Grade control has always been one of the most consequential activities in open-pit mining. Get it right and you’re sending the correct material to the right destination—ore to the plant, waste to the dump. Get it wrong and you’re either diluting feed grade with waste rock or, worse, sending payable ore to the waste dump where it’s gone forever.

Traditional grade control relies on blast hole sampling, assay results, and geological interpretation. It works, but it’s slow. Assay turnaround times of 12-24 hours mean that by the time you have results, the excavator has often already dug through the bench. Geologists make dig-line decisions based on incomplete information, which means grade control boundaries are estimates, not measurements.

AI is changing the speed and accuracy of that process. And in 2026, the results from Australian operations are strong enough to pay attention to.

Where the Traditional Approach Falls Short

In a typical open-pit operation, blast hole sampling captures grade data at intervals of 5-10 metres across a bench. That’s a sparse dataset for drawing dig-line boundaries, especially in geologically complex deposits where grade varies significantly over short distances.

The time lag compounds the problem. If assay results take 18 hours and the mine is running three shifts, the bench may have been partially excavated before grade control results are available. Operators dig based on planned boundaries, and corrections come after the fact—if they come at all.

Most open-pit operations accept 5-15% ore loss and 10-20% dilution as normal. On a gold mine producing 200,000 ounces a year, a 5% improvement in ore recovery is significant money.

What AI Grade Control Looks Like in Practice

The new generation of AI grade control systems integrate multiple data sources that already exist on most mine sites. They pull together blast hole assays, geological models, face mapping data, and MWD (measure while drilling) data from drill rigs.

Machine learning models find patterns across these datasets that human geologists might not spot. MWD data—drill penetration rate, torque, vibration—correlates with rock properties and mineralisation. Train a model on thousands of blast holes where you have both MWD data and assay results, and the system can predict grade between sample points with surprising accuracy.

An AI consultancy helped design a grade estimation system for a Western Australian gold operation that combined MWD parameters with geological domain modelling. The model’s predictions matched assay results within 8% accuracy on average, and it delivered those predictions in real time as blast holes were drilled—hours before assay results were available.

The practical impact is that dig-line boundaries become more accurate and available sooner. Excavator operators get grade boundaries overlaid on their GPS displays before they start digging, based on denser data than traditional methods provide.

Real Results From Australian Sites

A mid-tier gold producer in the Eastern Goldfields implemented an AI grade control system in mid-2025 and tracked results for eight months. Their dilution dropped from 14% to 9%, and ore loss decreased from 8% to 4%. That translated to roughly 7,000 additional recovered ounces over the period.

In iron ore, the value proposition is different but equally compelling. BHP’s South Flank operation has invested heavily in automated grade control systems that use sensor data and machine learning to optimise product blending. The goal isn’t just getting ore to the plant—it’s ensuring the right blend of lump and fines hits target specifications for shipping.

Queensland coal operations are using similar approaches. Coal quality varies substantially across a deposit, and AI systems that predict coal quality from drilling data are helping mines optimise dig sequences and reduce off-specification production.

Data Quality Matters More Than the Algorithm

I’ve seen operations bolt an AI system onto poor-quality data and wonder why it doesn’t work. The machine learning model is the easy part. The hard part is ensuring consistent, accurate data collection across drilling, sampling, assaying, and geological logging.

Operations that have invested in data quality management before implementing AI grade control have consistently seen better results than those that tried to fix data quality after deployment.

What’s Next

As sensor technology improves—particularly hyperspectral imaging and LIBS (laser-induced breakdown spectroscopy) for real-time elemental analysis—the data feeding AI grade control models will become richer and more immediate. Some companies are trialling bucket-mounted sensors that measure grade as material is excavated, creating a feedback loop that adjusts dig boundaries in real time.

We’re not at fully autonomous grade control yet. Geologists still need to validate model outputs. But the 5-12% grade improvement numbers I’m seeing aren’t theoretical. They’re real, measurable, and the operations achieving them aren’t going back to the old way of doing things.