How to Use AI in a Mining Contracting Company
The question most mining contractors ask is whether AI has any business near a mine site. The better question is which part of the operation it belongs in first. For most contractors, that answer is the office, not the pit.
This post is for the owner who runs 2-5 crews and watches dailies pile up on an office manager's desk every Monday morning. It is also for the office manager who suspects she is about to be automated out of a job. The question is how to use AI in a mining contracting company without disrupting what works at the face, without handing over data that could embarrass you, and without buying something that does not pay for itself inside the first 90 days.
What follows is the honest answer: what AI can read, what it must never decide, and the one workflow worth starting with.
Worked example
About 20 hours a week is what I see going into data entry and daily reporting at a working contractor. On most jobs I have been on it is a superintendent, an engineer or the owner doing it, at night, because they are the only ones who know what the numbers mean. So cost that hour at what their hour is actually worth, somewhere between $100 and $150.
| Step | Calculation | Result |
|---|---|---|
| Hours a year | 20 hrs/week x 50 weeks | 1,000 hrs/year |
| Cost at $100/hr | 1,000 hrs x $100 | $100,000/year |
| Cost at $150/hr | 1,000 hrs x $150 | $150,000/year |
Answer: 1,000 hours a year, worth $100,000 to $150,000, goes into moving numbers that were already written down once. That is the exposure, not the saving. What actually comes back depends on your paperwork, which is why the first step counts it on your own 30 days of reports instead of quoting you a percentage.
How to Use AI in a Mining Contracting Company: What It Can Read
AI reads documents. That sentence is simple enough to dismiss and concrete enough to build on.
Daily reports from underground come in by crew. They carry metres advanced, tonnes hauled, headcount by shift, equipment hours, and cost codes. They also come in with errors: a transposed cost code, a date from the wrong week, a crew count that does not match the shift schedule. Right now your office manager catches those by reading each report against the last one and against the contract. AI does that same pass before she does, so she is looking at a flagged exception list instead of a stack of unread PDFs.
The document types a contractor handles are well suited to this kind of reading:
Daily production reports. The structure is consistent: shift, crew, activity, quantity, cost code. An AI reader configured to your format will catch a mismatched code or a missing headcount entry faster than a person scanning a handwritten form.
MSHA 7000-1 records. Every injury, illness, and days-lost entry carries a filing requirement under 30 CFR Part 50. A structured reader can cross-reference your incident log against what has been filed and flag gaps before an inspection catches them. Knowing your incidence rate at any point in a contract is also the starting position for any performance conversation with the mine. You can run that calculation at Big Dog Mining's incidence rate calculator.
Contract billing schedules. Rate sheets, unit prices, scope definitions. AI can read the contract and flag when a submitted quantity falls outside the agreed rate structure. It cannot resolve the dispute. It can tell you the dispute exists three days earlier than you would have found it on your own.
Progress reports for the mine. Most clients want a fixed format. AI can draft that report from your daily data. A person approves it before it goes out. That draft used to take 90 minutes to produce from scratch. With the underlying data already read and structured, it takes 15.
What AI does not do is understand context. It does not know that last week's low advance rate is because the heading hit a fault, not because the crew was short. It does not know that the mine's format changed because someone on the client side mentioned it in passing. It reads what is on the page. The person running the office knows what the page is not saying.
What AI Must Never Decide
The list of things AI should not decide in a mining contracting company is shorter than most contractors expect, but the items on it are non-negotiable.
Safety calls. Whether a ground condition is safe to advance. Whether a scaler result changes the shift plan. Whether a near-miss gets filed as a reportable. These require a person's judgment and a person's name on the decision.
Scope changes. When a client asks for work outside the contract, someone with authority has to say yes, confirm the rate, and get it in writing. AI can flag that a work order does not match the contract scope. It cannot negotiate what happens next.
Billing disputes. When the mine's quantity reconciliation disagrees with yours, a person has to own that conversation. AI can give you the paper trail. It does not have the relationship.
Anything that goes to the mine under your signature. Progress reports, variance explanations, incident notifications. AI can produce the draft. You send it. That distinction matters because the consequence of a wrong number in a client report falls on you, not on the tool that produced the first draft.
The line is not about capability. It is about accountability. Keep the accountability where it belongs, and the tool stays useful.
Your Data Is Already in the Cloud
The objection that surfaces first is data. Your rates are in those contracts. Your client's name appears on every daily. Why would you hand that to anyone?
You would not hand it over without conditions, and you do not have to. The structure that works for contractors starting this is simple: one contract, one month, nothing else. Redact whatever you want before anything is sent. The NDA is signed before anything is seen.
The more important point is this: your data is already in the cloud. Your email runs on Google or Microsoft servers. Your accounting lives in QuickBooks Online or Xero. The photos of paper dailies in the site WhatsApp group are on Meta's infrastructure. The question is not whether your data is in the cloud. It is whether anything can finally read it.
The training question deserves a direct answer. Yes, the AI learns from your data. That is what makes it useful for your operation specifically. It learns your cost codes, your contract structure, your crew naming conventions, your exceptions. That learning does not get pooled with any other contractor's operation. When a second contractor uses the same tool, they are not learning from your rates. Their system is learning from their data.
"Nothing trains on your data" is the standard vendor answer. The honest version is different: it trains on your data, and that is the point. A vendor who tells you otherwise is either wrong about how the product works or hoping you will not ask. A contractor who asks that question plainly and gets a clear answer knows something about who he is working with.
How to Use AI in a Mining Contracting Company: The First Step That Pays for Itself
The smallest starting point is one workflow, not one company. Pick the workflow that costs the most time and produces the fewest judgment calls. Daily report processing is usually that workflow.
Here is what 30 days looks like in practice.
You send one month of daily reports for one site. Nothing from a second contract, no rate sheets, no client contact information. The AI is configured to your format: your cost codes, your production units, your crew structure. The configuration step takes a few hours. It is not a project.
After 30 days you have a count. How many reports had missing data. How many had cost code mismatches. How many quantities were outside the expected range for that crew and shift length. You compare that against what was caught manually. The gap between those two numbers is the business case, and it is a number you measured, not one a vendor quoted you.
At that point the next step is visible. Connect the same reader to your invoice drafting process. Feed it the contract rate sheet and let it check each billing line before the invoice goes out. Start producing the mine's progress report from the same daily data instead of retyping the numbers into a second document. None of those steps require a new platform. They require one more workflow pointed at documents that already exist.
How to Use AI in a Mining Contracting Company Without Touching the Face
Your crews change nothing. This point needs to be said plainly because the objection comes up immediately: "my guys won't adopt new software." They do not have to. They submit their daily the same way they do now: paper form, WhatsApp photo, PDF, tablet app. The AI reads the document downstream in the office. The face never sees it.
The connectivity objection falls away for the same reason. Whether your site has signal underground is not relevant to how the daily report gets processed. The report is submitted at surface, and the processing happens in the office afterward.
There is no server to run, no database to administer, no IT department required to maintain it. If the contract ends, the system stops. There is no obligation past the work.
The format change question is worth addressing directly: what happens when the mine changes its report template and your daily format has to change with it? The AI is reconfigured to the new format. It is the same kind of work as setting it up the first time, and it is shorter because the cost codes and crew structure are already known.
The Person This Does Not Replace
The office manager question is the one nobody asks out loud in a meeting, but it is the one that decides whether the internal champion forwards your email or lets it sit.
If AI reads the dailies and drafts the progress reports, what does she do?
She stops doing the part of the job that has always felt like data entry: reading each report, pulling the numbers out, checking the codes, entering the result in a separate spreadsheet. She starts spending that time on the part that has always required judgment: deciding what an exception actually means, knowing that a headcount discrepancy on a specific crew probably reflects a quit that has not been reported yet, recognizing that the client's format request last week was actually a signal about what the mine's PM is watching.
A system that reads documents does not know any of that. The person who has run that office for three years does. The job does not go away. The character of it shifts toward the part that was always harder and more valuable.
Nobody I have talked to about this wants fewer people. They want the person they have to stop spending Sunday night on it, and to have the room to take on a second contract. That is the real economic argument, and it is the one worth making to the person whose support you need before anything starts.
Common questions
AI hallucinates. I cannot have it making up metres drilled.
The system does not generate numbers. It reads the numbers in your daily report and checks them against a structure you define. When something does not reconcile, it flags that row for a person to review. A person approves before anything moves downstream. The hallucination risk applies to generated text, not to extraction against a fixed schema.
Is this going to replace my office manager?
No. The data-entry layer changes: reading reports, pulling cost codes, finding missing entries. The judgment layer grows: deciding what an exception means, knowing what the client actually expects, managing the conversation when a quantity disputes. That second layer is not automatable, and it is where the job has always lived.
What is the smallest possible first step?
One contract, one month. Send 30 days of daily reports for one site, redacted however you want. The system learns your format: your cost codes, your production units, your crew structure. After 30 days you have a count of exceptions caught and a concrete business case. Nothing else in your operation has to change first.
My rates and my client's name are all over these documents. Why would I share that?
You start with what you are comfortable with: one contract, redacted where you choose, after an NDA is signed before anything is seen. Your rates stay yours. The same data concern applies to QuickBooks Online and your email provider, and you already use both. What changes is that something can finally read what is already there.
Do I have to replace Excel?
No. Most setups run alongside it. Excel holds the master data. The AI reads incoming documents, surfaces exceptions, and produces draft outputs. You still control the spreadsheet. Nothing in your current stack has to change before you start, and nothing is forced to change after.
Getting this off paper
Most contractors already know these numbers. The cost is in re-keying them into a report, then a timesheet, then an invoice. If that is where your week goes, book a twenty minute call and we will look at your actual paperwork.