Exploration Copilot
Most exploration budget is spent proving where the ore isn't.
Exploration Copilot fuses every layer you already hold into ranked targets — greenfield and brownfield — each carrying its evidence, its confidence by depth, and how reachable it is. The geologist decides where to drill. The model learns from every hole.
Evidence contributing
Confidence by depth
Toggle a layer or select a target — the model re-scores live.
Section A–A'
Cut the block open and the confidence changes with depth.
The same model, viewed in section. Drag the depth window to see how prospectivity — and how much you can trust it — falls away below the reach of your survey methods.
Spectral, geochemical and shallow geophysical evidence all reach this depth, and three targets sit within it.
The problem
The data has been collected for decades. The decision is still a judgement call.
Most exploration teams are not short of data. They are short of a defensible way to combine it — and of the time to look at every layer against every other layer across a whole licence block.
What it does
Three jobs, one model of the ground.
Greenfield and brownfield are different questions asked of the same fused subsurface picture.
Multispectral and hyperspectral satellite imagery, radar and InSAR, terrain models, regional magnetics and gravity, and public survey data — fused to find alteration signatures, structural intersections and lithological contacts over ground nobody has drilled.
- Alteration and mineral mapping from spectral response
- Structural framework from lineament and terrain analysis
- Regional ranking before a single crew mobilises
- Coverage of licence areas too large to walk
Legacy drillhole logs and assays, ground geophysics, structural and lithological mapping and mine records, re-interpreted together against a current geological model — looking for extensions along strike, down dip and beneath existing workings.
- Historic assay and logging data in one model
- Strike, dip and depth extensions of known mineralisation
- Faster payback than greenfield, on ground you hold
- Infrastructure already in place near most targets
Both routes produce the same artefact: a depth-resolved prospectivity model where every target carries a probability, the evidence layers that drove it, confidence banded by depth, and a terrain and access assessment.
- Ranked targets with evidence attribution per target
- Confidence stated by depth band, not as one number
- Access and terrain scoring alongside geological merit
- Every score traceable to the layers behind it
Inputs
It reads what you already own.
Public and licensed data where it helps, your own data where it matters. Licensed sources are only brought in with your prior permission.
Confidence
Depth is where most targeting models quietly stop being honest.
Geophysical methods lose resolution with depth, and every method loses it differently. A model that reports one confidence number for a whole target is hiding that. Exploration Copilot bands it.
Indicative bands. Actual depth sensitivity depends on which survey methods exist over your ground and at what line spacing — established during the data audit, not assumed.
Near-surface, spectral, geochemical and shallow geophysical evidence agree often enough to support a firm call. Below that, you are relying on fewer methods with coarser resolution, and the model should say so rather than extrapolate a confident-looking colour downward.
This matters commercially. A target ranked highly on shallow evidence and a target ranked highly on deep inference carry very different drilling risk, and treating them as equivalent is how exploration budgets get spent proving nothing.
Governance
The model ranks. The geologist decides. Every hole teaches it.
Targeting is probabilistic. Exploration Copilot is built to be argued with, not deferred to.
Evidence attribution
Every target score decomposes into the layers that produced it, with each layer's contribution weighted and visible.
Geologist override
Ranking can be overridden with a reason code. Overrides are recorded and become part of the training signal.
Drill-result feedback
Assay results from each hole are fed back so the model is re-scored against what the ground actually contained.
Data provenance
Public, licensed and proprietary sources are tracked separately. Licensed data is only used with your prior permission.
Questions we get asked
AI mineral exploration, answered.
What exploration managers, chief geologists and mining leadership ask before starting a targeting programme.
What is Exploration Copilot?
Exploration Copilot is AI decision-support for mineral exploration targeting. It fuses satellite imagery, airborne and ground geophysics, geochemistry, legacy drillhole data and structural mapping into a depth-resolved prospectivity model, then produces ranked drill targets — each with the evidence that drove its score, confidence banded by depth, and a terrain and accessibility assessment. It supports both greenfield exploration over undrilled ground and brownfield targeting around existing operations.
What is mineral prospectivity mapping?
Mineral prospectivity mapping is the process of combining multiple geoscientific datasets — geophysical, geochemical, remote sensing, structural and geological — to estimate the likelihood that mineralisation occurs at a given location. Traditionally it is done by weighting layers manually or by expert judgement. Machine-learning prospectivity mapping learns those weights from known occurrences and drill results instead, and can evaluate far more layer combinations across a licence than a manual workflow can.
How does satellite data help greenfield exploration?
Multispectral and hyperspectral satellite imagery detects surface mineralogy and alteration signatures associated with mineralising systems, radar and InSAR reveal structure and ground movement, and digital elevation models expose lineaments and terrain controls. Together they let a team rank a large licence area before mobilising crews — narrowing where to spend on ground geophysics and sampling rather than covering everything at the same intensity.
Why is brownfield exploration often the faster return?
Brownfield ground is already held, already has infrastructure, and usually carries decades of drillhole and assay data that was interpreted under an older geological model. Re-interpreting that history against a current model regularly identifies extensions along strike, down dip or beneath existing workings. The distance to first ore is shorter and the permitting and access burden is lower than opening new ground.
How deep can a prospectivity model resolve?
It depends entirely on which survey methods cover your ground and at what resolution — depth sensitivity is a property of the data, not of the model. Near-surface, spectral, geochemical and shallow geophysical evidence typically supports higher confidence; confidence decreases through intermediate depths and becomes indicative deeper, where fewer methods contribute and resolution is coarser. Exploration Copilot reports confidence in depth bands rather than as a single figure, and the achievable bands are established during the data audit.
Can AI replace our geologists?
No, and a model that claims to should be treated with suspicion. Exploration targeting is probabilistic and depends on geological reasoning the model does not hold. Exploration Copilot ranks and evidences options at a scale and consistency a team cannot match manually; the geologist interrogates the evidence, applies judgement the data cannot carry, and decides where to drill. Overrides are expected, recorded and used to improve the model.
What data do we need to start?
Whatever exists. A typical starting set is regional and any proprietary geophysics, geochemical sampling, available satellite coverage, historic drillhole logs and assays, and geological and structural mapping of the licence. Gaps are normal — the data audit at the start of a pilot establishes what is present, what is usable, and which additional public or licensed sources would materially improve the result before any are acquired.
How does this relate to the rest of SmeltIQ?
SmeltIQ builds decision layers across the metals chain. Exploration Copilot decides where to look for ore; Recipe Copilot and Furnace Copilot decide how it is melted and refined once it reaches the plant; Data Copilot reports across the whole operation. They share the same platform, governance model and audit approach, so a group operating both mines and plants runs one decision layer rather than several.
Rank your licence before you drill it.
Start with one block, one commodity and the data you already hold.