What DEI metrics should we track, and should we still track them given the legal and political risk?

You are really asking two separate questions, and they carry different legal weight. One is whether to measure DEI outcomes at all. The other is whether to run programs that treat people differently based on a protected characteristic. Measuring and analyzing aggregate workforce data is generally lawful, and some of it is legally required: employers with 100 or more employees have to file the EEO-1 report, a government form that collects headcount by job category, sex, and race or ethnicity [5]. Federal anti-discrimination law does not forbid you from counting who works where. What it forbids is basing an individual employment decision, whether a hire, a promotion, a layoff, or a pay change, on that person's race, color, religion, sex, or national origin [1].

That distinction between aggregate analytics and individual decisions is the load-bearing line in this whole area. A dashboard showing that women hold 18 percent of engineering director roles is analysis. A rule that the next director must be a woman is an employment decision about an individual, and it is the kind of action that draws a Title VII claim, now from any group rather than only from historically underrepresented ones [1].

So the metrics half of your question and the should-we half are not the same debate. Many organizations are continuing to measure while re-examining or pausing specific programs, because the reporting obligations and the litigation risk point in different directions. None of this is legal advice; the specifics depend on your facts and on counsel who knows your jurisdiction.

Representation metrics count who works where, broken down by level, by function, and by change over time. The basic version is headcount by demographic group at each layer of the organization. A pattern the numbers often surface is a pipeline that narrows toward the top. The metric is honest about the what and silent about the why. It cannot tell you whether a gap comes from hiring, attrition, promotion, or the external labor market, so representation is a starting point for questions rather than an answer on its own. Larger employers already assemble much of this data for the EEO-1 report, which collects headcount by job category, sex, and race or ethnicity [5].

Hiring and applicant-flow metrics

Hiring metrics track the composition of the applicant pool and how different groups pass through each stage of selection. You look at who applies, then the pass-through rate from application to screen, to interview, and to offer. Where a group's share drops sharply at one stage, that stage is worth examining. The standard screen is the four-fifths rule from the federal Uniform Guidelines on Employee Selection Procedures: a selection rate for any race, sex, or ethnic group that is less than four-fifths, or 80 percent, of the rate for the highest group is generally treated by enforcement agencies as evidence of adverse impact [2]. Adverse impact is a signal, not a verdict. A procedure that produces it can still be lawful if it is job-related and consistent with business necessity, the standard the EEOC's guidance on employment tests describes [3]. The limit of these metrics is that they measure outcomes, not intent, and small applicant pools produce noisy ratios that can cross the four-fifths line by chance.

Pay metrics come in two forms that answer different questions. The unadjusted, or raw, pay gap compares average or median pay across groups with no controls, so it reflects both structural patterns, such as one group being concentrated in lower-paid job families, and any within-role differences. The adjusted pay gap uses a regression that controls for factors like job level, tenure, location, and performance, and reports the difference that remains after those controls, which is the portion that may point to inequity. Both matter, because they call for different fixes: a structural gap is a sourcing and promotion issue, while an unexplained adjusted gap is a compensation issue. The Equal Pay Act requires equal pay for equal work regardless of sex on jobs requiring equal skill, effort, and responsibility under similar working conditions, subject to defenses such as seniority and merit systems [4]. Many employers run these analyses under attorney-client privilege, a point covered further below.

Advancement and retention

Promotion rate and time-in-level measure whether groups move up at similar rates and speed. If two groups are promoted at different rates, or one sits longer at a level before advancing, the advancement data flags it for review, though again the number describes the pattern without explaining it. Retention metrics track turnover by group and, more usefully, regretted attrition, meaning departures of people the organization wanted to keep. A high overall turnover figure can hide that you are disproportionately losing strong performers from one group. Both are aggregate analytics, so collecting and analyzing them sits on the lawful side of the line, and both feed the adverse-impact reviews that some compliance work requires [2].

Inclusion and belonging metrics come from employee surveys, not from headcount data. The common instrument is an inclusion index: a set of survey items on fairness, psychological safety, manager support, and whether people feel free to contribute, combined into a single score you can track and compare across groups. It measures how the workplace feels to the people in it, the part of the picture that headcount misses entirely. The catch is confidentiality. To compare inclusion scores across demographic groups without exposing individuals, survey platforms suppress results for any group below a minimum size. Microsoft Viva Glint, for example, defaults to showing rated scores only when at least five people responded, and it also withholds the next-smallest group so scores cannot be reverse-engineered by subtraction [11]. That threshold protects respondents, and it also means small teams and small demographic groups often cannot get a breakout at all, a real limit on how granular inclusion reporting can be. Because the data is self-reported and anonymized in aggregate, it stays clear of the individual-decision line, though it is only as trustworthy as the response rate and the honesty the survey design earns.

Title VII of the Civil Rights Act makes it unlawful to fail to hire, to discharge, or to otherwise discriminate against any individual in compensation or terms of employment because of that individual's race, color, religion, sex, or national origin [1]. The word that matters is any. The same rule that protects historically underrepresented groups protects everyone, so a program that advantages one group in an individual decision can be challenged by a member of another group. Title VII also states expressly that nothing in it requires an employer to grant preferential treatment to any group because of a demographic imbalance in the workforce [1]. That is the statutory root of the line between analyzing an imbalance, which is allowed, and correcting it through individual decisions, which is where claims arise.

Measurement itself is not only permitted; parts of it are mandatory. Employers with at least 100 employees must file the EEO-1 report each year with workforce headcount by job category, sex, and race or ethnicity, and some federal contractors have historically filed at a lower threshold [5]. Adverse-impact analysis under the four-fifths rule is likewise built into federal selection guidance [2]. An organization cannot meet these obligations without collecting the same demographic data that DEI dashboards use.

The privilege consideration

Pay-equity and adverse-impact analyses are often run under attorney-client privilege. An analysis that goes looking for disparities can surface numbers that, if produced in later litigation, become a roadmap for a plaintiff. Structuring the work so that counsel directs it for legal advice can keep it privileged, but privilege is fact-specific: it turns on how the analysis is set up, who sees it, and whether it is kept separate from routine compliance reporting, and copying a lawyer on an email does not by itself create it [10]. Whether privilege attaches in your situation is a question for your own counsel, and this remains general information rather than legal advice.

The legal text on employment has not changed, but the enforcement climate around it moved sharply since 2023, and that shift is what most people mean by legal and political risk. Four developments matter, and they are worth stating precisely, because headlines tend to blur them.

The Supreme Court's 2023 admissions ruling

In Students for Fair Admissions v. Harvard, decided June 29, 2023, the Supreme Court held that Harvard's and the University of North Carolina's race-conscious admissions programs violate the Equal Protection Clause of the Fourteenth Amendment [6]. The case is about university admissions, not private employment, and it did not amend Title VII. Its effect on the workplace has been indirect, changing the litigation posture and encouraging challenges to employer diversity programs without setting a new employment rule.

The 2025 executive orders

On January 21, 2025, Executive Order 14173 revoked Executive Order 11246, the 1965 order that had required affirmative-action programs of federal contractors, and directed the Office of Federal Contract Compliance Programs to stop enforcing it [7]. A second order, Executive Order 14281 of April 23, 2025, set a policy of eliminating disparate-impact liability to the maximum degree possible and directed agencies to deprioritize enforcement [8].

The 2026 Justice Department opinion

On June 9, 2026, the Department of Justice's Office of Legal Counsel issued an opinion concluding that the EEOC's disparate-impact liability guidelines are unconstitutional, reasoning that Title VII guarantees equal treatment rather than equal outcomes [9]. That opinion is the executive branch's legal position, not a court ruling and not a repeal: the disparate-impact provisions of Title VII remain on the books, and the underlying selection guidance was still publicly posted as of this review [9][2]. Read together, these actions are best understood as a shift in federal enforcement posture and a rise in litigation risk around certain programs, rather than a settled prohibition on measurement. State and local laws apply independently of federal executive action.

That is a risk-and-goals decision for your organization and its counsel, and the honest answer is that the measurement question and the program question deserve separate calls. On the measurement side, several forces point toward continuing. Some demographic data collection is legally required, the EEO-1 chief among it [5], and adverse-impact analysis is part of defending selection procedures under existing guidance [2][3]. Measurement is also how an organization would catch a pay-equity or promotion problem before a plaintiff or a regulator does.

On the other side sit real reasons for caution. Analyses that document disparities create records that can be discovered in litigation unless they are handled under privilege [10]. Collecting demographic and belonging data raises employee-trust and data-privacy questions, especially where groups are small enough that anonymity gets thin [11]. The political environment has also raised the reputational stakes of how metrics are framed and used.

The program question is where most of the current legal exposure actually sits. Continuing to measure representation or pay gaps is different from running a quota, a group-restricted development program, or a protected-characteristic tie-breaker in an individual decision, and it is that second set that draws Title VII challenges now from any group [1]. Many organizations have responded by keeping the analytics and re-examining the programs, treating the two as separable. Which combination fits you depends on your obligations, your risk tolerance, and advice from counsel who knows your facts and jurisdiction. Nothing here is legal advice.