When a diversity target is missed in early-career hiring, the post-mortem usually examines the interview process. The arithmetic almost always points earlier.
The funnel composition problem
If the top of your funnel does not reflect the composition you are targeting, no amount of interview-stage adjustment will produce it. Representation varies enormously between institutions, so an inherited campus calendar largely determines your outcome before anyone is interviewed.
Widening the institution set is the primary lever
Adding institutions with different composition profiles changes funnel composition directly and immediately. It is a sourcing decision, it is measurable, and it does not require changing your bar.
Structured screens reduce variance
Unstructured interviews amplify whatever assumptions the interviewer brought. A defined rubric applied identically across candidates narrows that variance, and it improves signal quality for everyone, not only for underrepresented candidates.
Measure at every stage, not at the end
Track composition at each funnel stage. A drop concentrated at one transition is a specific, fixable problem; an outcome measured only at offer stage tells you nothing about where to intervene.
Key takeaways
- Funnel composition at the top determines the outcome; interview-stage fixes cannot compensate.
- Widening the institution set is the primary, measurable lever and does not lower the bar.
- Structured rubrics reduce interviewer variance and improve signal for all candidates.
- Measure composition at every stage transition to locate where drop-off actually happens.
Questions
Does widening the funnel mean lowering the bar?+
No — it means sampling from a larger population before applying the same bar. The two are frequently conflated, and the conflation is what stalls most diversity programmes.
How do we source for composition without discriminating?+
Change where you look, not who you accept. Adding institutions and channels is a sourcing decision; the selection criteria stay identical for everyone.