This summary describes six steps for introducing diversity management and says where research is thin. It is not legal advice. Legal position and retrieval date: 13 November 2026.

Under section 12(1) of the General Equal Treatment Act (AGG), the employer must take the necessary measures to protect against disadvantage, including preventive ones. Suitable training of employees counts as fulfilling these duties (section 12(2)). The AGG and the complaints bodies under section 13 must be made known (section 12(5)). A programme beyond that is voluntary. The bill amending the AGG (21/6178) was not adopted on 13 November 2026; the existing text applies.

The Charter of Diversity is voluntary; four companies started it in December 2006, and an association has run it since 2010. In a 2025 press release the association speaks of more than 6,000 organisations with around 15 million employees.

Six steps

The order is my own.

  1. Baseline analysis. Data on ethnic origin, religion, health or sexual orientation are special categories under Article 9 GDPR, so consult data protection officers early.
  2. Goals. Verifiable, derived from the analysis.
  3. Measures. Kalev, Dobbin and Kelly (2006) compared measures across 708 US workplaces from 1971 to 2002 (16,265 workplace-years). Training and evaluations aimed at managers' bias showed little or no positive effect.
  4. Responsibility. The same study found the clearest effects from affirmative action plans, diversity committees and full-time diversity staff.
  5. Indicators. Structure figures, process figures and survey data, each with an interpretation and a comparison period.
  6. Transfer. Training embedded in other diversity measures achieved higher effects (Hedges' g = 0.57 against 0.36).

What training achieves

Bezrukova and colleagues (2016) analysed 260 studies with 29,407 participants. The overall effect was g = 0.38, falling to 0.20 after correcting for publication bias. Effects were weakest for attitudes and affect (0.30), stronger for knowledge (0.57) and behaviour (0.48). Knowledge stayed stable; effects on attitudes and reactions faded. 89 percent of participants were from the USA.

Limits of the business case

Van Dijk and colleagues (2012) analysed 146 studies with 612 effect sizes. Demographic diversity related slightly negatively to team performance on subjective ratings (r = −0.05), task-related diversity slightly positively (r = 0.04); for objective measures the groups did not differ significantly. Green and Hand (2024) found no significant link when trying to replicate McKinsey analyses. A sound case therefore need not rest on profit; it can rest on section 12 AGG, fairness and verifiable goals.

This is a summary. The full German article with all details is at the same address under the German site.