When Artificial Intelligence meets corporate finance in the USA

We propose an Artificial Intelligence (AI) model that provides a transparent, scalable tool for monitoring nonlinear production relations in corporate finance. Linear models can distort inferences about scale, productivity, and efficiency when production technologies are nonlinear. We address this by developing a parsimonious neural production model that uses a single-hidden-layer feedforward neural network with one hidden neuron and delivers observation-specific elasticities, returns to scale, technical efficiency, and total factor productivity via closed-form derivatives. Estimated on 174,190 Compustat North America firm-year (2000–2025) observations using within-year demeaned logs of labor, capital, and R&D to explain sales, the model reveals increasing returns to scale, high technical efficiency among adopters, and sizable positive total factor productivity movements. A two-hidden-neuron specification yields only marginal improvements in fit and preserves these economic patterns. The architecture keeps the parameter count low, produces interpretable unit-level diagnostics, and is fully replicable for applied research and financial decision-making. Also, the model systematically outperforms the popular Cobb–Douglas production function. Operationally, the framework extends to supervisory and investment use, such as tracking nonlinear production, comparing units, and prioritizing interventions.