Want 10× from your enterprise AI whilst slashing its cost? Stop upgrading the model. Instead, fix what it reads.
Most mining boards I sit across from have signed off an enterprise AI licence in the last twelve months. Very few can tell me what it is reading.
Over the past few weeks the same conversation has come up with a miner, a commodity trader, a private equity fund and a royalty company. Different businesses, one complaint: the AI doesn't deliver across functions and it's becoming extremely expensive.
What we have learnt building data cores on clients' own information:
The data is rarely missing. Whether it's in Dropbox, SharePoint, Snowflake, an AWS bucket, a CRM or a drill database kept in three formats and forty years of scanned reports. Multiple sources, no common structure. A model reads what it is handed, and it is handed fragments.
General AI cannot tell you what it did not find. Ask it for every hole drilled on a licence and it will answer confidently from whatever surfaced. Without a system of record you don't know what you don't have, and neither does the model.
Data structure changes everything. When the AI reads structured, normalised and categorised data rather than raw documents, a query can use up to 300× less compute and cost 8 to 20× less to answer. That is where my 10× statement comes from (it's actually often a lot more than 10x).
So forget the bigger, new shiny model. A bigger model just makes it more wrong and more expensive (a lot faster).
You do not need to replace the AI you already run. Claude, ChatGPT and Gemini all improve when they sit on one source of truth, with a mining ontology underneath and every figure traced to the page it came from. Pulse is the layer your AI runs on.
We built our own data core first. Across every part of the industry, Pulse now holds:
- 100 million+ pages processed
- 83 million+ drill intercepts
- Normalised, source-traced technical mining metrics across the mining life cycle
- 3.5 million pages of historical Cyrillic archive, every figure traced to its source page
The judgement call stays with your people. What changes is how much of the company they can see when they make it.
If your AI programme has stalled, look at what it is reading before you look at the model. Our Board Guide to AI in Mining sets out 22 questions for the board. Two worth asking this quarter: what is our AI reading, and how would we know if something was missing?
The guide is free: download The Board Guide to AI in Mining.
Less searching. More strategising.™
Where does your team's data infrastructure sit today?
Answer 10 questions. Get a private diagnostic on your AI readiness, in minutes.
Less Searching. More Strategising.™
See the platform running on real mining data, and what it would look like for your team.