You are Patricia Calderón-Mora, chief human resources officer of Caldorra, an energy group headquartered in San José, Costa Rica. Your Audit Committee wants, in eighteen days, a written operating model for the eleven artificial-intelligence pilots running in your function. You start with the hardest one to call: the model that predicts who is going to leave.
The situation
Caldorra billed about US$4.8 billion in 2025 and employs 7,000 people in four countries: about 2,400 in Costa Rica, 1,950 in Brazil, 1,400 in Mexico and 1,250 in Spain. Four different labour regimes, with Spain inside the European Union's AI regulation. Over fourteen months your HR function, about 110 people, piled up eleven AI pilots with US$8.9 million of direct spend; US$11.4 million counting internal time and integration.
On 21 April the Audit Committee asked you to present, at its 9 May session, which decisions are made by people, which by software, and on what basis the line was drawn. And that no new pilot start in the meantime. Your first task is to give each of the eleven a fate: kill it, scale it, re-scope it or put it on hold.
Pilot 5 is the retention-risk model sponsored by Ileana Castro-Brenes: a commercial product with custom work on top, US$0.6 million, live in Costa Rica and Spain. Last year it flagged in advance 14 of the 22 valued people who ended up leaving. To do it, in Spain, it uses metadata about employees' email patterns that your counsel has not cleared.
Five people will read your paper: the CEO, who wants a story he can tell outside without later retracting a number; the chief operating officer; the CFO, who is looking for the saving; the lead independent director, who reads one page per topic and asks one question per meeting; and Marisol Jiménez-Cordero, of the refinery workers' council, who wants to know who an operator talks to when software says no.