The deterministic schedule is a fiction everyone signs
Every capital project starts with a schedule that says it will finish on a specific date. Most don't. The variance has well-understood causes — productivity assumptions are optimistic, vendor lead times slip, weather windows close, regulatory reviews take longer than the team modeled — but the schedule that gets sanctioned almost never accounts for them. Then variance reports document the slip after it has already happened.
AI project scheduling fixes this by treating each activity as a probability distribution, calibrated against your firm's historical performance on similar work. Instead of "this activity takes 14 days," the platform says "P50 14 days, P80 21 days, with weld-rate variance driving 60% of the spread." Sanction at P70 instead of P50, and you're suddenly working from a plan that has a fighting chance of being met.
How AI project scheduling works
Four layers stacked on top of your existing CPM schedule. Each one tells you something the deterministic plan doesn't.
Probabilistic activity durations
Each activity gets a duration distribution — not a point. The distribution is calibrated against your firm's historical productivity for similar scope: weld inches per day, line meters per shift, m³ of concrete per pour. The schedule's underlying probability becomes explicit instead of hidden in the optimism of the original estimate.
Probabilistic critical-path identification
The deterministic critical path tells you which activities matter today. The probabilistic critical path tells you which activities are most likely to matter at finish. Often they're different. Activities with high duration variance can become the actual driver even when they sit off the deterministic critical path.
Reference-class forecasting
Compare the proposed schedule against historical outcomes from structurally similar past projects — your firm's, plus a curated reference dataset. When the bottom-up estimate disagrees with the reference class, that's a signal the planning team is missing something.
Buy-down analysis
The platform identifies the activities where buying down variance is cheapest — accelerated vendor commitments, parallel engineering, redundant fabrication slots — and quantifies the schedule improvement per dollar spent. Management reserve stops being a debate and starts being a quantified investment decision.
Why Capital Project AI
- Calibrated to your firm. Productivity distributions come from your historical performance. The forecasts reflect your reality, not industry averages.
- Sits on top of your scheduler. Reads Primavera P6 and MS Project, runs probabilistic analysis, writes recommendations back. You keep the tools you already trust.
- Fast enough to use live. Run a probabilistic schedule analysis in under a minute. Re-run it during the planning meeting itself.
- Built by an ex-Shell capital owner. Founded by an engineer who has lived through $800M of delivered megaprojects and the slip-data behind them.
Stop sanctioning best-case schedules
Upload your CPM schedule — get the probabilistic forecast, the variance drivers, and the buy-down options in under a minute.
Open the Dashboard →What it looks like in practice
A petrochemicals owner is sanctioning a $250M debottleneck. The deterministic schedule from the EPC says 28 months from FID to mechanical completion. The deterministic critical path runs through the long-lead heat exchangers. Contingency is 14% of capex; schedule contingency is 4 weeks.
Capital Project AI runs the schedule as probabilistic. The deterministic critical path holds for the first 16 months but loses its grip in the construction phase. The probabilistic critical path actually runs through structural steel erection — it has lower mean duration than the heat exchanger path but much higher variance, driven by historical weld-rate variance at the firm's typical contractors. P50 finish moves from 28 months to 30. P80 finish is 35 months. The 4-week schedule contingency covers about P55. The recommendation: sanction at 33 months (P70), spend $1.4M on parallel structural steel fabrication that buys down 60% of the variance on that path, and shift schedule contingency from 4 weeks to 9. Total schedule expectation improves by 4 weeks at P80; total cost increases by $1.4M against a project of $250M.
The same engine powers capital project management software at the portfolio layer and Monte Carlo project simulation on the underlying math. For megaprojects, see megaproject risk management.
Frequently asked questions
What does AI add to traditional CPM scheduling?
Three things: probabilistic activity durations calibrated against your historical productivity (instead of single-point estimates), automatic identification of which activities are actually likely to drive slippage (not just the deterministic critical path), and reference-class forecasting against similar past projects to catch optimism bias before sanction.
Do we keep using Primavera or MS Project?
Yes. Capital Project AI reads from Primavera P6 and MS Project schedules and writes recommendations back. The scheduling tools stay; we add the probabilistic and historical-calibration layer on top.
How much historical data do we need?
Useful results start at 8-10 completed projects of similar scope. Strong calibration starts at 20-30. If you don't have that yet, the platform falls back to a curated reference dataset of public megaproject outcomes for the relevant industry.