Why most capital allocation is sub-optimal
The standard process — rank projects by IRR, fund the list until capital runs out — leaves enormous value on the table. It ignores how risky each project is, how correlated they are with each other, and how much downside the portfolio can absorb before it's in trouble. The result is portfolios that look acceptable on a pitch deck but underperform what the same capital, allocated with proper position sizing, would have produced.
The fundamental insight from quantitative finance — that the right amount to commit to a bet depends on its edge and its variance, not just its expected return — almost never makes it into capital allocation decisions. Capital Project AI applies that math directly: each project gets a recommended commitment level that maximizes long-run growth subject to your drawdown tolerance.
How position-sized capital allocation works
Capital Project AI combines three layers of math: project-level expected value, portfolio-level optimization, and Kelly-style position sizing. Each layer answers a question the others cannot, and the combination is what produces a defensible recommendation.
Expected value, not point estimates
Each project is modeled as a distribution of outcomes — capex, schedule, oil price, throughput — rather than a single number. The expected value is the probability-weighted average across that distribution. Two projects with identical NPV but different risk profiles get treated very differently by the allocator.
Fractional Kelly position sizing
The Kelly criterion gives the mathematically optimal fraction of capital to commit to a bet given its edge and variance. The full Kelly is too aggressive for capital projects (bankruptcy is not an acceptable outcome), so we apply a fractional Kelly — typically 0.25 to 0.5 — with explicit drawdown constraints layered on top.
Correlation-aware optimization
Two projects exposed to the same oil price, the same long-lead vendor, or the same regional permit risk aren't independent. The optimizer accounts for correlations explicitly — which often surfaces hidden concentration risk that the IRR ranking missed.
Drawdown constraints
You set the maximum acceptable portfolio drawdown — say, 20% of equity. The allocator finds the highest-growth allocation that keeps the P5 portfolio outcome above that floor. Bounded risk, optimized growth.
Why Capital Project AI
- Quantitative-finance math, applied to physical capital. Position sizing, drawdown control, and correlation modeling — borrowed from where the math is most rigorous and adapted for capital projects.
- Defensible recommendations. Every allocation comes with explicit assumptions, sensitivity analysis, and a clear walk-away rationale. No black-box outputs.
- Built for the capital committee. Outputs are the priority list, the trade-off table, and the "what changes if we relax this constraint" answer — not a 200-row analyst dashboard.
- From an operator who has done the work. Founded by an engineer who allocated capital across $800M of delivered megaprojects at Shell.
See your portfolio with proper position sizing
Upload your project list — get a Kelly-sized, correlation-aware allocation with drawdown bounds in under a minute.
Open the Dashboard →What it looks like in practice
An infrastructure investor has $400M of dry powder and seven candidate projects: two solar farms, a wind asset, a battery storage facility, a transmission line, and two gas peaker upgrades. The IRR ranking puts the gas peakers first — they have the highest unlevered IRR. The conventional allocation would commit 60% of capital to them.
Capital Project AI runs the same data through fractional Kelly with correlation modeling. The two gas peakers are highly correlated with natural gas price — funding both concentrates exposure. The optimizer recommends one peaker (not two), expands the battery storage commitment because it's anti-correlated with the renewables, and sizes the transmission line conservatively because its returns are highly variance-sensitive to permit timing. Expected portfolio return drops by 80 bps, but P5 outcome improves by 340 bps and maximum drawdown falls from 28% to 14%.
This complements the broader portfolio question covered in project portfolio optimization with AI and the project-level inputs come from AI capital project risk management. EPCs evaluating which jobs to bid use the same engine — see AI for EPC contractors.
Frequently asked questions
What does capital allocation software actually do?
It takes a list of investable projects, your available capital and constraints, and your risk tolerance, and returns a quantified recommendation: how much to commit to each, in what order, with what expected return and downside. Capital Project AI uses Kelly-criterion-style position sizing combined with portfolio-level optimization.
Is the Kelly criterion appropriate for capital projects?
A modified, fractional Kelly is appropriate for portfolios where bankruptcy is unacceptable and projects have fat-tailed downside — which describes most capital portfolios. The full Kelly is too aggressive; we apply a fractional Kelly with explicit drawdown constraints.
How does it integrate with our existing financial models?
Project-level NPV, schedule, and risk inputs can be uploaded from your existing models, or maintained directly in the platform. The optimization runs on whatever inputs you provide — the value is in the math layer that sits on top of them.