Closing the AI gender gap
IMD, Media Trust, and Code For Good Now are building a coalition to shape how AI is developed, deployed, and governed, so it becomes a force for inclusive growth, responsible innovation,...
Diversity, Equity, and Inclusion
You have the data. So why is inclusion lagging?
If your metrics don’t distinguish between the “D” and the “I” in DE&I, your efforts to become more inclusive will likely fail. Here’s how to use your existing data to get inclusion right.
Rapid read:
- Diversity data does not measure inclusion, and confusing the two can lead organizations to the wrong conclusions.
- Existing HR and operational data already contain powerful signals about inclusion at the team and manager level.
- Reframing three common metrics can help leaders identify where inclusion is working, where it isn’t, and where action is needed.
When Gallup’s seminal 2015 State of the American Manager report found that managers account for at least 70% of the variance in team engagement, the finding was treated as a leadership-development problem. Organizations responded with coaching programs, 360-degree feedback, and competency frameworks; all geared toward improving managers to close the variance from team-to-team. Almost none responded by confronting the more uncomfortable question that the data also posed: if the experience of work is overwhelmingly determined at the managerial level, why does virtually all inclusion measurement operate at the organizational level?
The answer reveals a category error that sits at the center of most corporate DE&I strategies: organizations have built their measurement infrastructure around diversity, not inclusion, and have proceeded as if progress in one guarantees progress in the other. It does not.Â
The common category error in corporate DE&I strategies
“Diversity” describes a characteristic of a workforce: how many people from how many backgrounds are present. “Inclusion” describes an experience: whether those people are resourced, heard, developed, and given a fair opportunity to advance. You can have one without the other, and the evidence increasingly shows that many organizations do.
McKinsey and Lean In’s Women in the Workplace research – now running for over a decade across more than 1,000 companies – clearly illustrates the gap. Women’s representation in the C-suite at top-performing companies has risen from 31% to 38% since 2021, but the experience of inclusion – measured through access to sponsorship, equitable promotion, and career support – has not kept pace. The McKinsey report also found that only 31% of entry-level women have a sponsor, compared with 45% of entry-level men. Four in 10 entry-level women have not received a promotion, stretch assignment, or leadership training opportunity in the past two years. Representation improved; for large parts of the workforce, the lived experience of inclusion did not.Â
These findings were echoed in the IMD white paper Why leadership systems fail women and how to fix them, which showed that women’s representation in executive roles had dropped below 31% entering 2026. These studies do not suggest a failure of commitment. The people who built today’s DE&I measurement systems intended to capture whether their organizations operate fairly. The problem is that the instrument they built – the representation report – measures one thing and is assumed to capture another.
Diversity data does not proxy inclusion data: they require different questions but, for the most part, only one set of questions has been operationalized.Â
Every operational system in your organization carries, alongside its intended signal, a secondary signal about whether the experience of working there is equitable.
The data is already there
Inclusion data is not missing from your organization. It’s present in systems that were built for other purposes – HR information system platforms, finance tools, internal mobility records, even meeting analytics – but it is simply not labeled as inclusion data. And because it is not labeled, it is not read.Â
Think of it as a “data shadow”. Every operational system in your organization carries, alongside its intended signal, a secondary signal about whether the experience of working there is equitable. Many important elements – such as budget allocation, career progression, internal mobility, the distribution of development resources and upward visibility – do not appear in DE&I reports, yet carry information about inclusion at the level where it happens (or fails to happen): the individual manager and the individual team.Â
Deloitte research into The six signature traits of inclusive leadership puts a number on the scale of this effect. Across a study of 4,100 employees in four organizations, it found that the behavior of leaders, whether senior executives or line managers, can drive up to 70 percentage points of difference between the proportion of employees who feel highly included and the proportion who do not. Among the top quartile of included employees, 72% reported high levels of inclusive manager behavior; among the bottom quartile, just 2% did. Inclusion is not an organizational atmosphere: it is a set of specific behaviors by specific managers, and those behaviors leave traces in operational data that most organizations already collect but have never thought to read as an inclusion signal.Â
None of this requires new platforms, new surveys, or new budget: it requires a new question directed at data you already
hold.
Three data signals worth reading differentlyÂ
The following metrics can be found in most organizations’ systems. Each has a conventional purpose; each also carries an inclusion signal that becomes visible the moment you change the question you are asking of the data.
Promotion velocity, disaggregated by manager
Promotion rate is a standard people metric. Most organizations report it by gender, ethnicity, or grade level: what proportion of a given demographic group reached a given seniority band, and how quickly. This aggregate view is the backbone of pipeline reporting.Â
The reframe is to calculate the same figure at the level of the individual manager. Promotion velocity measures the average time between promotions for employees within a specific manager’s team. When calculated at the manager level rather than the function level, the variance is almost always larger than the aggregate report suggests. A function that looks broadly equitable when averaged across all managers frequently contains individual managers whose direct reports advance at markedly different rates from peers in equivalent teams.Â
The McKinsey and Lean In data provide the macro picture: for 100 men promoted to manager, only 93 women were promoted in 2024. For women of colour, the figure was 74. But the aggregate does not tell you where in your organization this gap is concentrated. Manager-level disaggregation does. It answers a different question: not “do we have a promotion gap?” but “which managers are producing it, and which are not?” Those are materially different findings, and only the second points toward a specific intervention.Â
Internal transfer requests, read as a flight signal
Internal transfer data records who has requested to move teams, roles, or managers within the organization, whether those requests were approved, and what happened afterwards. Its conventional purpose is workforce planning: understanding internal talent flows, identifying teams that attract or lose people, and managing the pipeline.Â
The inclusion reading is different. An internal transfer request is a data point that says: this person is trying to stay in the company, but cannot stay where they are. The Gallup State of the American Manager report found that one in two employees has left a job to get away from their manager at some point in their career. The external departure is the visible event. The internal transfer request is the earlier, subtler signal: the same impulse to leave, directed inward first.Â
The diagnostic value is amplified when you disaggregate not by the role being sought but by the manager being left. A cluster of transfer requests from a single manager’s team, particularly when those requests are withdrawn, declined, or followed by eventual resignation, maps an inclusion problem that no annual engagement survey will surface in time. By the time an exclusionary environment registers in survey data, the people experiencing it have already tried to solve it themselves. Their transfer requests are the record of that attempt.Â
Discretionary resource allocation, disaggregated by team
Every organization allocates development resources unevenly. Training budgets, project exposure, conference attendance, mentoring access, stretch assignments: the distribution of these investments across teams is tracked for cost management, but seldom analyzed as an inclusion signal.
There’s a straightforward reframe here: take the discretionary development spend, including training, external development, and access to high-visibility projects, and divide by headcount at the team level. Rank the results. The distribution will almost certainly reveal something that your diversity report does not.Â
McKinsey’s 2025 data gives this signal direct empirical grounding. Only 21% of entry-level women are encouraged by their managers to use AI tools, compared with 33% of men at the same level. The difference is not a policy gap; both groups have the same access. It is a gap in who receives encouragement, investment, and opportunity from their direct manager. The newly launched IMD initiative and white paper Closing the AI Gender Gap, which finds that “women are grossly underrepresented in both AI development and use,” highlights the potential for AI to entrench existing fault lines, unless organizations, policymakers, and educators collaborate to ensure the systems shaping our future are inclusive.
The same pattern replicates across training spend, stretch assignments, and sponsorship. Deloitte’s inclusive leadership research identifies access to development and sponsorship as among the strongest predictors of whether an employee experiences their organization as inclusive. Discretionary resource data provides a proxy measure for that access without requiring any new diagnostic instrument.Â
If you do not have enterprise-wide access, the exercise is smaller in scope but identical in logic.
How can you get new insights from existing data?Â
None of this requires new platforms, new surveys, or new budget: it requires a new question directed at data you already hold.Â
For senior leaders
If you have enterprise-wide responsibility for people strategy, the starting point is a set of specific requests to your HR information system and finance teams: produce promotion velocity broken down by individual manager, not by function; produce internal transfer requests broken down by the manager being left, not the role being sought; produce discretionary development spend per head broken down by team. Most systems can generate these analyses within days. The barrier is not technical. Rather, it is that these insights have not been commissioned because the underlying question has not been asked: “Where is inclusion being produced or withheld, and by whom?”
For team leadersÂ
If you do not have enterprise-wide access, the exercise is smaller in scope but identical in logic. Look at your own team: who has been promoted in the past two years, and at what rate relative to their peers elsewhere? Who has requested to leave? Where has your discretionary budget actually gone? The answers describe the inclusion environment you have built – which may well not correspond to the one you believe you have built.Â
Key learningÂ
The three signals above serve to answer a key question: does your organization work inclusively, and if not, where exactly is it falling short? For too long, the data to answer it has been sitting in your systems, unread. These simple steps will help you ensure that inclusivity is truly embodied and not simply assumed by the presence of “diversity.”
Authors
Related
Beyond visibility: What the inclusion metrics miss
Leaders are tasked with measuring inclusion more rigorously than ever today – but are they measuring the wrong signals? ...
Six strategies to navigate charged DE&I discussions
Explore key techniques to keep DE&I dialogue constructive in hostile settings, including asking questions, reframing language, and knowing when to exit....
Beating the backlash: A cohesive vision for the future of D&I
In an era of political polarization, organizations can no longer afford to engage in 'performative' inclusion. Luca Condosta offers four essential ways to build strategic cohesion to avoid virtue signaling and drive...
Audio available
Learn Brain Circuits
Join us for daily exercises focusing on issues from team building to developing an actionable sustainability plan to personal development. Go on - they only take five minutes.
Â
Explore Leadership
What makes a great leader? Do you need charisma? How do you inspire your team? Our experts offer actionable insights through first-person narratives, behind-the-scenes interviews and The Help Desk.
Â
Join Membership
Log in here to join in the conversation with the I by IMD community. Your subscription grants you access to the quarterly magazine plus daily articles, videos, podcasts and learning exercises.
Â