What a Bankable Feasibility Study's Accuracy Band Actually Leaves Out
A Bankable or Definitive Feasibility Study almost always gets quoted with a cost accuracy figure attached — something in the order of ±15%, ±20%, sometimes tighter. That number tends to get treated as the finished picture: the study said the number, the number has a margin, the margin is the risk. It isn't that simple, and the gap between what the accuracy figure actually promises and what people assume it promises is where a lot of later trouble starts.
“A Bankable Feasibility Study’s accuracy figure — typically an AACE Class 3 estimate — is based on only 10–40% of the project’s total engineering, with the remainder filled in by factored and historical cost data rather than project-specific engineering. The accuracy band assumes the BFS flowsheet carries unchanged into FEED, which often isn’t the case, and it describes methodology rather than guaranteeing the outcome. The gap between expected and actual cost/schedule step-up widens further on orebodies with low or variable grade, difficult or expensive mining conditions, complex extraction technology, or projects intersecting culturally or ecologically sensitive land.”
What the accuracy band is actually measuring
The industry's own classification system is more precise about this than the shorthand version usually is. AACE International's* cost estimate classification system — the standard reference most feasibility-stage estimates are implicitly built against — defines a Class 3 estimate, the level typically associated with a bankable or definitive feasibility study, as one built on roughly 10% to 40% project definition. Its expected accuracy range is around −10% to −20% on the downside and +10% to +30% on the upside.
Read that carefully and two things stand out. First, the accuracy figure is explicitly a statement about the quality of the estimate given how much of the engineering exists at that point — not a guarantee about where the final number lands once the rest of the engineering gets done. Second, the band is asymmetric: the downside risk is smaller than the upside risk, which is a formal way of saying a BFS is structurally more likely to understate final cost than to overstate it.
The 60% of the project that isn't engineered yet — it's estimated by proxy
Ten to forty percent project definition doesn't mean the other sixty to ninety percent is simply missing or left blank. It's filled in — but by factored and parametric methods, not by project-specific engineering. A Class 3 estimate is built substantially on semi-detailed unit costs and assembly-level line items drawn from cost databases, historical project data, and standardised factors, rather than from a fully engineered, quantity-surveyed design. That's normal, appropriate practice at this stage — it's precisely what makes a BFS achievable in a reasonable timeframe at a reasonable cost.
But it carries a specific, easy-to-overlook assumption: that this project resembles the reference projects and historical data the factoring is drawn from, closely enough for the factors to hold. For a fairly conventional orebody and a well-proven flowsheet, that assumption is usually reasonable. For anything unusual — and it's worth being honest about how often "unusual" turns out to be the case — the factored two-thirds of the estimate is carrying more risk than its accuracy class implies, because it was never actually testing this project's specific conditions in the first place.
The flowsheet the accuracy band was actually built on
There's a second, quieter assumption baked into the whole exercise: that the flowsheet the BFS accuracy figure was calculated against is the flowsheet that actually gets carried into FEED and final engineering. It often isn't. Testwork gets refined, technology selections get revisited, and — as anyone who's watched a project move from BFS into detailed design will recognise — it's common for meaningful elements of the process design to shift once the engineering that wasn't there at BFS stage actually gets done.
When that happens, the accuracy band stops meaning what it appeared to mean. It was calculated as the expected variance around a specific design basis. If the design basis itself changes, the comparison between the BFS number and the eventual FEED or execution number isn't really measuring estimate uncertainty on the same project anymore — it's comparing two different designs, and the accuracy band was never built to describe that gap. This is a second, separate way the true step-up in cost and schedule can end up well outside what the stated accuracy range implied, layered on top of the factored-estimate risk above rather than replacing it.
Where the expected step-up stops being manageable
None of this means every BFS-to-FEED transition is a crisis waiting to happen. Most boards anticipate some step-up in cost and schedule as engineering matures — that's a reasonably well-understood feature of how these studies work, even if the full extent of it sometimes still comes as a surprise. The step-up becomes genuinely dangerous, rather than merely expected, when it's compounded by specific characteristics of the orebody or the project itself:
Low-grade or highly variable mineralisation, which is harder to represent reliably in early-stage sampling and testwork and more sensitive to head-grade assumptions carrying through into the revenue model.
Orebodies that are difficult to mine selectively or cleanly, where dilution and mining recovery assumptions are easy to understate at BFS stage.
Deposits that are expensive to access — depth, geotechnical complexity, or unconventional mining methods — where the factored cost data is least likely to reflect the project's actual conditions.
Extraction processes that are inherently complex or technically demanding — nickel pressure acid leach (PAL) circuits are a well-known example of a technology category with a track record of underestimated capital and operating cost at early study stages.
Projects that intrude on land of cultural or ecological significance, where risk can emerge gradually through community engagement processes running concurrently with the engineering — alongside the more general risk of local opposition delaying approvals or forcing design changes.
Any one of these raises the stakes on getting the BFS-to-FEED transition right. Several of them together — which isn't a rare combination — is exactly the situation where the gap between the accuracy band's implied risk and the project's real risk is largest.
Why this particular gap is hard to see from the outside
None of this shows up in a board pack as a red flag, because nothing about it looks wrong. The study meets its stated accuracy class. The financial model is built correctly on top of it. Every individual number can be defended. What's invisible is the combination — the asymmetry in the accuracy band itself, how much of the estimate is factored rather than engineered, whether the flowsheet is actually stable, and whether the specific orebody or project carries one or more of the risk amplifiers above. Each on its own is a known, well-understood feature of how feasibility studies work. Together, they're rarely added up explicitly by anyone in the room.
That's precisely the kind of thing an independent read of the technical basis, done specifically at the BFS-to-FEED transition rather than only at FID, is built to catch. In practice that means something quite specific: tracing which line items in the estimate are engineered versus factored, and checking whether the factored ones sit against the project's actual risk amplifiers rather than a generic reference case. Confirming the flowsheet the accuracy band was calculated against is still the flowsheet being carried into FEED. And weighing the accuracy class against the specific orebody in front of it, not the average project the classification system was built around. None of that requires the original study to have been done badly — it requires someone with no role in having written it to go looking, specifically, for where the compounding happens.
*Source: accuracy ranges and project-definition percentages are drawn from AACE International Recommended Practice 18R-97, "Cost Estimate Classification System — As Applied in Engineering, Procurement, and Construction for the Process Industries."