# BHP’s Digital Push: A Numbers Guy Reads the Fine Print
BHP’s digital officer Mikko Tepponen recently laid out how the miner is deploying AI, automation, and data-led decision-making across its operations, claiming a run-rate of almost US$500 million in incremental EBITDA by the end of FY2026, with a target of US$650 million by end of FY2027.
Here’s what this really means: for a company that churns out roughly US$25-30 billion in EBITDA depending on the iron ore cycle, half a billion in tech-driven incremental EBITDA is about 2% of the business. Real money, yes. A transformation, no. I’ve watched enough “once-in-a-generation opportunity” speeches from the dot-com era to know that when executives reach for that phrase, the actual numbers rarely match the rhetoric. To Tepponen’s credit, he’s quantifying his claims — US$500 million is a verifiable figure, not a vibe. That puts him ahead of most AI evangelists in heavy industry, which is a low bar, but a bar nonetheless.
## The Strategy: Sensible, Unsexy, Mostly Credible
BHP formally launched its digital strategy at the end of FY2025 with four stated goals: reduce high-potential safety incidents, embed data and AI into daily operational decisions to cut unit costs, build and scale solutions with frontline buy-in, and shift culture toward pace and adaptability.
The approach is pragmatic — develop solutions with the operating assets rather than imposing them from head office, let the people who use the tech own it, then scale what works globally. In mining, where a bad rollout at an isolated Pilbara site can idle a crusher for weeks, this bottom-up model beats the top-down consultant-driven programs that routinely burn nine figures at other industrials. This isn’t the first time we’ve seen this circus — big miners have been promising “the digital mine” since sensors got cheap in the 2010s. Rio tried it. Anglo tried it. The ones who succeeded did it incrementally, at the pit face, not in a boardroom slide deck.
## Where the Value Actually Shows Up
The concrete examples are the credible part, because they’re boring:
– AI-assisted resource characterisation — better orebody knowledge, the cheapest margin improvement in mining.
– Autonomous haulage trucks running tighter cycles.
– Better feedstock consistency in crushing and screening.
– Improved copper extraction recoveries.
– Rail optimisation to cut queuing on the iron ore export line.
– Predictive maintenance across the fleet.
None of this is speculative. Autonomous haulage has been paying off at Rio and Fortescue for a decade. Predictive maintenance is industrial AI’s one genuinely proven use case — machines tell you they’re failing before they fail, you fix them on your schedule instead of theirs. If BHP is only now banking savings here, it was leaving money on the table for years, which tells you something about how slowly the big miners move even when the technology is mature.
The oversized rock detection system is a good example of the right scale of ambition. Cameras on diggers in the Pilbara feed images to models that flag oversize material before it gets loaded. On conveyors, the system detects rocks over 200mm and stops the belt automatically. Result: fewer crusher damage events, better throughput, less downtime, and fewer humans walking up to moving ore streams to chuck rocks off — which is exactly the kind of task that gets people hurt. If you want to know where AI actually earns its keep in heavy industry, it’s here: dull, repetitive, vision-based inspection tasks where a camera is cheaper and safer than a person. Not chatbots. Not copilots. Cameras watching rocks.
The Port Hedland stacker and reclaimer example follows the same logic. These machines used to reposition sequentially — one move finishes, the next begins. Smart relocation now lets some movements overlap, with edge detection preventing collisions. Less idle time, more throughput. Simple operational physics, no magic involved.
## The “Pit to Port” Pitch — Where I Get Skeptical
Tepponen’s next act is integration: moving from “single point” applications to an integrated, pit-to-port optimisation of the whole supply chain, dynamically, to liberate productive hours and lower unit costs.
Here’s the skeptical take. Integration is where every industrial digital program goes to die. Individual point solutions work because they have clear owners, clear metrics, and failure modes contained to one machine or process. End-to-end optimisation requires clean data across every node of the chain — mines, crushers, trains, ports — and mining data is notoriously messy, siloed, and inconsistent between sites. The integrators and systems people who promise this have rarely spent a night shift at an operating port. When the model says “hold the train for nine minutes to optimise the whole system,” and the train controller disagrees because the model doesn’t know about the stalled reclaimer, you get organizational friction that no slide deck anticipated.
That said, if anyone can pull it off, it’s an iron ore business — BHP’s Pilbara supply chain is a relatively linear, high-volume, single-commodity pipeline, far more tractable than a polymetallic underground complex. BHP has run integrated planning across the WAIO for years. So this isn’t fantasy. But the US$650 million target presumably embeds assumptions about integration savings materializing on schedule, and schedules in mining integration projects slip. Assume the out-year number is optimistic by a meaningful margin and you’ll be closer to reality.
## What Investors Should Actually Watch
Forget the narrative. Track three things. First, does the incremental EBITDA run-rate actually hit the stated targets in FY2027 filings — or does the definition quietly change? Companies under pressure to justify capex have been known to reclassify operating improvements as “technology-driven.” Second, watch unit costs. The entire point of this program is lower cost per tonne. If BHP’s WAIO unit costs don’t trend down over the next few years, the digital program is theater. Third, safety statistics — high-potential incident rates are a leading indicator of whether the technology is doing real work or just generating press releases.
One thing worth saying plainly: BHP is not overpromising here, at least not by industry standards. Half a billion dollars is modest relative to the hype cycle around AI generally, and the stated use cases are proven technologies being scaled, not speculative bets. The irony is that this sober approach is exactly why it might actually deliver. The companies claiming AI will “transform everything” rarely transform anything. The ones quietly cutting crusher downtime and queuing times at the port are the ones banking cash.
Mining is a capital-intensive, cyclical, commodity-price-driven business. No algorithm changes that. But shaving costs per tonne in the middle of the cost curve, safely, compounds over decades. If BHP executes even 70% of what Tepponen is describing, shareholders will be fine with it — provided management doesn’t treat the tech savings as an excuse to get sloppy on capital discipline elsewhere. Watch the numbers, ignore the language, and remember: in this business, the orebody and the cost curve decide your fate. Technology just decides how far from the bottom quartile you sit.
