We develop a framework to estimate latent participation thresholds from binary choices and translate them into the adjustments required for participation. Participation occurs when an observed attribute exceeds an agent-specific threshold drawn from a parametric distribution whose parameters depend on observable characteristics. We select among competing parametric distributions using Vuong tests. Inverting the expected threshold yields agent-specific adjustment requirements. For artificial intelligence adoption by French firms, we find that most firms lie below their workforce threshold. Under our preferred scenario, with complementary digital investments and technical progress, latent demand for ICT engineers reaches about 63,000 full-time equivalents.