Algorithms now decide who gets the corner office
The Monday promotion meeting used to smell of espresso and fear. Now it smells of nothing: a silent dashboard ranks the staff, green bars inching up or down while humans watch like spectators at a horse race they no longer bet on.
Inside Coastal Code, JPMorgan and Unilever, the next head of EMEA is chosen by models that chew through 200,000 Slack messages, vacation-day patterns and the tone of quarterly-self reviews. The winner is whoever triggers the fewest risk flags in a sandbox where fake markets crash in fast-forward. HR calls it “leadership potential”; engineers call it overfitting on charisma proxies.
The spreadsheet that deletes gut feeling
Old-school assessments lasted two days, a Marriott ballroom and a plate of soggy croissants. Today the assessment is permanent: keystroke latency, calendar reciprocity, who you invite to a meeting versus who actually shows. One retail giant discovered its future VPs reply 17 % faster to emails from subordinates than peers; that micro-signal now outweighs 15 years of “experience”. The algorithm cannot spell empathy, yet it proxies it with response time. The irony is sold as objectivity.
But data only apologizes for bias once caught. When a French telecom fed its model ten years of appraisals, the machine learned to downgrade anyone whose tenure began with a fixed-term contract; 83 % of those were women. The code was fixed, the pattern resurfaced six months later under a different variable name. Bias does not die; it recompiles.
What still can’t be compressed
Ask the board of Novo Nordisk why they overruled their own ai last winter and the CFO stubs his pen: the model hated the Copenhagen candidate for refusing to relocate. She stayed, revenues in diabetes care rose 12 % in two quarters. Algorithms track movement; they miss momentum created by sheer stubborn conviction.
Meanwhile start-ups sell synthetic empathy: VR simulations where you fire an avatar that cries on command, then an index rates your cortisol stability. Trainers insist it predicts resilience. Skeptics call it emotional deepfake training. Both are right; neither knows the long-term cost.
The quiet truth inside every talent analytics suite is a checkbox labeled “human override”. It is clicked less often each quarter, because opting out drags the recruiter back into litigation territory: if you ignore the machine’s ranking, be ready to explain why in court.
Half-code, half-conscience
Hybrid models are emerging. Google’s re:Lead program keeps the algorithm for the first screen, then deletes every score before the final interview. Recruiters receive a single sticky note: “Ask about the time they apologized in public.” The anecdotal becomes the differentiator precisely because it cannot be scaled.
Investors smell the next compliance wave. Insurance vendors now pitch “algorithmic malpractice” policies in case a rejected candidate sues for digital discrimination. Premiums rise with model opacity. Explainability is becoming a line item next to dental coverage.
By 2026 Gartner predicts 60 % of executive hires will involve a synthetic assessment twin—an ai ghost that shadows the human for a year, updating the probability of a future fall from grace. The file will live beside the dental records. Promotions will hinge on keeping your twin alive.
Corner offices are no longer rewards; they are regression outputs. The view is the same, but the mirror reflects a number. Remember to smile at it—latency matters.