Innovation is the introduction of something new … after you were born.
Nothing about us, without us.
Any book highlighting innovations runs the risk of quickly becoming outdated. In fact, the moment this book hits bookshelves the clock is already ticking that there will be a new concept, tool, technique, technology, model, methodology, program, platform, or approach that redefines the game for data‐driven DEI. While I'm reasonably confident that the five‐step cycle of Data‐Driven DEI is sufficiently durable to stand a test of time, I humbly accept that certain innovations I have profiled will inevitably become obsolete (and therein lies the opportunity to write future editions of this book!).
Nonetheless, I conclude this edition by offering my thoughts on the future of Data‐Driven DEI and, more specifically, a few thematic areas that will likely transform the landscape of Data‐Driven DEI. While I highlight specific innovations, I also emphasize guiding principles to help you navigate whatever future lies ahead.
At various points throughout this book, I have referenced Equitable Analytics™ and the Equitable Impact Platform™ (EquIP™), which is powered by geospatial Equitable Analytics™. Pioneered by Peter York, BCT's principal and chief data scientist, Equitable Analytics™ points to the future of Data‐Driven DEI both methodologically and ideologically.
As a methodology, Equitable Analytics™ is an equitable solution for figuring out what works for whom or, stated differently, what “causes” a result or outcome for whom. Equitable Analytics™ is a part of what Judea Pearl, in his book The Book of Why: The New Science of Cause and Effect, refers to as the “the Causal Revolution,” because the underlying concept of “what works” is cause and effect. Referring to our discussions in Step 3: DEI Insights about “What Works” models and in Step 5: DEI Impact about correlation vs. causation, “what works” does not equate to what is associated or correlated with a desired result or outcome, but rather, what causes a desired result or outcome. “Nowadays, thanks to carefully crafted causal models, contemporary scientists can address problems that would have once been considered unsolvable or even beyond the bale of scientific inquiry… . The mere mention of ‘cause’ or ‘effect’ would create a storm of objections in any reputable statistical journal,” writes Pearl. Even two decades ago, asking a statistician questions like, “Does DEI training mitigate unconscious bias?” or “Do ERGs increase feelings of inclusion and belonging?” would have been like asking if she believed in voodoo, according to Pearl. He continues, “But today [we can] pose such questions routinely and answer them with mathematical precision. To me, this change is nothing short of a revolution. I dare to call it the Causal Revolution, a scientific shakeup that embraces rather than denies our innate cognitive gift of understanding cause and effect.”1
Equitable Analytics™ uses BCT's Precision Modeling to build causal models that understand cause and effect for specific populations and accomplish this at a level of granularity that is more nuanced and more insightful than simply disaggregating data by demographics. While disaggregating data by demographics is a recommended and useful practice, we also know that our ability to understand what works for whom is insufficient when examined solely through a demographic lens. Equitable Analytics™ can provide deeper insights and will increasingly supplant disaggregating data by demographics alone (see Figure 6.1). According to Peter during a one‐on‐one conversation, “With the tools we now have available, we can make tailored and precise recommendations that predict a desired DEI outcome in the same way Netflix and Amazon can make a tailored and precise recommendation that predicts enjoyment from watching a specific movie or reading a specific book. This is revolutionizing DEI, social and community programs.” Understanding the Equitable Analytics™ process helps you to fully understand its power.
The first step in the Equitable Analytics™ process is to determine the focus of analysis or the unit of the analysis. This breaks down to either cases, which includes people and groups of people, or places, which includes various geographic locations. More specifically, there are three choices for the unit of analysis:
The focus of analysis determines the entity you desire to say something about. When Equitable Analytics™ is applied to cases, it helps determine and recommend the optimal mix of initiatives and strategies for individuals to improve personal DEI and for groups to improve organizational DEI. When Equitable Analytics™ is applied to places via EquIP™, it helps determine whether a community is receiving the locally accessible government support, philanthropic support, public contributions, social services programming, and volunteerism it needs to improve diversity, inclusiveness, and well‐being, equitably, along with recommendations of the optimal mix.
The second step in the Equitable Analytics™ process is to build a causal logic model for the desired outcome with stakeholder engagement. A logic model is a visual diagram depicting the causal assumptions. It uses directional arrows to represent the relationship between contextual factors and resources that best support the workplace experiences that lead to the greatest level of equitable outcomes for all employees. Think of your logic model as a visual diagram of your organization's hypotheses as to what it takes for every employee to succeed. This framework will guide the next steps of Precision Modeling.
For example, let's assume that a desired outcome for your organization is that every employee have an equitable opportunity to advance and be promoted to the management level. However, your organization has experienced challenges with equitably advancing and promoting Black women to the management level. To begin the Equitable Analytics™ process, a team comprised of DEI experts, HR professionals, and diverse colleagues who have had positive and negative experiences reaching the management level, would be engaged in building a causal logic model for advancement and promotion to the management level within your organization. The logic model could include variables such as the number and type of learning and development courses, number and location of departmental rotations, existence of mentorship and sponsorship relationships, and number and type of leadership development programs (see Figure 6.2).
When it comes to diversity, equity, and inclusion, it is critically important not to assume that an employee's race, gender, sexual orientation, or other identity characteristics should causally affect their workforce experience. That would be antithetical to valuing DEI. So, your causal logic model should not include workplace context variables reflecting demographic characteristics such as race/ethnicity, gender, age, sexual orientation, disability status, and so forth, with arrows connecting them to one's workplace experience. For example, being a Black woman should not “cause” a different workplace experience, and therefore, demographic and identity variables (e.g., race/ethnicity, gender, age, disability status, etc.) should not be represented in your causal logic model with arrows that affect the workplace experience. In doing so, and recognizing that, unfortunately, it is often the case that demographic and identity variables are sometimes the reason people are selected (and not selected) for certain workplace experiences, such as mentorship and sponsorship, Equitable Analytics™ mitigates this selection bias by excluding these variables from the causal logic modeling (and subsequent Precision Modeling) process.
Instead, your causal logic model should incorporate identity into the outcome of promotion and advancement. Specifically, demographic and identity variables should be represented in your causal logic model as a part of the end result or outcome. For example, demographic and identity variables should be in your causal logic model as the effect of a positive workplace experience at the end of an arrow where your outcome statement states something like “Advancement and promotion of more Black women to management level.”
The third step in the Equitable Analytics™ process is to build an analytical framework that will serve as the instructions for conducting the data modeling. The key step in developing the analytical framework is to determine which variables in your administrative data (i.e., existing data in an HRIS, LMS, ERP, CRM, or other DEI‐related data system) align with each of your workplace logic model components (i.e., context factors, experiences, and outcomes). For example, your workplace logic model will likely include contextual factors like employee educational background, position, and level. Your analytical framework will need to identify the specific variables in your administrative data that align with these logic model elements.
The fourth step in the Equitable Analytics™ process is to conduct Precision Modeling, which includes the following two tasks:
Continuing with the previous example, Equitable Analytics™ could find that there are nine matched comparison groups (refer to Figure 6.3) based on the workplace contextual factors that will affect their likelihood to experience what it takes to advance. Let's assume that matched comparison Group #2 is “project managers in the operations department with 3+ years of experience and a bachelor's degree.” As you can see in Figure 6.3, the current success rate with advancing and promoting members of Group #2 to the management level is 55%, whereas the overall success rate across all groups is 61%.
Continuing the previous example, Equitable Analytics™ would determine the optimal mix of strategies—including learning and development courses, departmental rotations, mentorship/sponsorship relationships, leadership development programs, and the like—that maximize the likelihood of everyone in Group #2 advancing and being promoted to the management level. The analysis could reveal that an emotional intelligence course, a departmental rotation into marketing and sales, a formally assigned mentor/sponsor, and an inclusive leadership development program represent the most effective mix of strategies (i.e., what causes or “what works”) for members of Group #2 to advance and be promoted to the management level. In fact, when people in this group receive what works, their success rate is 70%, and when people in this group do not receive what works, their success rate is only 5%, as shown in Figure 6.4.
Once these determinations have been made of what works for whom, the fifth step in the Equitable Analytics™ process is to perform an equity assessment by putting demographic characteristics such as race/ethnicity, gender, age, disability status, and so forth, back into the algorithmic model to determine whether what works for a matched comparison group has been equitably administered, as shown in Figure 6.5. Continuing with the previous example, Equitable Analytics™ could look at all project managers in the operations department with 3+ years of experience and a bachelor's degree to determine if Black women have been equally as likely as white men to receive what works. As you can see in Figure 6.6, 65% of white men got what works while only 45% of Black women got what works within the same Group #2. Additionally, Equitable Analytics™ could identify which mentors/sponsors have been the most effective (and ineffective) in helping Black women to be promoted and advance to the management level. This offers very nuanced and powerful insights regarding exactly by whom and where inclusivity and allyship are enabling people to advance and exactly by whom and where bias or discrimination may be impeding advancement, which could lead to very targeted actions such as unconscious bias and conscious inclusion training.
Mitigating Algorithmic Bias In addition to mitigating selection bias, Equitable Analytics™ also mitigates algorithmic bias. All too often, data scientists uphold predictive accuracy as the measuring stick for an effective algorithm, but it is not an equitable criterion. For example, an algorithm designed to predict who is the most ideal candidate to become the next CEO of a corporation, nonprofit organization, or private foundation will increase its accuracy by including race/ethnicity as a variable in the model. However, because the majority of CEOs are white men,2 doing so would translate a race/ethnicity bias into an algorithmic bias that advantages certain groups and disadvantages other groups. Hypothetically speaking, based on historical data, the algorithm could predict with 99% accuracy that whites are the most probable selection to become CEOs. However, this would be due to the fact that prior successful CEOs were much more likely to be white as a result of historical racial biases. Once again, race/ethnicity should not be a factor in making this determination. If this hypothetical algorithm were then used to select candidates deemed to be most likely to become CEOs for a leadership development program, it would advantage whites and disadvantage people of color, much like the prior example of how Amazon's computer models translated a gender bias into an algorithmic bias that advantaged men and disadvantaged women seeking jobs. An algorithmic model that removes race/ethnicity would be less accurate; however, it would better reflect the unbiased truth, or representation of what is fair and equitable, that race/ethnicity should not predict who becomes a CEO. By removing demographic variables as predictors of experiencing what it takes to succeed, the Equitable Analytics™ methodology mitigates the algorithmic biases that can result when demographic variables are included in the algorithmic modeling process, thus resulting in algorithms that prioritize the unbiased truth over predictive accuracy. Cathy O'Neil, author of Weapons of Math Destruction: How Big Data Increases Inequality and Threatens Democracy, advocates for such an approach when she writes, “But wait, many would say. Are we going to sacrifice the accuracy of the model for fairness? Do we have to dumb down our algorithms? In some cases, yes.”3
Finally, the results from Equitable Analytics™ produce summaries of both the current success rate that is attributable to the strategies and interventions each person currently receives and the best possible success rate if each person received the strategies and interventions that work for their group. Put another way, the results present “counterfactual” findings of what happened when a group was and was not appropriately supported to experience what it takes to advance. Once you and your organization determine which strategies and interventions are feasible from among those recommended for each group, such as an emotional intelligence course and a formally assigned mentor/sponsor only for Group #2, Equitable Analytics™ offers a prediction of the success rate you can now anticipate for that group, as shown in Figure 6.6. The solid bar depicts the attributable success rate; the line marked with “X” depicts the best possible success rate if everyone received what works for their group; the gap between the two lines is a performance gap, or a gap between our current performance and our best possible performance given the strategies and interventions currently being offered and tracked. As you can see, if everyone in Group #2 received what works for their group, their success rate would increase from 55% to 91%. Equitable Analytics™ can automatically generate these and other actionable insights in reports and dashboards.
Methodologically, Equitable Analytics™ can be applied to any number of desired outcomes including identifying the mix of strategies to maximize inclusion and belonging, increase intercultural competence, and achieve equity for every employee to advance and be promoted to the executive level, and much more. Ideologically, as you can clearly see from the preceding description, Equitable Analytics™ also embraces and espouses several tenets that should guide the responsible use of data analytics in DEI including machine learning, NLP, and other branches of AI. They include:4
Based on these principles and its seven‐step process, Equitable Analytics™ offers both an ideological and methodological vision for a more precise future of Data‐Driven DEI.
The term neural networks refers to a branch of machine learning that loosely mimics the biological neurons and learning processes of the human brain. Neural networks are comprised of a series of algorithms that learn by endeavoring to recognize patterns in data. Artificial neurons represent levels of abstraction or layers in the neural network that perform basic to increasingly more complex pattern recognition. For example, a neural network to identify the numbers 0 through 9 might begin with a layer that identifies pixels as black or white, and then pass this information on to a subsequent layer that identifies different lines and curves, and then pass this information on to a final layer that identifies each number by its unique combination of lines and curves.
Deep learning refers to a process by which a neural network seeks to identify patterns or solve a problem through a never‐ending cycle of trial and error called “training” that strengthens its understanding of certain underlying relationships in data and weakens others leading to more complex understandings. The deeper the layers of the neural network, the deeper the learning. Deep learning has deep implications for the future of Data‐Driven DEI. Before I share these implications, I will provide brief background on a specific and growing genre of deep learning programs, large language models (LLMs), that have the deepest implications.
Large language models (LLMs) are trained on massive data sets using natural language processing (NLP) and natural language understanding (NLU) to strive toward artificial general intelligence, or the ability of machines to learn or understand general tasks performed by human beings (which is yet to be achieved). NLP is a branch of AI that deals with the structure of human language. NLU analyzes the syntactic and semantic elements of text to derive meaning. Combining the use NLP and NLU can help identify patterns and meaning in unstructured data (i.e., raw text and narratives), such as the internet and its vast library of content, including articles, blogs, and social media posts, that would be difficult for a human to find. Unstructured data often contains powerful patterns within nuanced context, but these patterns go unnoticed because it is difficult for humans to process and make meaning from large amounts of text data. This is where LLMs, NLP, and NLU can offer mind‐blowing assistance.
Arguably, the most recognized LLM program is the Generative Pre‐Trained Transformer (GPT)5, and its associated chatbot, ChatGPT, by OpenAI, an AI research laboratory consisting of a for‐profit and parent nonprofit, whose mission is to ensure that artificial general intelligence benefits all of humanity. GPT is currently the largest LLM program, trained on 45 TB of text, but there is a growing list of others including GPT‐J, GPT‐NeoX, DeepMind, Google's BERT (Bidirectional Encoder Representations from Transformers), Facebook's BART (Bidirectional Auto‐Regressive Transformers), and DistilBERT, to name a few.
These programs have trained on content as vast as the internet (filtered), collections of digitized books, the entirety of Wikipedia, and more, to solve a single problem: predicting the next word in a text (you've experienced something comparable if your phone, computer, or search engine has offered suggestions to help complete a sentence while you are typing). The results are impressive to say the least.
LLM programs are able to produce original, coherent, often thoughtful, and well‐written paragraphs of prose in response to simple prompts. For example, here is what GPT produced in response to my prompt, “Write an essay that compares and contrasts diversity, equity and inclusion with racial equity”:
There are many ways to think about the relationship between diversity, equity and inclusion (DEI) and racial equity. Both DEI and racial equity are about ensuring that everyone has an opportunity to participate in and benefit from society. However, there are some important distinctions between the two.
DEI is about creating a society in which everyone can participate and feel included, regardless of their background. This includes things like making sure that people of all races, ethnicities, genders, sexual orientations, abilities and socioeconomic backgrounds have an equal opportunity to succeed.
Racial equity, on the other hand, is specifically about ensuring that people of all races have an equal opportunity to succeed. This includes things like addressing historical and current inequities that have put people of color at a disadvantage, as well as making sure that people of color have an equal voice in decisions that affect them.
Both DEI and racial equity are important for creating a just and fair society. However, racial equity is specifically focused on addressing the disparities that exist for people of color, while DEI is about creating an inclusive society for everyone.
I thought this was pretty good! Not only can GPT and other LLM programs create an essay in a fraction of a second, but they can produce an unlimited number of original essays in response to the same prompt (they can also produce poems, stories, computer code, and more), and the prompt can include more nuanced instructions such as a particular style or voice (i.e., “Write an essay … like a hip‐hop song”).
LLMs will revolutionize DEI in several ways such as automatically, quickly, and accurately:
With all of their power, potential, and promise, LLM programs, like all computer programs, must be governed and used in a proper ethical context that benefits humanity. This raises the question: What constitutes a benefit to humanity? In the New York Times Magazine article “A.I. is Mastering Language. Should We Trust What It Says?,” Steven Johnson asks related and important questions such as, “How do we train them to be good citizens? How do we make them ‘benefit humanity as a whole’ when humanity itself can't agree on basic facts, much less core ethics and civic values?”6 In the article, Tulsee Doshi of Google says that one of its principles is “making sure we're bringing in diversity of perspectives—so it's not just computer scientists sitting down and saying, ‘This is our set of values.’ How do we bring in sociology expertise? How do we bring in human rights and civil rights expertise? How do we bring in different cultural expertise, not just a Western perspective? And what we're trying to think through is how do we bring in expertise from outside the company. What would it look like to bring in community involvement? What would it look like to bring in other types of advisers?” While these are the right questions, the answers are yet to be found and must be vigorously sought after to safeguard the future of Data‐Driven DEI.
Under Peter York's leadership, we have an ever‐expanding range of experiences at BCT with deep learning including fine‐tuning LLM programs and developing and training NLP and NLU models to accomplish the kinds of tasks described earlier. Perhaps more importantly, we fully understand the importance and obligation to include safeguards against the improper use of any machine learning techniques, including NLP and NLU. When analyzing text using LLM programs, it is important to be aware of potential biases that may be present in the data. We therefore espouse several guiding principles to monitor for bias when using these models:
While I explicitly cautioned against the use of neural networks and deep learning algorithms in the previous section due to their lack of transparency—it is difficult to extract exactly how they do what they do—it's clear they are likely here to stay. Their proliferation must adhere to these principles. The diligence and rigor that is applied to training deep learning algorithms must be surpassed by the diligence and rigor of ensuring that myriad voices are reflected in how models are built, what is used for training data, and how the training takes place. This will signify more than just deep learning, but rather deep, diverse, equitable, and inclusive learning as the future of Data‐Driven DEI.
Throughout this book, I've made repeated references to transparency as a guiding principle that engenders trust and sets the tone for an organizational DEI journey. Interestingly, a growing cadre of tools are no longer making transparency the exception but rather the norm by providing three types of insights:
As these tools become more ubiquitous and it is less obvious that personal data is being collected, analyzed, publicized, and used to make decisions, it will become increasingly important that explicit prior permission (opt‐in) is obtained from users for any data collection, so they can make informed decisions, and the data is collected in a safe and ethical way. For example, in Europe “consent must be freely given, specific, informed and unambiguous through a clear affirmative action, which means that pre‐checked boxes or other types of implied consent is not sufficient. The recipient must also be told exactly how their data will be used. Senders must keep evidence of the consent and provide proof if challenged.”8 As another example, Streamlytics, a next‐generation data ecosystem, provides “safe and ethical access to accurate consumer activity from all aspects of their lives.” Data is currency. It has value. People must be fully informed about how their data is being used—how their currency is being spent—as we move toward a more transparent future of Data‐Driven DEI.
At BCT, we envision and are working toward a future where a single platform can integrate all DEI functions such as assessment, planning, learning, development, reporting, evaluation, and more. The entire Data‐Driven DEI five‐step cycle will be captured:
In doing so, the DEI integrated platform will also centralize DEI data including assessment data, experience data, HR data, output data, and outcomes data, which often reside in disparate and siloed data systems such as an HRIS, ERP, CRM, and LMS. By integrating DEI functions and integrating DEI data, the platform will paint a hyper‐personalized picture of DEI for people and a comprehensive picture of DEI for organizations: What is the optimal learning and development pathway for an individual to increase diversity? What are the optimal KSAs for an individual to foster inclusivity? What are the optimal activities such as mentorship, sponsorship, and allyship for an individual to foster equitable outcomes?
This vision is rooted in the science of the individual and the belief that, according to Todd Rose in The End of Average: How We Succeed in a World That Values Sameness,9 “If we want a society where each of us has the same chance to live up to our full potential, then we must create professional, educational, and social institutions that are responsive to individuality.” Rose espouses three guiding principles of individuality:
As I've said before, organizations don't change. People change. These individual principles are therefore useful in guiding the future of Data‐Driven DEI for both people and organizations. They also expand upon BCT's vision for a DEI integrated platform that will support personal and organizational transformation via a library of DEI learning journeys, Microcommitments, VR immersions, and more, that are all adaptive, personalized, and tailored to the unique preferences and competences of people and the unique people, practices, and policies of an organization.
There is a growing number of DEI integrated platforms that begin to approximate this vision including Kanarys, Mathison, Co:Census, GlobeSmart, OpenSesame, Blueprint Strategy Platform, Included.ai, MESH Diversity, and Emprising™, but none of them fully encompass all pieces of this futuristic puzzle. BCT has begun to bring this vision to reality through our Equitable Analytics™, which helps determine the optimal mix of strategies and interventions for individuals and/or organizations, our Through My Eyes™ VR immersions, and our partnerships with Rali's Change Experience Platform (CxP), Intrinsic Inclusion™, and The Inclusion Habit®. Together, they provide an integrated suite of methods and features that can optimize personal and organizational impact by applying multi‐dimensional thinking, context‐specific understanding, and a unique pace and sequence for DEI pathways and journeys that represent the future of individualized Data‐Driven DEI.
Throughout this book, I have tried to show how data can be used in an efficient, effective, and ethical way along with the tools and metrics to improve your personal and organizational DEI journey. In addition to five specific steps and a plethora of tools and metrics, I have outlined several principles to guide the present and safeguard the future of Data‐Driven DEI. With this roadmap as both an anchor for grounding and a compass for guidance, I place my hope in what lies before us, and I place my faith in what lies within us.
As I stated in the introduction, our world is increasingly comprised of “communities of the like‐minded.” Far too often, we surround ourselves and associate with people who are like us—people who share the same values, beliefs, race/ethnicity, religion, socioeconomic status, political affiliation, and other identifiers, as ourselves. As a result, we are increasingly less likely to befriend people who are not like us, which only leads to greater division and misunderstanding. DEI represents a unique and unparalleled opportunity to break down the walls that can separate us in our personal lives, within our organizations, and throughout our society. While I firmly believe that Data‐Driven DEI is a means to these ends, I once again acknowledge that data is not a panacea to DEI; it is simply a tool. What ultimately makes the difference is you and the intentional steps you take to move beyond your comfort zone into your growth zone; be a bridge between people that would otherwise remain separated; and get comfortable with being uncomfortable. The true solution to a better tomorrow is what lies within you today.
I have often said to people that the ultimate objective is to make DEI a part of your DNA. In other words, your overarching aim is that the five‐step, never‐ending, continuous cycle of Data‐Driven DEI becomes a natural part of who you are and what you do. Inasmuch as DNA is the hereditary materials in humans, I now say to people that the ultimate objective is to also make our DNA a part of DEI. In other words, our overarching aim is that our humanity, empathy, and service to one another remain first and foremost and are fully embedded in any DEI tools and techniques, methodologies and technologies, platforms and programs, and deep learning algorithms and databases. That is, a future of DEI that is certainly driven by data, but is fundamentally centered on love, benefits all of humanity, and unites us as a people.