Why most portfolio decisions are bad portfolio decisions
Capital committees ration capital across competing projects. The math that should drive that rationing — expected value, risk-adjusted return, opportunity cost, correlation — usually lives in a spreadsheet that one analyst maintains, that nobody on the committee fully trusts, and that quietly encodes whichever assumptions were live the last time it was rebuilt. The result: project champions with the loudest narrative win, the highest-NPV opportunities sometimes get cut, and the portfolio drifts further from intent every quarter.
Re-baselining is reactive. By the time the variance reports show that the portfolio's expected return has fallen, the deferral or scope-cut decisions are constrained by sunk-cost commitments and contractor mobilization clauses. The result is a portfolio that looks acceptable in the executive summary but underdelivers against what the same capital, allocated differently, would have produced.
How AI portfolio optimization works
Capital Project AI treats portfolio decisions as a quantitative optimization problem: maximize risk-adjusted value subject to capital, schedule, and resource constraints. The math has been around for decades — what changes with AI is the speed of running it on the full scenario space and the quality of the inputs at the project level.
Expected value across scenarios
Each project is evaluated across thousands of price, schedule, and execution scenarios. The output is not a single NPV but a distribution — with means, P10s, P90s, and downside risk separated cleanly. Two projects with the same expected NPV can have very different distributions, and the optimizer treats them differently.
Capital constraint solver
An integer-programming solver handles indivisible projects (you fund the LNG train or you don't), capital phasing across years, and shared engineering or procurement resources. The output is a ranked priority list with explicit substitutions: drop project D, accelerate project A, and the portfolio NPV moves by $42M.
Real options for stage-gated projects
Projects waiting at a stage gate are modeled as options: you pay a small amount today to keep the right to commit later, when more information is available. The optimizer values that flexibility correctly — which is why stage-gated FEED studies often look much more attractive than a deterministic NPV would suggest.
Correlation handling between projects
Two projects exposed to the same oil price, same long-lead vendor, or same site permit aren't independent bets. The optimizer accounts for the joint risk and prevents you from quietly concentrating exposure even when each individual project looks acceptable.
Why Capital Project AI
- Portfolio-native, not project-native. Most tools optimize one project at a time. Capital Project AI optimizes across the portfolio in a single pass — the math that capital committees actually need.
- Built around the capital committee workflow. Outputs are the priority list, the sensitivity analysis, and the walk-away recommendations executives ask for — not a 200-row dashboard.
- Fast enough to use mid-meeting. Re-run the optimization with a different constraint and have the answer before the next slide.
- Built by an ex-Shell capital allocator. Founded by an engineer who oversaw $800M of delivered megaprojects and learned the hard way which decisions matter.
Run a portfolio optimization on real data
Upload a project list and a capital constraint — get a ranked priority list and the trade-offs in under a minute.
Open the Dashboard →What it looks like in practice
A common pattern: an upstream operator has $500M of annual capex and twelve candidate projects ranging from a brownfield expansion to a greenfield development to two pilot CCS projects. The current plan funds eight of them, ranked by IRR. Capital Project AI re-runs the allocation as an optimization with three constraints: total capital, total engineering hours, and a maximum portfolio variance target.
The result reorders the priority list. The two highest-IRR projects stay, but the third — a debottleneck the team had championed for years — drops to the bottom because its return is highly correlated with the brownfield expansion already funded. A pilot CCS project the IRR ranking had cut moves into the funded set because it adds uncorrelated upside that lowers portfolio variance. Net portfolio NPV moves up by $58M; expected variance comes down 11%.
The same machinery powers capital allocation software at the position-sizing layer, and the project-level inputs come from AI capital project risk management. For EPCs evaluating which contracts to bid, the equivalent question is covered in AI for EPC contractors.
Frequently asked questions
What is AI project portfolio optimization?
It uses constraint solvers, expected-value math, and scenario simulation to choose the best mix of capital projects to fund under a fixed budget, schedule, or resource constraint. It replaces gut-feel ranking with a quantified, defensible recommendation that updates as inputs change.
How is this different from a custom Excel model?
Excel models age badly. They encode last year's assumptions, get hard-coded by whoever built them, and silently break when the portfolio shape changes. Capital Project AI runs the optimization on live inputs with full scenario coverage and re-runs in seconds — so the answer is current, not historical.
Does it work with stage-gate processes?
Yes. Stage-gated projects are modeled as real options: you pay to keep the option open, with a future right to commit at the next gate. The optimizer values that flexibility correctly rather than treating the project as all-in or all-out.