This is worth being blunt about, because the industry is not. Remote sensing and machine learning rank ground by how much it resembles ground that was productive elsewhere. That is a prioritisation tool, and a good one. It is not a discovery.
What it changes is the budget. Conventional coverage spreads a fixed exploration spend evenly across a licence. Ranked ground concentrates the same spend on the fraction most likely to repay it and, as importantly, takes the rest out of the programme.
Nothing is drilled on a screen alone. Every ranked target is walked by a crew before it enters a drilling programme, and targets that do not survive that walk are reported as failures rather than quietly dropped.
Delivered as georeferenced layers you can load into your own software, plus a written interpretation. Not as screenshots.
Pipeline
Multispectral and radar imagery over the whole licence: every hectare, not a sample of it.
Our own machine-learning models map alteration and structure band by band, and score and rank every cell of the ground.
Crews walk the highest-ranked ground and sample it. Nothing is drilled on a screen alone.
Samples and drill holes build the model of the ground, and the model decides where the equipment goes.
We built this system ourselves: our own machine-learning models, and the large geological datasets behind them. It joins AI-powered satellite imagery analysis to our geological studies, to find the areas with the strongest exploration potential. The models read what the satellite sees (rock alteration, iron oxide, structure and terrain, in images taken before any digging) alongside what is already known on the ground (old workings, measured strikes), and give every target a confidence. That sharpens the targets and shows the best places to investigate next. A geologist checks every ranking before it reaches the field.
Field guide · example
Where to dig
The rule: if T1 cuts the structure, dig T2 and T3. If it does not, stop and review before any more trenching.
An example drawn for this page. It shows no real licence, place or result.
Our own AI for land evaluation. Send us a point or the outline of a licence: the system reads the satellite data over that ground together with geological knowledge, scores it against the known deposits of Tanzania, and recommends go or no-go, with the reasons behind it. A geologist reviews every answer before it reaches you.
The map is drawn for this form. The dot marks your point; it is not a result.
Stated here so that the first conversation can be about the work rather than about how we charge for it.
| Item | Detail |
|---|---|
| Input required | A licence boundary or a set of coordinates. Existing data helps but is not required |
| Typical turnaround | Two to four weeks for processing and ranking; verification is scheduled separately |
| Delivery format | GeoTIFF and shapefile layers, plus a written interpretation |
| Confidence | Stated per target, with the evidence listed. Targets are never presented without it |
| Verification | Field crew mobilised on request; findings reported whether or not they support the ranking |
| What it is not | It is not a resource estimate and it is not a substitute for drilling |
It is a ranking, not a prediction, and we report it as one. The useful measure is how much ground it removes from a programme: typically a licence reduces to a small fraction of its area for first-pass fieldwork.
No. A boundary is enough to start. Where previous sampling exists it is used to calibrate the ranking, which materially improves it.
Yes, over open ground, and it is one of the more sensible uses of the method: it is far cheaper to rank ground before applying for it than after.
It is written up as a negative result with the reasons. A method that never reports failures is not producing information, it is producing reassurance.
A licence boundary or a single point is enough to begin. We will come back with what the imagery supports, what it does not, and what a verification programme would cost.