Redteam recently conducted some research showing that, for major gold mines started in the past 10 years, metal prices have been 45% higher than expected and ore grades have been 10% lower. If the goal is accurate forecasting during project evaluation, both deviations are a problem; however, the inherent uncertainty of metal price and ore grade feels very different. The one truth about the gold price is that it will be different tomorrow, while the true ore grade couldn’t have changed between project forecasting and operation. Metal price and ore grade represent the two very different types of uncertainty that a mining project can face – those are aleatory and epistemic uncertainty, to use the technical terms. To manage uncertainty (and risk) effectively, we first need to understand what type of risk we’re dealing with.
On one hand, aleatory uncertainty refers to a lack of predictability; this type of uncertainty is fundamentally unknowable today. It includes things like the temperature outside three years from now, the outcome of a coin toss, and the results of the next UK election. While the quality of predictions can vary for aleatory uncertainty, the uncertainty itself is irreducible – a fair coin will land heads or tails and there’s nothing you can do ahead of time to predict the outcome with more certainty.
On the other hand, epistemic uncertainty refers to a lack of knowledge; this type of uncertainty is fundamentally knowable. It includes things like the current temperature outside, how many jellybeans are in a jar, and the location of your car keys. In these cases, the information exists today, and the uncertainty could be reduced by applying effort and resources to collect it. You could take a thermometer outside, you could empty out the jar and count the jellybeans and, for the love of God, those keys must be somewhere! The critical question in cases of epistemic uncertainty is whether the reduction in uncertainty – or risk – is worth the effort.
In a mining context, epistemic uncertainty shows up as things like the price of capital equipment, the surface elevations for the construction of a new haul road, orebody grades, and metal recovery. In each case, the information exists and is fundamentally knowable; however, the effort required to gather the information differs. In the case of equipment prices, you can solicit firm quotations or sign a binding, fixed-price purchase contract. The effort required to obtain the information is the time spent soliciting quotations, negotiating the contract and perhaps the value of any financial deposit required. The surface elevations for the construction of a new haul road exist, and the information can be collected through a routine land survey. In both cases, the effort needed to gather the information is modest and the management decisions to collect it don’t require much brainpower – it’s unusual to execute a capital project without equipment quotations or build a haul road without a survey. However, as the information gets more resource intensive to collect, the management decisions get more difficult.
The metal content of an orebody is information that exists today and will not change within the timescale of building and operating a mine – the mining company can have all this information by the end of the mine life. However, with current technology and processes, the true metal content of an orebody is prohibitively expensive to obtain during mine planning. So, we collect some information through drilling and then model and argue and drill and model and argue until we have a satisfactory estimate of the metal content for planning and forecasting purposes. Similarly, the uncertainty around life-of-mine metal recovery is dealt with in a similar way. Samples are collected representing different mineralogical zones in the orebody and tests are run to simulate plant conditions and predict metal recovery during operation. Confidence in the estimates can be increased by running tests with larger and more varied samples. Both examples illustrate a key method of trying to reduce epistemic uncertainty – collecting samples of the information and creating models to predict the unknown. With epistemic uncertainty, the management challenge lies in valuing the collection of information and modelling effort against the likelihood that additional knowledge will change your decisions.
In the early stages of a project, especially before any significant decisions have been made, it’s easy to justify efforts in gathering information, running experiments, and creating models to make big reductions in uncertainty. As the uncertainty diminishes, decisions get harder and can be uncomfortably subjective. What’s the likelihood $1M in drilling will change a $20M mine design decision? More than 5%? You’ll be lucky if the answer seems obvious when the question comes around. But as more information is collected, as the estimates become solidified, and as the uncertainty reduces to the point where more information is unlikely to change the decisions of record, you can become comfortable with the remaining epistemic uncertainty because you understand it and have made a conscious and informed decision to leave that risk on the table.
In the gold project data we collected, the actual ore grades ranged from 65% higher to 67% lower than expected. On the other hand, average gold recovery was 1% lower than expected with a range of +3% to -10%. Would different decisions have been made at the planning stage of these mines if there had been better knowledge of the true ore grades? What about with better knowledge of actual recovery rates? The answers may seem obvious in hindsight, but the critical skill is understanding the uncertainty you’re dealing with ahead of time so economically rational and risk-informed decisions can be made before it’s too late.