If “we use 10%” is your contingency policy, your CFO is one investment committee away from asking the question you can't answer: why 10? Why not 7, or 18?
This guide replaces gut-feel percentages with a defensible, board-ready method based on Quantitative Cost Risk Analysis (QCRA) and Monte Carlo simulation.
Contingency = (P80 cost from QCRA) − (Deterministic base estimate). It will rarely be 10%. It is almost always traceable.
Why percentage-based contingency fails
Three failure modes show up again and again on UK and GCC capital projects:
- It's untraceable. Nobody can tell you which risks the 10% covers, so when a risk materialises, nobody knows if it was “in the pot” or not.
- It under-prices the tail. Cost distributions are right-skewed — the worst 10% of outcomes are far worse than 10% over base.
- It ignores correlation. When steel goes up, concrete usually does too. Independent line-item buffers double-count where risks are correlated and under-count where they compound.
The 5-step QCRA contingency method
Step 1 — Build the deterministic base estimate
You need line items, not a top-line. Quantities × unit rates × productivity factors, by package. This is your deterministic view, also called the “point estimate”.
Step 2 — Identify and quantify risk events
From your quantitative risk register, capture each discrete event with probability of occurrence, min / most likely / max cost impact, and distribution shape (PERT is the safe default).
Step 3 — Quantify continuous uncertainty
Some cost variables don't “happen or not” — they always vary. Examples: productivity, material rates, FX, labour escalation. Model these as continuous distributions (PERT or triangular for sparse data; lognormal for highly skewed).
Step 4 — Run the simulation
10,000+ Monte Carlo iterations in @Risk, Crystal Ball, Safran Risk cost module, or Argo. Validate the model before reading the outputs.
Step 5 — Pick the confidence target & extract contingency
For UK and GCC EPC work, P80 is the most common board-ready target. Some boards want P90 for safety-critical or first-of-a-kind work. See P50 vs P80 vs P90.
Figure 1 — Contingency is the gap between your deterministic estimate and your chosen confidence target on the QCRA S-curve.
Worked example: £44M EPC package
Deterministic base estimate: £44M. After QCRA across 24 risks, 6 continuous uncertainty variables, and Spearman correlation on commodity-linked items:
- P50 = £48.2M → contingency £4.2M (10%)
- P80 = £54.0M → contingency £10.0M (23%)
- P90 = £56.8M → contingency £12.8M (29%)
Why your inputs decide everything
Most QCRAs fail not at the model, but at the inputs. Workshop 3-point estimates routinely understate uncertainty. The simulation faithfully reproduces this bias.
This is why IQRM developed the Risk Data Engine™ (RDE™): it derives uncertainty distributions empirically from your historical project performance data — giving QCRA inputs that actually reflect how your projects behave.
Frequently asked questions
What's a typical contingency for EPC construction?
It varies massively (12–35%+) depending on project type, novelty, and definition maturity. Anyone quoting a fixed figure without QCRA evidence is guessing.
Is management reserve the same as contingency?
No. Contingency covers known unknowns sized via QCRA. Management reserve covers unknown unknowns and is held above the project budget by the sponsor.
Should contingency be held against the P50 or P80?
The contingency you transfer to the project is usually sized to the chosen target (often P80). Some organisations release contingency to P50 and hold the P50→P80 gap as a programme-level pot.
Can I do QCRA in Excel?
For very simple models, yes. For anything with >10 risks, correlations, or integrated cost-schedule, you need purpose-built tools.
Build defensible contingency capability
The QRM Professional Programme teaches QSRA, QCRA, and integrated JCL with hands-on Safran Risk and Argo workshops.
Explore the Programme →Related: What is QCRA? · P50 vs P80 vs P90 · Quantitative Risk Register
