Long-form essays on the math, the operating reality, and the institutional failure modes behind capital project decisions. Written for the people in the room when sanction happens.
The McKinsey, IPA, and Flyvbjerg datasets all converge on the same answer, and it's not the one most committees act on. A walk-through of what consistently goes wrong on capital projects above $1B, and the small set of decisions that explain most of the variance.
~10 minute read · April 2026
Reference-class forecasting and Monte Carlo simulation are well understood. So why do sanctioned projects still come in at 1.4× their P50 estimate? A practitioner's walk-through of what to actually do — calibrate inputs, model correlations, decompose variance, and write it back into the sanction case.
~12 minute read · April 2026
"AI for capital projects" gets used to mean everything from chatbot summaries to genuine portfolio optimization. A sober breakdown of the four places AI actually moves the answer for a capital committee — and the four places it doesn't.
~11 minute read · April 2026
These pieces are written by the Capital Project AI research team — engineers and researchers with deep operational and academic experience in capital allocation, megaproject risk, and probabilistic decision-making.
If a piece changes your mind on something, or if it gets something wrong, please tell us. The category benefits from honest argument.
Capital Project AI operationalizes the math we write about. If you run a capital program where the decisions matter, we'd like to help.
Talk to the team →