What DEI Programs Actually Do Inside a Company, and What Works

This is commentary. The facts, cases and studies cited are sourced and checkable; the conclusions are mine.

Most people's entire experience of DEI is a 45-minute e-learning module with a progress bar that won't let you skip ahead, a scenario about a fictional employee named Devon, and a checkbox at the end certifying you completed it. You click it. You go back to work. Nothing in your job changes.

That module is the visible tip of something bigger, and the bigger thing is worth understanding before anybody argues about it. So let's take the whole apparatus apart and look at what the research actually found.

The machinery, piece by piece

A corporate DEI program is usually six or seven separate mechanisms wearing one name.

There's mandatory training, the module. There's diverse slate policies, which require a certain demographic mix among candidates interviewed before a hire is approved. There are employee resource groups, affinity networks with a budget and an executive sponsor. There's supplier diversity, which sets targets for spend with certified minority- or women-owned vendors. There are representation goals, often published in an annual report with a date attached. There's executive compensation tied to those goals, where part of a leader's bonus depends on hitting them. And there's a dashboard somebody presents quarterly.

Some of those are innocuous. Some of them are a lawsuit waiting to be filed. And the reason to know which is which is that they fail and succeed for completely different reasons.

The best evidence against the training came from people who believed in it

Frank Dobbin and Alexandra Kalev are sociologists, not culture warriors. Dobbin is at Harvard. In a 2016 Harvard Business Review piece called "Why Diversity Programs Fail," they laid out what they'd found looking at roughly three decades of data from more than 800 U.S. firms.

Their conclusion was blunt. Mandatory diversity training, hiring tests and formal grievance procedures — the three things companies spend the most on — didn't improve managerial diversity, and in some cases the numbers moved the wrong way afterward. Their earlier paper with Erin Kelly in the American Sociological Review in 2006, covering 708 workplaces over three decades, had reached a similar place: structures that assigned somebody clear responsibility did something measurable; training and evaluations did close to nothing.

The mechanism they proposed is one any manager recognizes. Tell an adult they're required to sit through a lecture about their own moral failings, and they don't come out humbled. They come out annoyed. Compliance training is for fire exits and expense fraud, not for how a person feels about a coworker.

A 2016 meta-analysis in Psychological Bulletin by Katerina Bezrukova and colleagues, pooling forty years of diversity training evaluations, found the same shape from a different angle. Knowledge gains stuck reasonably well. Attitude changes faded. The longer the gap between the training and the follow-up measurement, the less was left.

Unconscious bias was the weakest link

The intellectual foundation under a huge share of corporate training is the Implicit Association Test and the idea of measurable unconscious bias you can train away.

That foundation isn't holding up. A 2019 meta-analysis led by Patrick Forscher, covering hundreds of studies, found that interventions could shift scores on implicit measures — but those shifts didn't reliably produce changes in actual behavior. Anthony Greenwald, who co-created the IAT, has publicly said for years it isn't suitable for diagnosing or predicting an individual person's conduct.

Which leaves a lot of corporate programs in an awkward spot. They're built to fix a thing the test can't reliably detect, using methods that don't reliably change what people do.

The business case has a replication problem

For a decade the argument that closed every boardroom debate was McKinsey's Diversity Matters series, starting in 2015, reporting that companies in the top quartile for executive diversity were meaningfully more likely to post above-average profitability.

In March 2024, Econ Journal Watch published a paper by accounting researchers Jeremiah Green and John Hand attempting to reproduce that result using S&P 500 data. They couldn't. They found no statistically significant relationship between executive racial and ethnic diversity and the financial performance measures they tested, and they were unable to quasi-replicate McKinsey's published findings.

McKinsey disputes the critique. Fine. But the honest position is that the headline claim isn't settled science, and it was sold as settled science to a lot of executives who never read past the summary.

Then the law moved

Students for Fair Admissions v. Harvard, decided in June 2023, was about college admissions. General counsels read it anyway.

In April 2024, in Muldrow v. City of St. Louis, a unanimous Supreme Court lowered the bar for what counts as a harm under Title VII — an employee no longer has to show "significant" injury from a discriminatory job transfer, just some disadvantage in the terms of employment.

In June 2025, in Ames v. Ohio Department of Youth Services, the Court went 9-0 again, this time holding that plaintiffs from majority groups don't have to clear an extra "background circumstances" hurdle to bring a Title VII claim. Justice Jackson wrote it. There was no dissent.

Title VII never contained an exception for good intentions. It says you can't make employment decisions because of race, color, religion, sex or national origin. It doesn't say which race.

And in January 2025, an executive order revoked Executive Order 11246, the 1965 order that had required affirmative action programs of federal contractors for sixty years.

The retreat, with dates

Tractor Supply in June 2024. John Deere in July. Ford, Lowe's and Harley-Davidson in August. Boeing dismantled its global DEI department in November. Walmart announced that same month it would wind down its Center for Racial Equity and drop racial targets in supplier programs. Meta, McDonald's and Target all announced changes in January 2025. Verizon and T-Mobile made commitments while merger approvals were pending in 2025.

That's not a trend signal. That's a public record of corporate filings and statements.

What I'll concede

Not everyone is running. At Costco's annual meeting in January 2025, shareholders rejected a proposal to report on DEI risks by roughly 98 percent. Apple shareholders turned down a similar proposal a month later. Some boards looked at the same evidence and decided to stay put.

And plenty of people working in these roles are decent professionals who inherited a program somebody else designed. Hitting the policy is fair game. Sneering at the person administering it isn't, and it doesn't win anybody over.

What the same research says does work

This is the part that gets skipped. Dobbin and Kalev didn't conclude that nothing helps. They found real effects from:

• Targeted recruiting — going to schools and pipelines you weren't visiting before, which adds candidates instead of excluding them
• Mentoring and sponsorship, formally assigned rather than left to chance
• Task forces of line managers with actual accountability, not an HR committee
• Structured hiring — the same questions, in the same order, scored on the same rubric, for every candidate

That last one is the sleeper. Structured interviews beat unstructured ones on predicting job performance in the industrial psychology literature going back decades. They're also the cheapest thing on this list, and the most boring, which is probably why nobody sold a keynote about them.

If you want to do something Monday

Pull your own company's policy documents. Not the poster in the break room, the actual policy. Then ask one question about every program in it: is any employee excluded from this on the basis of race or sex?

A mentoring program open to everybody is a mentoring program. A leadership track open to some employees and not others is a Title VII exposure with a nice logo. That's the line, and after Ames, it's a line with unanimous Supreme Court authority behind it.

If the answer makes you uneasy, that's a question for an employment attorney licensed in your state, not for a comment section.

The modules were never the problem. They were the alibi.