Graduates entering the same profession from the same institutions start at close to the same salary. This is well documented and is frequently cited as evidence that the problem has been solved.
Then, over the following decade, the lines separate. Analyses of administrative earnings data in several countries find that the majority of the eventual gap opens during this period, and that it opens gradually rather than at a single point.
Understanding what happens in that window is more useful than any amount of general advice, because the mechanisms are specific and some of them are visible in advance.
Mechanism one: assignment
Within any organisation, work varies in career value. Some projects are visible to senior decision-makers, involve novel problems, and create a track record. Others are essential, invisible, and create nothing.
Studies of work allocation have found that non-promotable tasks — committee work, organising, mentoring, documentation, administrative coordination — are distributed unevenly, with women receiving and accepting more of them.
Experimental work on this has found that in mixed groups, women are more likely to volunteer for such tasks and more likely to be asked, and that both effects contribute.
The compounding effect is large. Two years of high-visibility project work and two years of committee work produce two very different cases for promotion, from two people with identical ability.
Mechanism two: the first promotion
The transition to first-line management is the single largest leak in most corporate pipelines. Analyses of large employer datasets have consistently found that the ratio of women promoted to manager is lower than their representation at entry level, and that this first step accounts for more of the eventual seniority gap than any subsequent one.
This finding is important because it contradicts the ceiling metaphor. The problem is not at the top. It is at the bottom of the management ladder, and everything above it is arithmetic consequence.
The mechanisms proposed include promotion on demonstrated performance versus promotion on assessed potential — a distinction that consistently disadvantages the group whose potential is assessed more conservatively — and differences in sponsorship, discussed below.
Mechanism three: sponsorship versus mentorship
The distinction is worth being precise about because it explains a persistent puzzle.
A mentor gives advice. A sponsor spends their own credibility advocating for you when you are not in the room — putting your name forward, arguing for your promotion, assigning you to something visible.
Survey research has consistently found that women report having mentors at rates similar to or higher than men, and report having sponsors at lower rates.
That combination explains why mentoring programmes have produced disappointing results against seniority metrics. They supply the input that was not scarce.
Sponsorship is harder to programme because it requires a senior person to take a risk, and people take risks on those they identify with, which reproduces the existing composition of the senior group.
Mechanism four: the negotiation compound
Starting salary differences are small but they compound, because raises are usually percentages of current salary and because subsequent employers frequently anchor on prior pay.
A modest initial difference, compounded across a decade with percentage-based increases, produces a substantial gap without any subsequent unfairness at all.
Jurisdictions that have banned salary history questions provide some evidence here: several studies have found that such bans are associated with increases in pay for women changing jobs, consistent with the anchoring mechanism being real.
Mechanism five: the timing collision
The window in which first promotions occur coincides with the window in which people who have children generally have them.
Research on the child penalty using administrative data finds a sharp and persistent drop in women's earnings following the birth of a first child, and little or no equivalent drop for men. This has been replicated across countries with widely differing policy regimes, though the magnitude varies substantially with policy.
The mechanism is not only time out. It includes hours, role changes, employer changes, and the timing relative to promotion decisions.
It is worth being precise about what this does and does not show. It does not show that the whole gap is caused by children — the gap exists among those without them, though it is smaller. It does show that the largest single discontinuity in the earnings data is located here.
What is actually actionable
Track assignments, not just performance. At the end of each year, the question is not only "did I do well" but "what can I now point to that someone senior saw."
Decline non-promotable work sometimes, and negotiate its distribution rather than its existence. The evidence that it is unevenly distributed is strong enough to make the request reasonable rather than awkward.
Distinguish mentors from sponsors and be explicit about asking for the second. "Would you put my name forward for X" is a different request from "would you give me some advice," and people frequently agree to it when asked directly.
Understand that the first promotion is the gate. Effort concentrated there has higher returns than effort spread evenly across a decade.
And treat the first offer as a compounding decision rather than a starting figure, because that is what it is.
The limit of individual action
None of the above fixes work allocation policy, promotion criteria or parental leave design, and it would be dishonest to present personal strategy as sufficient.
What it does is ensure that a person navigating the system knows where the gates are. Almost nobody is told, and most people discover the assignment mechanism several years after it started operating on them.