Inclusion metrics that actually move
Representation percentages are easy to count and almost useless. The metrics that predict retention and promotion equity are harder to track and almost never reported.
Most DEI dashboards I see open with a pie chart of headcount by gender and ethnicity. It's the wrong place to start. Representation is a lagging outcome of a hundred upstream decisions — and reporting it without those upstream metrics turns DEI into theater.
1 = low, 10 = high. Based on cross-company benchmarking 2024-26.
- Promotion rate by group+9high predictive value, medium effort
- Pay equity gap (controlled)+9high value, medium effort
- Manager-level inclusion eNPS+8high value, low effort
- Interview pass-through by stage+7medium value, low effort
- Voluntary attrition by group+7medium value, low effort
- Headcount representation %+3low value, very low effort
A 40% female engineering org sounds healthy until you find out 80% of those women are at IC1-IC3 and 0% are in staff+ roles. The single number hid a promotion bottleneck. Always pair representation with flow — hiring rate, promotion rate, attrition rate — by level and by group.
The eNPS question that does the work
"At work, my manager treats people fairly regardless of background."
One question, agree/disagree, run quarterly. Cut by manager. The variance between managers in the same org is usually wider than the variance between companies.
- Promotion equity ratio — promotion rate by demographic divided by representation in the eligible pool. A ratio below 0.9 means you're under-promoting that group.
- Voluntary attrition delta — voluntary leaves by demographic vs. the company baseline. A gap of >3pts is a leading retention signal.
- Performance rating distribution by demographic, with a calibration override rate. If managers' first-draft ratings differ by group but calibrated ratings don't, your calibration is doing the work — and you should investigate why first-drafts diverge.
- Manager-skip rate — % of employees who would not recommend their manager, segmented by demographic. The most actionable inclusion signal in the dataset.
Because the numbers are uncomfortable, and once published they create an obligation to act. That is exactly the point. A metric that doesn't move behaviour is a vanity metric — even when it's labelled DEI.
You do not need a new analytics platform. You need a quarterly extract from the HRIS, a stats analyst for two days, and a leadership team willing to look at the numbers without flinching. Start with promotion equity and voluntary attrition delta — both can be computed from existing data in under a week. Publish them internally, name an owner per metric, and commit to revisit at the next quarterly business review. Iteration beats instrumentation.