Every underground shift produces four data streams that someone has to do something with before the month closes: development meters, tons hoisted, ground support installed, and ventilation readings. Each stream feeds a billing claim, a safety plan, or both. The same number gets written at the face, typed into Excel, and retyped again into the claim.
The question is not whether AI belongs in a mining operation. It is which steps in that chain a piece of software can handle reliably and which steps require a named, qualified person. Underground mining has a regulatory answer to the second half of that question. The first half is what this post is about.
Worked example
A contractor runs 2 development headings on a single contract. Each heading fires 1 round per shift on a 2-shift day, advancing 3.0m per round. The month has 22 working days. Each daily carries 5 fields that must appear identically in three documents: meters advanced, bolt count, mesh panels installed, delay code, and crew on shift.
| Step | Calculation | Result |
|---|---|---|
| Monthly development meters billed | 2 headings x 2 shifts x 22 days x 3.0m per round = 264m | 264m advanced and billable |
| Daily reports generated | 2 headings x 2 shifts x 22 days = 88 dailies | 88 individual daily reports to process |
| Field values originating at the face | 88 dailies x 5 fields per daily = 440 source values | 440 field values, each written once at the face |
| Total entries before billing closes | 440 source values x 3 documents (daily, ground support log, billing claim) = 1,320 individual entries | 1,320 typed entries per month |
| Error exposure at 1% transcription rate | 1,320 entries x 0.01 = 13 wrong entries | 13 wrong entries per month; at 2% that is 26 |
Answer: A 2-heading operation generates 1,320 individual field entries per month before billing closes. At a 1% transcription error rate, that is 13 wrong entries. At 2%, it is 26. Most are caught in reconciliation. The ones that are not cost you on the invoice or in the ground support log, and sometimes both.
What the daily actually contains
An underground development operation records four data streams every shift. Development meters measure how far the face moved: the round fires, the muck clears, the heading advances. Tons hoisted counts how much material left the heading or stope and reached surface. Ground support records what was installed: bolt count, pattern, mesh panels, any shotcrete or split sets. Ventilation readings capture airflow at survey stations and confirm the heading is receiving enough fresh air.
Each stream has a reason beyond bookkeeping. Meters are the billing unit on most development contracts. Tons are how hoisting and trucking are measured on a per-ton basis. Ground support records satisfy the ground control management plan the mine requires, and they appear in safety audits. Ventilation surveys demonstrate compliance with the mine ventilation plan and with regulatory requirements under 30 CFR Part 57.
All four streams originate in the same document: the daily report written at the face, by the driller or the shifter, usually on paper.
The recording chain and where it breaks
After the shift, the daily travels. On most operations it goes by photograph: a phone shot dropped into a WhatsApp group or uploaded to SharePoint. The office manager receives it, reads it, and types the numbers into a spreadsheet. That spreadsheet feeds the billing claim, which goes to the mine's clerk, who enters it into the mine's own system. The mine's surveyor reconciles the billing meters against their own survey at month end.
That is 4 handling steps and 3 separate typing events before the meter figure is confirmed on both sides. The ground support data takes a parallel path. It goes from the daily into the office spreadsheet, and also into the ground control management plan log, a separate document on the contractor's safety side. Those are 2 typed destinations for the same bolt count.
This is where the chain breaks. The bolt count on the daily is typed into the billing claim. It is also typed into the ground support log, separately, often by a different person. Those two typed values should be identical. When they are not, you have either an invoice dispute or a compliance gap.
The ventilation reading takes a third path. It lands in a ventilation survey register, which the mine's ventilation officer may also want a copy of. If your shifter records velocity and flow at each station, someone has to put those numbers somewhere that is not the daily.
The daily is written once and retyped three or four times into documents that are supposed to agree. They often do not.
Ground support: the record that serves two masters
Ground support is where billing and safety intersect most concretely. Every resin bolt installed is a cost item: contractor-supplied, contractor-installed, billed per unit. It is also evidence that the ground control management plan was followed. The mine's geotechnical team set a bolt pattern for that heading. Your crew installed the bolts. The daily is your record that it happened.
The SME Mining Engineering Handbook defines a safe system of work as one built on competent people, fit-for-purpose equipment, and effective systems and practices in a controlled work environment (SME Mining Engineering Handbook, 3rd edition, 15.1.9.1 Competent People, p. 1562). The ground support record is part of what that third element requires. It is not an administrative option.
A wrong bolt count on the daily is not just a billing error. It is also a gap in the safety record. If the daily says 18 bolts and the plan required 20, the same error lands in two places: the billing claim and the ground control log. One wrong number, two problems.
The contractor who can show the installed bolt count matched the plan, round by round, has a different answer when the mine's safety team asks. The contractor who cannot match the records has a problem on both sides of the conversation at once.
What an AI agent can actually read
The agent's job in this chain is narrow and specific. It reads the daily. It extracts the four streams: meters, tons, ground support, ventilation. It compares each extracted value against what the plan says should be there. It flags the exceptions.
For meters, the comparison is against the round plan: expected advance per round, expected hole count, expected meters per shift. A value far outside the expected range gets flagged before it goes into the billing claim.
For ground support, the comparison is against the GCMP pattern: expected bolt count per round, expected mesh coverage. A round that shows 16 bolts where 20 is standard gets flagged. The agent does not make a compliance decision. That is the shifter's job and the geotechnical engineer's job. The agent makes sure the number is not silently wrong.
For ventilation, the agent reads the survey station entries and compares them against the mine's ventilation plan minimums. Readings that fall below the station requirement are surfaced immediately, not at week-end when the log is compiled.
An autonomous mine is information-driven, from the detailed mine model to product levels, payloads, and the location of every asset (SME Mining Engineering Handbook, 3rd edition, 9.8.4.4.1 Efficiency Gains though Process Control, p. 819). The information an AI agent needs for these comparisons is the same information being typed into spreadsheets today. The agent reads the source document. The spreadsheet work becomes verification, not transcription.
Integrated site awareness, tracking every machine and mobile object in real time, offers a significant safety advantage by the same principle that makes air traffic control effective (SME Mining Engineering Handbook, 3rd edition, 9.8.4.1.1 Integrated Site Awareness, p. 818). A mine's continuous, around-the-clock operation makes that tracking just as critical. A shift-end read-in is not real-time, but it closes the gap from days to hours.
What requires a person
This is the part that matters most for a contractor managing safety compliance on someone else's site, and it is the part technology vendors most often skip.
The SME handbook is direct: to achieve safe production in a mine, competent people must be employed in senior management, in line management, and at the coal face (SME Mining Engineering Handbook, 3rd edition, 15.1.9.1 Competent People, p. 1562). Evidence of competence includes training, assessments, and formal qualifications. In most Australian jurisdictions, the mine manager must pass a written and oral examination. The record of that competence is personal. It cannot be assigned to a system.
The ground control management plan review is signed by a named person. The pre-shift inspection is signed by the shift supervisor. The near-miss report is investigated by a person who is accountable. An AI agent that extracts and reconciles numbers does not change any of that. The agent produces a reconciled, flagged data set. A competent person reviews it. That person approves it. That person signs it.
MSHA's regulations for underground mines require pre-shift examinations, ground control inspections, and ventilation records to be kept on site and available for inspection. The regulation does not specify who types the number into a spreadsheet. It specifies who is responsible for the condition the number describes. Those two things are not the same.
What the agent takes over is the transcription. What stays with the person is the approval. The safety record is still the person's record. The difference is that the person is approving a reconciled, flagged data set rather than trusting that the retype was right.
The MSHA benchmark your operation probably does not have
MSHA publishes its accident and injury data publicly. Every citation, every incident rate, every contractor's record is in there. Most small contractors do not know their own incident rate relative to similar operations, because nobody has pulled the data and done the arithmetic.
The MSHA incidence rate calculator on this site runs that comparison. Feed it your hours worked and your incident count and it returns your rate against the industry benchmark. Before building any safety management system, knowing where you stand is the starting point. The data to find out is already public and free.
The safety data question from the contractor's side is not how to build a system. It is whether you even know your number. The daily that feeds billing already contains the crew hours that go into that rate. The same document, read once, produces both the production figure and the safety input.
Where this leaves the four streams
Development meters: read from the daily, compared against the round plan, flagged if implausible, then passed to billing without a second retype.
Tons hoisted: read from the hoisting log, compared against the ticket weight if your site runs one, then to the billing claim.
Ground support: read from the daily, compared against the GCMP pattern, flagged if the count is below plan. One read populates both the billing claim and the ground support log.
Ventilation: read from the survey station entries, compared against the mine's ventilation plan minimums, flagged if a station falls below requirement.
The office manager reviews the flagged exceptions, resolves anything the agent could not read with confidence, and approves the clean set. The billing claim is assembled from the approved data. The ground support log is updated from the same data. The ventilation register is updated from the same data.
The daily was written once. Everything downstream of it is read, not retyped.
Run your own numbers
Rather than work it by hand every time, the msha incidence rate calculator does this arithmetic and shows its working. The rest of the set is on the mining calculators page.
Common questions
Can the same daily that drives billing also feed the safety record?
Yes, and that is the practical starting point. The daily already contains the crew, the equipment, the meters, and the bolt count. One read of that document can populate billing, the ground support log, the ventilation record, and training attendance in a single pass. The current approach reads it four times separately.
AI hallucinates. I cannot have it making up meters.
The agent reads numbers, it does not generate them. It extracts what the driller wrote, flags what it cannot read, and shows you the source image alongside the extracted value. A meter it cannot read with confidence goes into a queue for you to resolve, not into the billing claim. The output is a flagged exception list, not an invented number.
What does MSHA require a contractor to keep for ground support and ventilation?
Under 30 CFR Part 57, underground metal and nonmetal mines must conduct and record pre-shift examinations, ground control inspections, and ventilation surveys. Records must be kept on site and available for inspection. The current text is at ecfr.gov. An AI agent reading the daily does not change any of those obligations.
Does my office manager still have a job?
The retype work goes away. Every field that arrives from the daily as a readable value stops being typed a second and third time. What stays: judgment calls on missing or implausible data, the client relationship, and every approval a named individual is required to make. The job gets better, not smaller.
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 20-minute call and we will look at your actual paperwork.
