The pitch from data platform vendors into Australian mining hasn’t changed much in five years. The reality on site has changed a lot. I’ve spent the last couple of months talking to operations people at three iron ore operations and one gold operation about where data and AI tooling is actually moving the needle in 2026.

The headline finding: the value is concentrated in three places, and most of it isn’t where the vendors are pitching.

Where the money is actually being made

Predictive maintenance on haul truck fleets. This was the original promise, and it’s the place where it’s most cleanly working. Operations running modern condition monitoring on Cat 793s and Komatsu 980Es are seeing 8-15% improvements in unplanned downtime year-over-year, with the gains compounding as the historical data base deepens.

The catch: it took six to eight years of clean data capture to get there. Operations that started the program in 2018-2019 are reaping the benefits in 2026. Operations starting now are looking at a 2032-2033 payback on the same investment. That’s not a reason not to start — but it’s a reason to be honest about the timeline.

Ore reconciliation and grade prediction. The gap between block model prediction and run-of-mine reality has been a known issue forever. The newer geological modelling tools — combining traditional kriging with machine learning on drill core and blast hole assay data — are starting to tighten that gap meaningfully. I’ve seen operations cut their reconciliation variance from 12-15% to 6-8%. That translates directly into better stockpile blending and better customer outcomes.

Energy management. Power is now the second-biggest operating cost line at many Australian operations. Sites running real-time energy optimisation on processing plants — adjusting mill draw, recirculating load, and reagent dosing dynamically — are pulling 4-7% off the energy bill. At industrial scale, that’s serious money.

Where vendors are pitching but value is thin

“Autonomous everything” platforms. The autonomous haulage story at Pilbara scale is real. The story for mid-tier operations is more complicated. The economics of full autonomous fleet conversion only work at scale, on long campaign life, and on well-defined pit geometry. If you’ve got 12 trucks running a 40-50% life-of-mine and complex pit walls, the business case is much thinner than the vendor deck suggests.

Real-time mine planning AI. The pitch sounds great: AI re-optimises your mine plan continuously based on actual material movements. The reality: most operations don’t have the data quality at the input layer for that to work. Garbage in, sophisticated garbage out. Get your fleet management data clean first, then think about layer two.

ESG and sustainability dashboards. A lot of vendor money is going into pitching these to operations that already have the underlying data. If you’ve already got compliance reporting working through your historian and SAP, you don’t need an additional $400K platform to put a dashboard on top of it. You need someone who can build the right reports on what you already have. For the integration work itself, an Australian AI company or specialist data team can typically deliver this faster than a full platform implementation.

The implementation lessons that keep recurring

A few patterns I see at every site I visit.

First, the people who own the data on site are not the people getting trained on the tools. The control room operator, the planning engineer, the geologist — these are the people who should be in the cockpit. Often the platform is bought at corporate, configured by IT, and operations get a half-day training session and a help desk number.

Second, the integration to legacy systems is always underestimated. Modern mining sites are running historians from the 1990s, fleet management from the 2010s, SAP since forever, and a few cloud-native point solutions stitched on top. Any new platform that doesn’t get the integration right ends up being another data island.

Third, the value compounds with data quality, not with platform features. The single highest-leverage investment most operations could make right now isn’t a new platform — it’s six months of work cleaning up the data they already have.

What I’d be doing if I were running ops

If you’ve got tight discretionary budget for tech in FY27, here’s the order I’d put things in.

Spend the first wave on data quality. Asset register hygiene, sensor calibration, integration cleanup. Boring work, big returns.

Spend the second wave on predictive maintenance for whichever asset class is your biggest unplanned downtime contributor. For most iron ore operations that’s still haul trucks. For underground gold, it’s often the mill or the hoist.

Spend the third wave on whichever value driver is most exposed to operational performance — for some that’s energy, for some it’s ore reconciliation, for some it’s throughput. Pick one. Don’t try to do all three at once.

And ignore the autonomous-everything pitches unless you’ve already done the first three. The economics only work when the foundations are in place.