Diversity, Equity, and Inclusion (DEI)

The 2020–2024 cycle made DEI highly visible and highly politicized; the 2025–2026 cycle is bringing a quieter, more operational phase. The companies still doing the work have stopped leading with language and started leading with measurement: representation at every level, pay equity by demographic, promotion velocity, attrition delta, and inclusion-index responses from employee listening. The companies who only did the language are quietly unwinding. The ones who did the operations are still there.

Take a 6,800-person financial-services firm. Diversity work: recruiting partnerships, structured interviewing, slate requirements for senior hires. Equity work: annual pay-equity audit with a remediation budget, promotion-rate parity review, benefit-access parity (e.g., does the parental-leave policy actually work the same for frontline and desk roles). Inclusion work: pulse surveys scored by demographic slice, manager training on inclusive-meeting practices, employee resource groups with executive sponsors and real operating budgets. The three streams share a dashboard but not a playbook.

Every company says it cares about DEI. The ones that actually move outcomes share four practices: (1) they measure by demographic slice quarterly and share the data internally, (2) they have a named executive accountable for each of the three streams, (3) they budget for remediation (pay equity adjustments, bias training, ERG support) as a fixed line item not a one-time initiative, and (4) they revise targets annually based on data, not political sentiment. Programs without these four practices become performative within eighteen months regardless of the budget.

In tech, DEI metrics have been tracked publicly for a decade (EEO-1 data, diversity reports), and the outcomes have moved slowly — particularly in engineering and leadership. In manufacturing, the diversity conversation is different: frontline representation is often strong, but equity and inclusion outcomes (pay progression, promotion into supervision, voice in shift decisions) are where the gap shows. In retail, the frontline is diverse by headcount but the corporate office often isn't — the gap is between layers. Same acronym, different diagnosis.

AI creates two specific DEI risks. First, agents trained on historical hiring, promotion, or performance data encode the historical inequities of that data — a recruiting agent that optimizes for "past successful hires" reproduces past bias unless explicitly corrected. Second, the people most likely to adopt agents are the people already comfortable with tech — which in most organizations correlates with demographic factors. The result, without intervention, is a productivity gap that widens along the same lines the organization is trying to close. DEI programs in 2026 have to include AI access and AI bias as first-class concerns, not an afterthought.

• Conflating the three disciplines. A recruiting push ("diversity") without equity or inclusion work produces a revolving door. The new hires leave and the number doesn't move.
• Optics-only programs. Statements, taglines, and annual reports without quarterly measurement are brittle and easy to unwind. The operational work is what holds.
• No executive accountability. Every senior leader "cares about DEI" in principle; almost none have measured goals tied to their performance review. Without that, the work doesn't compete with the business plan.
• Training without behavior change. A unconscious-bias workshop that isn't paired with process change (interview rubrics, promotion calibration, pay audits) is a check-the-box investment.
• Measuring aggregate, not slice. "Our engagement score is 78" hides that the score for your underrepresented population is 62. Slice every metric.
• Political whiplash. Companies that expand and then abruptly contract DEI programs do more damage than companies that never expanded. Pick a level you can sustain and sustain it.