The Reconciliation Myth #4 – “The Day-2 Conundrum” (with apologies to my accounting friends)

Picture this… A cold, winters morning. The sun is barely breaching the horizon. Everything is calm, peaceful and yet… It’s another end-of-month and time for the regular rush to publish production results from the mine and the ore treatment plant. I’m not looking forward to it. The accountants have got their hands on the process and it’s now a mandatory day-two closure for all production and cost/revenue reporting. Everything is rushed to fulfil the demands of the omnipresent reporting gods.

I know from practical experience that there’s no way we will have all the data we need to ‘close the books’ two days after the end of the month. I’m waiting on lab samples, the surveyors, in an attempt to meet totally unreasonable deadlines, started their end-of-month pick-ups three days before the end of month and have yet to provide closing stockpile balances, the mine production haulage data is riddled with incorrect sources and destinations and the mill samples were bolloxed when the automatic tailing sampling system blocked up for a week.

To cap it all off something seems different recently. Over the last 2-3 months our mine-to-mill reconciliation has drifted south and there’s now a 10% difference between the mine’s claimed head grade and the mill’s back-calculated feed grade.

My manager walks in… “So, I guess this difference between the mine and the mill must mean our grade control practices are stuff – what have you been doing to fix it?”

I’m sure this and similar scenarios have been played out month-after- month at mines world-wide and as an industry we’ve tried and failed to resolve the problems. It’s a wicked problem with incomplete data, competing agendas and totally arbitrary and unrealistic time pressures. We lurch from month-to-month managing the symptoms and never quite have time (or the ability) to dig in and understand what is happening.

There are a multitude of possibilities. Regardless, the finger pointing usually starts with the resource model (I’ll discuss the resource in a later blog) and the grade control system. Clearly the mill is always right and thus if we don’t correctly predict the mine’s tonnes and grade (i.e. match the mill) it’s a problem in the mine and the geologists’ responsibility – right.

I’d like to unpick that assumption for a minute. Let’s ignore the cases where there is blatant fraud (yes it happens) and look instead at why automatically jumping on the grade control system may not be a great idea.

Think about the day-two closure. Hmm… OK so the data we need (which is already incomplete) is not going to be fully finalised and available. We make our best guess on missing sample results or out-of-date surveys and even faulting mill sampling systems and we ‘close the books’. The accountants are happy – the mine-to-mill error is pushed back to the mine – but everyone else is left with a sense of frustration. Then the next fire emerges and the end-of-month outcome is put off for another month. What happens next? Well, we absorb the ‘error’ from the previous month and start again. So, absorbing the error? Is that likely to affect this month’s result? Yup, sure is. We’ve just increased the noise in the system. To make matters worse, if we don’t realise we increased the noise we may have reacted to a poor monthly result and made changes to the system. That will only increase the within system variability. Ouch.

That day-two closure is a total fabrication in any case. What’s the likelihood that you know exactly how much metal you’ve made and sold by the second day of the month? Vanishingly small. If you are a base metal concentrate producer it’s zero. You won’t ‘know’ how much metal you sold until all the smelter and shipping adjustments are determined. Do you include those adjustments in you reconciliation? Or do they just disappear? I worked with one operation where shipping adjustments were ‘taken to book’ once a quarter. Any differences between the ore treatment plant’s shipment records and the customers’ tonnes and grade based payments were accounted for at the end of each quarter. On the second day of the month. Regardless of what else happened during the month or the quarter.

I’m sure you can start to see the error and noise inherent in a monthly mine-to-mill reconciliation system. If we are reasonable we’d understand that errors exist – both in estimation and measurement. If you dig back far enough in your science training you may even remember the concepts of ‘significant figures’ and ‘precision’. If you recall, in mathematics we should be reporting results that reflect the reliability of the least precise measurement. Or, to put it another way we can’t make a very precise result when using imprecise measurements, even though a spreadsheet may have 20 decimal places!

Unfortunately, this whole concept doesn’t play well in accounting or management terms. If it did, I’m sure we see a lot less end-of-month reconciliation blues. Instead we slavishly report imprecise estimates and measurements as if they were the gospel truth then worry ourselves to death when our imprecision means numbers don’t match. What’s the definition of insanity again?

However, the challenge goes beyond significant figures. What exactly are we trying to do here again? Why are we reconciling? As I discussed in “The Reconciliation Myth #1”  there’s an understandable but naïve desire to have the mine and mill tonnes and grade match, be it over a day, a week, a month, a quarter or a year. This ‘everything must balance’ logic is, in my opinion, driven by an accounting mindset. I’ve gone so far as to call the end-of-month process ‘ore accounting’ instead of reconciliation. I think it’s a useful distinction. Let the accountants do there, admittedly important, job in terms of reporting production and tracking costs and working capital but let’s not delude ourselves that it has much in common with true scientific process and understanding the variance between our mine and ore treatment plant production estimates.

As a scientist or engineer I have different needs. I need to know how precise my measurements and predictions are. I need to know where the least precise data comes from. I need to understand the limits of the system and to be able to identify:

• Systemic measurement bias;
• Common cause variation (or system noise);
• Special cause variation.

If I don’t understand these three items I’m off to a bad start. I may spend scarce resources (people and time and money) trying to solve problems that don’t exist, or worse by trying to solve a noise-induced problem changing the system settings instead of changing the system.

After all, how do I know what is a ‘good’ reconciliation result and what is a ‘bad’ reconciliation result? When should I react and when should I understand that I’m working within the limits of the system? My manager may think that a 7% difference between the mine and the mill is unreasonable but how did he decide 7% was his tolerance limit?

I often hear of a simple +/-10% deviation on a month-to-month basis being used to say when a mine-to-mill variance is acceptable or unacceptable. How do we, as scientists, know it 10% is significant or not? Why would you spend time and effort trying to solve a problem that may not exist?

Fortunately, there are some ways to help with this all too common problem. It’s going to take some change and going to challenge some people’s paradigms but there’s value to be unlocked – even if it’s only to improve your mood when it comes to day-2 of the month.

Firstly, divorce your ore accounting and reconciliation practices. Let the ore accounting system wrap up on day-2. Let it absorb last month’s errors and historical shipping adjustments as and when it will. Monitor the system and its performance but don’t be fooled into believing it’s providing you with any meaningful information.

Secondly, adopt a scientific approach to reconciliation. Identify the measurements and estimates used and determine the precision of each. Understand that the precision of the calculation is limited by the precision of the least precise measurement.

Thirdly, adopt and understand the practices of statistical process control (SPC) for your reconciliation process. Use SPC and specifically control charts to monitor your system. Work out the system noise (common cause error) or the system capability. How good could the mine-to-mill comparison actually be if everything in the system were working to its best? Use this as a base line for comparison. Be warned… it may not be that +/-10% you are used to using. That’s ok, it just means your error tolerance is less than the system capability. If that’s the case you need to improve the system.

And lastly, for the more advanced reconciliation aficionados, think about testing your practices. While the SPC approach above sounds great and is very useful, it’s often plagued by a high degree of special cause errors – when (wo)man, machine or environment adversely impact on the system. Therefore, it can be hard to know if your reconciliation system is sending you a signal or not. With modern geostatistics and process modelling we can improve on traditional SPC practices.  We can create process models of our production systems and incorporate measurement precision and error into the process model itself. If we then inform the process model with a set of conditional simulation realisations we can create a probability picture of ‘plausible reality’. Because our process model includes production constraints and limits and our conditional simulation may include instances where those constraints are imposed, we get a much better picture of the reconciliation signal. Imagine, instead of a single -7% mine-to-mill variation we now have the techniques and the tools to say something much more meaningful along the lines of “This month’s mine production claim was within 2 standard deviations of the expected capability of our process”. Bingo… I’m within 2 standard deviations. I’m happy – onto the next fire.