Where it does not

Not every stage of the pipeline leaks equally — and some that attract the most intervention are close to flat.

The pipeline metaphor invites a particular kind of thinking: find each hole, plug it, measure progress. The trouble is that the metaphor implies leaks are distributed, when the data show they are clustered. Spending on broadly distributed fixes — generalist outreach, balanced shortlists at every level, workplace-culture programmes applied institution-wide — is reasonable only if the losses are broadly distributed. Several stages turn out to be roughly equal between men and women, which means resource spent there is solving a problem that exists more weakly than assumed.

Take conversion from application to first interview. In most large British employers with published recruitment data, the gap at this stage is small or absent when like-for-like roles are compared. Women who apply to technical roles in the same grade band are interviewed at rates close to those of male applicants. This does not mean the pipeline is healthy at entry — it means the problem at that exact point is not severe selection bias in interview invitation. The larger entry-side issue is who applies at all: the shortlist is already shaped before the recruiter opens it.

Similarly, promotion rates in the first five years of a technology career tend to show only modest gender gaps in the UK employer data that has been published under voluntary and mandatory reporting. Women promoted from junior to mid-level grades are not, on average, promoted substantially more slowly than men at the same tenure. The disproportion that produces a heavily male senior cohort is not chiefly manufactured at those early rungs. It accumulates later, by a different mechanism, and it accumulates faster than the early-career data would suggest.

Where the effort lands, and why it is displaced

Visible, measurable, early-stage metrics attract disproportionate intervention for a structural reason: they are easy to count and easy to attach targets to. Graduate intake percentages, first-round interview ratios and apprenticeship enrolments produce clean quarterly numbers. That legibility makes them attractive to programmes and to the organisations that report progress against them. The stages that are genuinely flat get improved, and the improvement is counted. The stages that are steep — typically the transition from mid-career specialist to senior, and the years in between when attrition concentrates — are harder to instrument, so they attract less structured effort.

Women promoted from junior to mid-level grades are not, on average, promoted substantially more slowly than men at the same tenure.

There is also a category error that recurs in published commentary. Because women enter many UK technology employers at lower rates than men, early entry stages are assumed to carry the largest losses. But a percentage-point gap at entry, applied to a large intake, and a smaller percentage-point gap ten years later, applied to a much-reduced population, can produce very different absolute outcomes. What looks like a small late-career leak may be removing more women in absolute terms than the more widely discussed entry gaps, precisely because it operates on a narrowed cohort where every departure has greater structural weight.

The graduation-to-employment transition also turns out to be flatter than is sometimes claimed. UK women who graduate in computer science move into employment in the sector at rates that, while not identical to men's, do not account for anything close to the full disparity visible a decade into careers. The substantial divergence happens on the other side of employment, not at the entry to it.

None of this means entry-stage work is wrong. Expanding who considers a technology career and ensuring recruitment processes do not systematically disadvantage women are worthwhile ends. But a pipeline analysis that concentrates resource at the flat points, because they are measurable and responsive to targets, is not the same as one that concentrates resource where the gradient is steepest. The difference matters financially and practically. Programmes designed around entry metrics will improve entry metrics; they will not, by themselves, close a gap whose steepest section lies elsewhere.