AND SELECTION
ESCP Business School · Human Resource Management · November 2026
Amala grows its own people and already hires women well at entry level. Two gaps hold it back: too few women apply, and the AI screen runs unchecked
analysts and associates are women, ahead of four peer boutiques. Seniors are grown from within, a model the research backs.
1 · of applications come from women: the commitment is invisible from outside.
2 · 250 CVs per class go through the standard Claude chat, with no written rules.
1 · Make the signal visible: inside first, then outside.
2 · Put the screen on written rules and a dashboard the partners read.
A partner-led boutique that sells judgement: 50+ people, three practices, no HR function
Acquisitions, disposals, fundraising and capital transactions.
Helps clients structure financing and find suitable lenders.
Strategy, financing and liquidity for private-capital fund managers.
People 54 on the team page, Paris
Grades Analyst → Associate → VP → Director → Partner
Missing an HR function, a careers page, job ads
Interns grow into analysts and VPs, and AI already sits at the first gate: every CV goes through the standard Claude chat
Amala's bet on home-grown talent is backed by research, and at entry level it is ahead of its peers
In an investment bank (5,260 workers), outside hires were paid about 18% more, yet performed worse in their first two years and left more often than people promoted from within.
Bidwell, 2011 · firm-specific human capital, Session 3
Problem 1: fewer women apply, because Amala's commitment is invisible from outside
Candidates cannot see inside a firm, so they read signals; only observable, costly ones are believed (Spence, 1973; Connelly et al., 2011; Session 4).
Fix 1: make the signal costly and visible, starting inside
Every employee knows Amala backs the Women in Finance Chair: they are its first ambassadors.
Women analysts and associates grown in-house meet candidates at events and interviews.
Publish what already works: 50/50 intern classes, 50% women asked at school events, women by grade once a year.
AI helps sourcing: it finds women-heavy programmes and clubs and drafts the outreach. It never selects.
Problem 2: the screen runs on the standard Claude chat, with no written rules, names in view and no record
| THE STANDARD CHAT, TODAY | WRITTEN RULES, AS WE PROPOSE | |
| Instructions | rewritten in each chat | the same every time |
| What the model sees | name and school: so gender and class | the criteria only |
| Memory | one long chat, or none | each CV read alone; decisions saved |
| Trace | none described to us | every decision logged, with a reason |
| Bias check | none: the AI charter covers privacy and GDPR | women and men counted at every step |
Validity and reliability: a screen nobody can describe cannot be shown to measure the right thing, consistently (Session 3, p. 32)
Proxies: school and names stand in for class and gender (Session 3, p. 39; Wilson & Caliskan, 2024)
Names in view: hiding identity raised women's chances in orchestra auditions (Goldin & Rouse, 2000)
Oversight is not a bias check: reviewers tend to follow the AI's hint (Kupfer et al., 2023)
Candidate reactions: AI-led selection is judged less fair (Açikgöz et al., 2020; Newman et al., 2020; Session 3, p. 37)
Fix 2: put the screen on written rules and a dashboard the partners can read
Partners write the criteria; rules run next to today's chat. Every employee briefed on the Chair.
Names and schools hidden; first monthly report. Careers page; first Chair event.
Compare cohorts: do screened-in candidates do better? Decide the next step.
Six smaller risks, each with a cheap fix
| THE RISK | THE CHEAP FIX | THE THEORY |
| “Fit” weighs a lot but has no definition | Write the competencies for each role; judge fit on values, not shared hobbies | Person–organisation vs person–job fit (Kristof, 1996; Session 3, p. 14) |
| Two evaluators must both say yes: one can block | Structured scorecards, scores recorded and checked by gender | Inter-rater reliability (Session 3, p. 32) |
| Unsolicited applications and referrals copy today's network | Keep them for retention; add Chair and school channels for reach | Informal sources lower turnover (Session 3, p. 26) |
| The school is the first thing the table shows | Hide it at screening; test skills instead | Social class, the forgotten diversity (Ingram, 2021; Session 3, p. 39) |
| Equal treatment is claimed, not measured | Track promotion, time at grade and exits by gender | Equity theory (Adams, 1965; Session 5) |
| Little support for women after they join | Link mentoring to careers; track how long women stay | Psychological contract (Rousseau; Session 4) |
A folder of CVs goes in, the rules read them, the dashboard updates
Three sources, cross-checked, every number recomputed from the raw pages
Analysts: Amala says 30% women; the page shows 6 of 17. The first interview's 50/50 is the intern classes.
Senior hires: 7 entries at VP level or above since November 2023, 1 a woman.
Exits: 7 of 16 women and 13 of 53 men in advisory roles have left the page since November 2023. Small numbers.
Seven problems, ranked by their link to AI and to women
Sources
Avgoustaki, A. (2026). Course slides, Sessions 3, 4 and 5. ESCP Business School.
Bidwell (2011) · Spence (1973) · Connelly et al. (2011) · Kristof (1996) · Adams (1965) · Ingram (2021) · Goldin & Rouse (2000).
Pew Research Center (2023) · Açikgöz et al. (2020) · Newman et al. (2020) · Kupfer et al. (2023) · Wilson & Caliskan (2024).
Regulation (EU) 2024/1689 (AI Act), Annex III 4(a) · France Invest & Deloitte (2025), Baromètre parité. Full APA list in our research file.