Most of a mining megaproject's cost overrun is locked in before the first shovel moves, hidden in an inaccurate estimate, not what happens in the field. AI can catch a bad estimate before the board votes.
A board sanctions a new mine at two billion dollars. Four years later, the mine opens, a billion-and-a-half over budget and eighteen months late. Everyone blames execution.
“It was that bad stretch of weather.”
“We couldn’t source the labor needed to meet our dates.”
“A subcontractor low-balled to win the deal and then clawed profit back with a litany of change orders.”
But these are just excuses. Most of the overrun was baked into the project before the first shovel hit the ground.
Overruns are everywhere
Most large mining projects miss their numbers. For builds worth a billion dollars or more, McKinsey puts the average cost overrun at 79% above the sanctioned budget and the schedule at 52% past plan. A separate review of 192 mining megaprojects conducted by EY found that 64% came in over budget, behind schedule, or both.
Here is how AI usually shows up on a capital project. After sanction, someone bolts a real-time execution dashboard onto the build, and it starts reporting cost and schedule from the moment dirt moves. It flags that you are over budget while the concrete is already setting.
The decisions that pre-ordained the outcome were set years before. The dashboard is a rear-view mirror. It shows you the damage from a turn you took a long time ago. No execution tool rescues a project that was mis-scoped before anyone broke ground.
Get upstream of the sanction decision
Think about where a capital project's value gets set. Your ability to influence cost and scope peaks at the front end, during concept and feasibility, then collapses the moment the board sanctions. Spending runs the other way, almost nothing early and a flood once execution starts.
So, the cheapest place to change the outcome is before commitment. That’s where AI can do the most good. Three moves matter there.
1. Stress-test the estimate before you sanction
The most valuable thing AI can do on a capital project is tell you whether the numbers are believable. Machine-learning models trained on hundreds of past builds can compare your estimate, scope, location, and assumptions against what happened on similar projects. They can flag the optimism bias behind that 79% overrun. Working from front-end data alone, ensemble models have predicted final cost and schedule with mean errors under 10%. That hands your estimators a comparison set drawn from projects that have already been completed.
2. Fix the design while changes are still cheap
AI can build a digital twin of the plant at feasibility, while the design still lives on a screen. Using this model, you can test layout, material movement, tolerances, and access before anything gets poured. Re-routing a conveyor in the model costs nothing. Re-routing it in the field costs a fortune and sets the project back three months.
3. Only then do you watch execution in real time
Once dirt is moving, AI on earned value and field data earns its keep. It catches schedule slip and cost creep weeks before monthly reporting would, while there is still room to react. This is the dashboard every vendor tries to sell to you first. It is useful, but it belongs last, because no execution dashboard saves a project that was mis-scoped at sanction.
The overruns start in the estimate
You cannot optimize your way out of a bad sanction decision. Most of the overrun comes from flaws in the estimate. They live in numbers that were already wrong the day the board signed off. No amount of disciplined execution can rewrite that reality after the fact. When you point AI at the estimate and the design before commitment, you work the one lever that moves the outcome. After sanction, the best you can do is slow the bleeding.
All of these tools exist today. You just have to aim AI in the right place. Put the models on the estimate and the design while the number can still move, and you change how the project ends. Wait for the execution dashboard, and it will only tell you how far off you already are.
More information
If you’d like more information about these and other use cases for the mining industry, we would be glad to talk it through. A consultation is the easiest place to start. We will look at where a first project will drive quick P&L gains, so you leave the conversation with a concrete starting point. You can also visit rapidcanvas.ai and read verified customer reviews on G2.
About me
I write about practical, ROI-first AI in heavy industry. I work with RapidCanvas, which helps enterprises move past pilots to production AI they own and can build on. Connect with me on LinkedIn for more insights on mining and heavy industry.
Additionally, I run a hands-on 2-day AI workshop for operations leaders who want to put these ideas to work.
