figure 1.1, for instance2:

Figure 1.1 Candidate Feedback Data by Gender Across Recruitment Stages

What do you observe in this chart? While women report higher satisfaction levels during the screening stage, their satisfaction drops during the subsequent stages of the hiring process. In addition to the qualitative scores, it's also important to analyze the survey comments to fully understand the candidate experience for various groups. We will discuss more about how to analyze comments later in this chapter.

Attrition Diversity

Another lever an organization can use to improve representation is retention, or attrition prevention. Should you care about the diversity of the employees who have left your organization? Sometimes described as the "leaky bucket" issue in DEI, attrition of diverse talent is one of the reasons many organizations haven't made significant progress.

If the number of women your organization has hired is similar to or lower than the number of women leaving the company, there will be no progress in the female representation. This is why it's critical to also measure attrition diversity.

How do you measure attrition diversity? Similar to the promotion diversity, you will measure the attrition rate for each group.

In the 2022 McKinsey and Lean In study, 10.5% of women leaders quit their jobs over the last year, which was the highest rate...