You finalised the blend plan on Tuesday. Two-thirds from Stockpile A, one-third from Stockpile B, target iron content at the contract midpoint, and moisture slightly above the comfort line but within tolerance. The sign-off happened in the weekly shipping programme review; everybody nodded, and the calculation was saved to a workbook nobody will open again until Sunday night, when the reclaim sequence goes live.
Somewhere between Tuesday afternoon and Sunday evening, the stockpile changed. A feed conveyor went down on Wednesday, and the swing shift redirected four hours of production from another face into Stockpile A. Thursday morning, a truckload from the satellite pit was tipped onto it. Friday, the grade control team found a small pocket of higher-ash material in the active bench, and it went into Stockpile B rather than the waste dump because that is where the feed was running. None of these was a wrong decision. Each of them was fine in isolation. The problem is that the blend plan, the document your vessel is loading against, is now calibrated to a stockpile that no longer exists.
This is the quietest, most common cause of Stockpile Blend Management failure in bulk commodity logistics. Not a bad analysis. Not a rogue operator. Not a contract dispute. A blend plan that was correct on the day it was calculated but is no longer correct on the day it is executed, because the stockpile the plan was built against has continued to evolve after the plan was locked.
This post walks through the five most common Stockpile Blend Management failure modes, explains why each of them is structurally invisible in most operations today, and lays out the system-level controls that replace human vigilance with automated reconciliation. If you have ever signed off on a blend plan and then felt a low-grade unease about whether it will still be true four days later, this is the piece for you.
Before the failure modes, a minute on why stockpile blending is hard in the first place. A bulk stockpile of iron ore, coal, bauxite, nickel concentrate, anything that is stacked and reclaimed rather than bagged and stored, is not a uniform pile of identical material. It is a geometric record of everything that was ever placed on it, in the order it was placed. The top layer was added yesterday. The core was there three months ago. The reclaim bucket or the bridge scraper is pulling from the boundary of those layers, not mixing them evenly.
This means the effective grade of material leaving the stockpile at any moment is a weighted average of several layers, with the weights depending on the geometry of the reclaim sequence. If the reclaimer works the stockpile front-to-back in even passes, the average converges on the stockpile’s blended profile. If the reclaimer starts at one end because that is where the current Capesize’s first hold is being filled, the material in the first 20,000 tonnes of the stockpile is disproportionately weighted toward whatever was added most recently at that end.
This is well-understood by mining engineers and is modelled in commercial stockpile management Software. The problem is almost never that the model exists. The problem is that the model, the blend plan, and the logistics programme live in three different systems, maintained by three different teams, reconciled once a week by email. By the time the physical stockpile diverges from the modelled stockpile, the people who built the blend plan do not know it has happened.
The most common rupture is the simplest. The blend plan was calculated from stockpile analysis data that were current as of Monday. New material was added on Tuesday. The blend plan did not update because the analysis data on which it was built were frozen at the time of sign-off.
In a typical mid-sized iron ore or coal operation, a stockpile will receive several thousand tonnes per day. Over a four- to seven-day window between blend signoff and vessel loading, the stockpile’s analysis profile will have drifted noticeably. How much depends on the variability of the feed and the tonnage through, but for a stockpile running at 40 per cent of its capacity with daily additions of 5 per cent, even a small drift in the feed grade becomes meaningful at the edges of the contractual tolerance.
The fix is not to update the blend plan more often by hand. That just makes a manual process more burdensome. The fix is to make the blend plan a live view of the stockpile’s current state, rather than a snapshot taken at signoff. Every new addition updates the modelled profile. The blend plan shows the current best estimate of what loading against this stockpile would produce, continuously up to the start of reclamation.
This is a data integration exercise, not a new piece of Software. The analysis data is being produced anyway. The stockpile model is being maintained anyway. The work is connecting the two, so the blend plan is not a file but a queryable view.
The second rupture is that physical stockpile movements, tonnes in, tonnes out, what was added from which face, where in the stockpile it was placed, are recorded in the mine operations system, not in the logistics system. The weekly shipping programme is built against a stockpile summary that was produced on request, usually late on a Friday afternoon, and saved as a snapshot.
The consequence is that decisions made between Friday’s summary and the following week’s load are invisible to the logistics team until they become visible as a variance, usually at the pre-load sample. A shift change decision to divert feed to cover a maintenance task on another belt is an operationally sensible call. It is not a call that the logistics team should need to review. But the fact of the call needs to reach the blend plan calculation, or the blend calibration becomes wrong.
In a connected operation, every stockpile movement above a meaningful threshold (e.g. 1,000 tonnes) automatically updates the logistics-side view of the stockpile’s profile, with the analysis value inferred from the face currently being mined or measured from the grade-control sample for that tonnage. The mine team does not change its behaviour. The logistics team stops being blind to the changes.
The typical test for whether this is working: when a logistics coordinator, on a Monday, asks what has changed in this stockpile since the blend plan was signed off, can they answer the question themselves in 90 seconds without emailing anyone? If yes, failure mode two is closed. If no, the operation is running a structural gap.
When the operations team recognises that the stockpile has changed, the normal response is to recalculate the blend manually. Somebody opens the blend calculation workbook, updates two or three cells, and the new ratio gets circulated by email. In most operations, this happens once or twice between signoff and load. It looks like adequate control.
It is not, for two reasons. First, the manual recalculation uses only the subset of data the person doing the recalculation is aware of, which, by definition, is a subset because the data landscape is not centralised. The recalculated ratio is better than not recalculating, but it is still calibrated to an incomplete picture of the stockpile.
Second, the manual recalculation is rarely logged against the original plan. There is now a tension between what was signed off on at the programme review and what is actually being executed at the berth. The variance is known to the people who did the recalculation and unknown to everyone else, including the quality team, who will sign the certificate, and the commercial team, who will own the customer conversation if something goes wrong.
The structural fix is to make the recalculation automatic and continuous. As the stockpile’s modelled profile updates, the required reclaim ratio updates. The only human decision required is whether to accept the system’s recalibrated ratio or override it with a different loading strategy; either way, the decision is logged, timestamped, and visible to downstream teams.
The quality team, in almost every operation, knows before anyone else when a blend plan is at risk. They see the grade control results as they come in. They notice the daily drift. They raise the concern, sometimes in a meeting, sometimes by walking over to the logistics desk, sometimes in an email that gets read three hours later.
The question is not whether the quality team is doing its job. They are. The question is whether the mechanism that carries their concern to the decision-maker is reliable. In a spreadsheet-and-email operation, it is not. A concern raised by the quality superintendent at 4 pm on a Thursday reaches the logistics manager on Friday morning. A concern raised while the logistics manager is on leave may not reach anyone in time.
Building the communication channel as a system capability rather than a human habit is a different exercise from building the analytics. It requires that deviations from the current blend plan generate a notification addressed to a role, not an individual, be escalated if unacknowledged, and be logged against the cargo to which they relate. The quality team no longer has to chase the logistics team. The logistics team no longer has to remember to check their email during a busy reclaim.
This is often the single highest-ROI change an operation can make on the blend-plan side, because the information is already being produced. The failure is entirely in the carriage. Fixing the carriage is not technically hard. It does require operational commitment because the new pattern replaces an informal norm with a structured one, and people who have worked well under the informal norm will occasionally miss it.
A reblend occurs when the operations team realises mid-load that the cargo is trending off-spec and takes corrective action: slowing the reclaim from one stockpile, opening the reclaim from a second stockpile, and mixing in material from a buffer. These are proper operational decisions, and a well-run operation makes them frequently.
The failure is that the reblend is almost never verified until the final surveyor’s result comes back, which is too late. The loading continues under the reblended strategy, the certificate is issued, and the vessel departs. If the reblend worked, nobody would hear about it. If the reblend did not quite work, the off-spec event happens at discharge.
The fix here is a measured one. Every reblend action, every stockpile switch, every feed rate adjustment, every buffer addition, is logged against the cargo with the expected quality impact. The pre-load and in-load samples are compared to the expected impact within the sampling cadence, not at the end. Any meaningful divergence triggers a pause-and-review, not a continue-and-hope.
In Practise, this means the surveyor’s preliminary results are being read in near real time against a running model of what the cargo should be. The sampling cadence does not change. The interpretation of the samples does.
Each of the five failure modes has a specific fix. But the five fixes share a common shape. All of them replace a manual reconciliation, a person checking, a person recalculating, a person remembering, with a system reconciliation that happens whether or not anyone is paying attention.
This is the single most important shift to name. A blend plan on a spreadsheet is an artefact of human vigilance. Human vigilance is a scarce, expensive, unreliable resource. The people on your team are good at their jobs; they are not, and should not be, expected to notice every stockpile addition, every feed diversion, every slow drift. A blend plan connected to a live stockpile model, a live logistics view, and a live quality feed is an artefact of system reconciliation. System reconciliation is cheap, consistent, and requires no leave.
The cultural work here is reframing the logistics team’s role from operator of the blend calculation to overseer of the blend system. The team does not stop using their expertise. They stop being the mechanism by which the reconciliation happens, and start being the people who intervene when the system flags that something needs human judgment.
This is a genuine upgrade for the people involved. Most logistics coordinators who have worked on a well-integrated platform for more than a few cycles will not go back. The work is more interesting, the decisions are clearer, and the anxiety is lower. The operation loses nothing and gains a structural resilience it did not have before.
Assuming you recognise one or more of the five failure modes in your operation, here is the practical sequence for closing them without undertaking a multi-year platform project.
The stockpile management system almost certainly has a place for analysing data. The LIMS almost certainly has an export. In many operations, the two are not connected because historically they did not need to be. Connecting them is a few weeks of integration work and produces the first visible capability: a stockpile view that updates as lab results come in, rather than as a human transcribes them.
Once the stockpile model reflects current analysis data, the logistics-side view of the shipping programme needs to pull that data into the blend calculation. The blend plan stops being a static workbook and becomes a live recalculation that uses the latest stockpile state. The interface the logistics team uses does not have to change significantly, and the numbers just start being right.
With live blend calculation in place, the next step is thresholding. The system knows the contract tolerance for the next-nominated cargo. It knows the current expected blend output. When the two drift within a configurable collision band, a notification is sent to the logistics and quality roles responsible for the cargo. The notification carries the data and the suggested corrective actions; it does not make the decision for anyone.
This is the stage at which the operation begins to catch problems before loading, rather than during or after loading. It is also the stage at which the business case pays back, typically within one or two prevented events.
At this point, the operation is preventing large-variance events and managing small-variance events within load cycles. The final stage is feeding the customer’s discharge sample back into the model so that the next cargo’s calibration starts from verified ground truth. Patterns become visible that were not visible before, not individually, but in aggregate, and the upstream operations team gets a closed feedback loop they have probably never had before.
A Hunter Valley thermal coal exporter running a five-stockpile reclaim system had historically averaged three off-spec events per year, two of which were attributable to blend-plan drift between signoff and load. The blend plan was produced on a Wednesday from Tuesday’s stockpile summary; the vessel typically loaded on the following Monday or Tuesday.
After stage one of the reconciliation work, the stockpile summaries became live views. After stage two, the blend plan became a live calculation. The first tangible catch came within the first month: a Sunday-evening reblend had been executed without updating the ratio, and the first load sample on Monday morning came back at the edge of tolerance. The system flagged it; loading was paused for 6 hours while the reclaim sequence was adjusted, and the cargo was loaded on spec.
The operational cost of the six-hour pause was modest, including some vessel demurrage and some berth idle time. The cost of the off-spec cargo it prevented would have been a significant rebate on a 75,000-tonne shipment, plus cascade effects on subsequent loads. The economics paid back on that single event.
Subsequent events over the following year tracked a pattern most operations see after they close the blend-plan thread. The number of near-miss catches increases in the first six months as the system surfaces previously invisible drifts. The number of actual off-spec events drops sharply. After eighteen months, the near-miss rate stabilises as the upstream operational habits, feed diversions, reblend discipline, and stockpile rotation adjust to the new visibility.
There are three common objections to closing the blend plan thread, and all of them are reasonable enough to address directly.
The blend plan is a commercial document, not an operational one: This is partly true. The blend plan is the contractual commitment to produce a cargo of a particular quality. What the playbook proposes is not to replace that commitment; it is to ensure the commitment is continuously verified against the reality of the stockpile, so it is kept.
Our blend calculations are proprietary, and we do not want them in a third-party system: Legitimate. The blend calculation itself can remain in whatever system the operation trusts. The live reconciliation does not need to know how the blend ratio is calculated; it needs to know the current expected output of the stockpile and the contract tolerance. Most mature implementations keep the proprietary math in the existing system and integrate with the reconciliation layer through standard interfaces.
Our operators would lose their feel for the stockpile: This is the most interesting objection because it is partly true and partly not. A senior reclaim operator who has worked the same stockpile for ten years has pattern recognition that no system will match. The goal is not to replace that pattern recognition. It is to make sure that when the senior operator is on leave, or has moved on to another role, the operation is not suddenly blind. The system captures what the senior operator already knows and makes it available to the next person in the seat.
The practical output of this work is a blend plan that stays true to the stockpile it was built against. The cultural output is more interesting and, for most operations, more valuable.
The first visible change is the shape of the weekly programme meeting. When the blend plan is a live artefact rather than a snapshot, the meeting starts with a reconciliation argument over whether the stockpile number is right, and ends with a decision conversation. The numbers are shared, agreed, and current. The twenty minutes reclaimed from reconciliation get spent on the questions that actually matter: which cargo is at highest residual risk this week, which stakeholder needs to be notified of what, which commercial trade-off is worth making. Most logistics managers who go through this shift describe it as the single biggest change in how their week feels.
The second change is in how new people come up to speed. In a spreadsheet-and-email operation, a new blend planner learns the stockpiles by osmosis, sitting with the incumbent, watching how decisions get made, slowly absorbing which subtleties matter. It is a craft apprenticeship that works, but it takes months and creates a long-tail dependency on the incumbent. In an operation where the stockpile state is visible as a living model, a new planner can learn about the stockpiles by observing them. The craft knowledge still matters, but it sits atop a shared foundation rather than substituting for it. Onboarding time drops.
The third change is in how commercial and operational teams talk to each other. Today, when the commercial team asks whether we can accept this spot cargo opportunity for the stockpile we have on hand, the operational answer is usually a careful hedge: we think so; let us check; we will get back to you by tomorrow. With a live blend model, the answer is specific: at the current stockpile state, this is the blend that would meet the spec; this is the headroom we have against the other commitments; and here is the probability of meeting tolerance given the current feed variability. Commercial decisions get made faster and with more confidence. Spot opportunities that previously got declined because the operational answer took too long, and start to get taken.
The fourth change is harder to name, but the one most logistics managers mention spontaneously once it has happened. The low-grade anxiety that sits around every sign-off, every load, every discharge sample, starts to lift. Not because the operation has become easier, but because the information it depends on has become trustworthy. You stop wondering whether the number on the screen is up to date. You stop double-checking figures you should not have to double-check. The work begins to feel more like management and less like firefighting.
It is worth separating two concepts that often get conflated in vendor conversations. Blend optimisation is the mathematical exercise of finding the best reclaim ratio across a set of stockpiles to meet a contract spec while minimising value loss. It is a well-defined optimisation problem, solved by linear programming or more sophisticated variants depending on the commodity and the complexity of the constraint set. Good blend optimisation Software has been available for decades.
Blend reconciliation is the operational discipline of ensuring the optimised blend calculated remains accurate over the hours and days between calculation and execution, and that deviations are flagged and addressed in time. Reconciliation is not a mathematical problem. It is an information-flow problem. It is what this playbook is about.
The distinction matters because many operations have invested in blend optimisation and then concluded that the investment did not pay back. Often, the problem was not the optimisation. The problem was the reconciliation. The optimised blend was correct when it was calculated and drifted before it was executed, for all the reasons this article has walked through. Adding reconciliation to existing optimisation is often the smallest, highest-impact investment an operation can make in its quality performance.
If your operation is just beginning to think about blend management, the sequence is the reverse of what intuition suggests. Start with reconciliation, get the information flow right. Optimisation can come later. A simple blend ratio that stays true to the stockpile beats a sophisticated optimised ratio that is stale by the time it is executed.
If you are considering whether blend-plan drift is a real problem in your operation, the most useful, cheapest exercise is a 90-minute whiteboard session. Take the last four cargoes you have loaded. For each one, pull the signed-off blend plan, the final surveyor’s result, and crucially, the stockpile additions and reclamation movements that occurred between the two.
Plot the assumed stockpile profile at signoff against the actual stockpile profile at reclaim start. Write down how you knew, or did not know, at the time, that the two had diverged. Write down how you would have known earlier, and who would have needed to act.
The pattern almost always jumps off the page. In operations with stable feeds and disciplined stockpile rotation, the divergence is modest, and the current manual vigilance is probably enough. In operations with variable feeds, satellite pits, or shift-level decision authority, the divergence is material, sometimes material enough to have been the cause of the last off-spec event the operation is still trying to explain.
Either way, the 90 minutes are well spent. It converts an abstract concern about blend-plan drift into a concrete artefact about this operation, with these stockpiles, and these recent cargoes. That artefact is the best possible input for a decision about whether the investment in automated reconciliation makes sense and, if it does, where to start.
The blend plan is one strand of a larger quality thread. For the end-to-end picture, our piece on building a continuous quality thread from mine to vessel shows how blend-plan discipline fits into the wider reconciliation between pit, stockpile, port and customer. For the commercial framing of quality variance, see how to stop off-spec commodities from reaching the vessel. For the organisational work, the shift from manual vigilance to system reconciliation, our piece on moving operations from reactive to proactive is the natural follow-on. External references on the underlying stockpile mathematics can be found in the long-running body of SME and CIM blending research and in AusIMM’s publications on grade control and stockpile management.
The blend plan does not have to fail. It fails today because the thread from stockpile to plan to load to ship is carried by human vigilance across multiple systems that were never designed to talk to one another. Rebuild the thread, one stage at a time, and the failures stop looking like bad luck and start looking like the system doing its job.
Nick Ogle has over 30 years of experience in Enterprise IT, spanning engineering, sales, and marketing roles across Australia, the USA, and APJ for various IT vendors.
Nick is passionate about entrepreneurship and Software innovation that drives positive change. Currently, he is the Sales & Marketing Manager at SCIAR Systems, a Newcastle-based SAAS startup, where he is helping commercialise their groundbreaking Bulk Commodity Logistics solutions.
For more information on Nick and to find articles that have been written on the IT sector in the past, feel free to look at his LinkedIn profile or browse some of the additional articles Nick has written for SCIAR.