Every mining contractor running a development contract has the same daily office problem: the same set of production numbers gets entered into three different places by at least two different people before an invoice goes out. Most of the work in that chain requires no judgment. It requires time.
This post draws the line between what the paperwork chain actually looks like, what a document reader can take off the plate, and where the human judgment that keeps a mining operation safe and profitable is irreducible. The line matters because a vendor who blurs it is selling a promise.
The FAQ bank for this post has forty questions, twelve of which are about data security before a single question about what the system actually does. That ratio tells you the real objection. The data question gets a direct answer below.
Worked example
Two active development headings, reporting 6 days a week. Each daily passes through 3 office steps: a consistency check, transcription to the invoice spreadsheet, and reformatting for the mine's daily template. Month-end reconciliation adds time when the mine's countback differs from the reported total. Wage rate and entry time are illustrative inputs; substitute your own to size your operation.
| Step | Calculation | Result |
|---|---|---|
| Reports per month | 2 headings x 26 reporting days = 52 reports | 52 reports/month |
| Entry time per report | 3 entries x 15 min per entry = 45 min per report | 45 min/report |
| Monthly entry time | 52 reports x 45 min = 2,340 min = 39 hours | 39 hours/month |
| Month-end reconciliation | 1 disputed submission per week x 1 hr x 4 weeks = 4 hours | 4 hours/month |
| Total monthly office time | 39 + 4 = 43 hours | 43 hours/month |
| Annual cost | 43 hours/month x 12 months x $35/hr = $18,060 | $18,060/year |
Answer: At these inputs, the daily report chain consumes 43 hours of office time per month and costs $18,060 a year in admin wages.
What the Daily Report Chain Costs
A contractor running 2 active development headings submits daily production reports 6 days a week. Each report carries the data that drives the invoice: advance in meters, blast details, shotcrete consumption, any deviations from the round plan.
Before that report becomes an invoice line, it passes through the office at least three times. First pass: the office manager reads it, checks internal consistency, and flags anything that does not reconcile with the previous day's position. Second pass: she enters the advance meters and consumption figures into the invoice spreadsheet, where they are multiplied against contract unit rates. Third pass: she reformats the same advance and blast data into the mine's daily reporting template, which uses different field names and a different layout from the invoice sheet.
Three passes. The same numbers each time. None of them requires a decision. All of them require time.
At month end, the mine runs a survey countback. Their measured advance differs from the sum of reported dailies. It almost always does. She goes back through the month's reports to identify where the gap opened, then resolves the discrepancy with the mine's quantity surveyor.
This chain is what can be automated. Every figure already exists in the field submission. What happens next is reading and re-entry.
What the Reader Actually Does
A document reader ingests the daily in whatever format the supervisor submits: a photographed paper form, a PDF from a site tablet, a structured WhatsApp message. It extracts the production fields and checks them for internal consistency at the moment of submission. A round advancing 4.2 meters on a charge designed for 3.8 meters is flagged immediately, not three weeks later when the invoice is disputed.
It populates the invoice spreadsheet from the extracted figures. It formats the mine's daily from the same source data. Every entry carries a timestamp and a source record, so the chain from field submission to invoice line has no undocumented steps and no re-entry by hand.
MSHA Part 50 (30 CFR Part 50) requires contractors to report accidents, injuries, and illnesses to the Mine Safety and Health Administration. The rule applies to every contractor at every mine site. An automated system can populate the Part 50 forms from your existing records, flag filings approaching the required reporting window, and maintain the documentation log that inspectors ask to see. It cannot decide whether a near-miss event crosses the threshold for a reportable injury. That determination carries real legal weight and it is yours.
The system reads what has already been decided and recorded. It does not make decisions.
Where Automation Stops
Ground conditions are the largest category of things that do not automate. They are also the ones that determine whether people go home at the end of shift.
A development heading that was stable last week can show signs of stress this week because of water infiltration, a change in rock mass classification, or a blast that disturbed the support pattern. The reading happens in the heading. A superintendent walking through notices bolt plate deflection, the drip pattern on the back, the sound the ribs make underfoot. He makes a call about whether to proceed with the next round, pull the crew for additional ground support, or bring in a geotechnical engineer.
That call draws on what he has seen in hundreds of headings over years of underground work. No daily report captures that experience, and a system with access to every daily ever submitted from that mine would still not have enough to make it. The daily captures what he decided and when. That record matters for safety compliance and for contract disputes. It does not contain the judgment that produced it.
Contract disputes sit in the same category. When the mine's countback comes in two meters short of your daily total across a full month, there is arithmetic on one side of the conversation. There is also a relationship, a project history, and a negotiation that depend on how well the contractor reads the room. A document reader can show you exactly which daily the discrepancy opened in and what the submitted record said that day. It cannot sit in that meeting or decide how to approach it.
Crew decisions do not automate either: whether to push through on a late round, who goes to which heading when two critical dates land in the same week, whether a face is safe to enter after a misfire. These involve a read of the crew's state and the ground's state that no submission record captures fully. The record of those decisions, once made, is exactly what the system is built to preserve.
The Data Question
The FAQ bank for this pillar has twelve data questions before a single question about function. That ratio is not a curiosity. It is the real barrier, and it needs a direct answer rather than reassurance.
Start with one contract, not the whole company. One month of one site. Redact whatever you want before you send it. The NDA is signed before anything is seen.
And no, your data does not train it. The AI is Claude on Amazon Bedrock, running in your own AWS account. AWS states that your content is not used to improve the base models and is not shared with any model providers. The system reads your codes, your rates, your handwriting and your exception patterns right where they sit. You own the account, you pay the bill, and you can lock us out with one click.
Your data already lives in the cloud today. Your email, your accounting system, the photos in the WhatsApp group on the site superintendent's phone. What changes with this system is that something can finally read all of it consistently and give you a uniform record across every heading.
What Happens When a Daily Is Missing
It happens every week. A supervisor submits a partial report. A field note goes to the wrong thread. A tablet dies overnight and the backup copy is missing the blast record.
An automated system handles this the way a sharp office manager handles it: it flags the gap the same day. The supervisor gets a prompt asking for the missing fields while the shift is still in memory. The gap shows in the daily dashboard as incomplete, not as a silent hole that surfaces three weeks later during invoice reconciliation.
A missing daily caught on the day it is missing is a five-minute fix: one message to the supervisor, one response. The same gap found at month end is an afternoon of backtracking, a phone call to site, and a discrepancy in the mine's records that now has to be resolved against their survey countback.
Automation does not prevent the missing daily. It prevents the three-week delay before it surfaces.
Where This Has No Business Going
The questions most vendors skip are the strongest trust signals in the set.
If your operation has no consistent written daily from the field, a document reader has nothing to read. Verbal reporting and mine-survey-only records mean the first project is building the written chain, not automating it. That is a different service, and an honest vendor names the difference before talking about the reader.
Exploration contracts where the reporting format changes every few months present a separate problem. A reader is set up for your format. If that format changes faster than it can adapt, the overhead of maintaining the system exceeds the time it saves. For exploration work, the better first step is a stable field submission template, not the reader.
Contracts below about 30 reporting days a month sit at the lower edge of where the system justifies itself. A well-designed spreadsheet and a clear submission process from the field handles that volume at lower cost and less complexity. This is an honest threshold, not a sales one. Below it, the spreadsheet is the right tool and a system like this one is overhead. A vendor who cannot name that line is telling you something about how they handle the rest of the conversation.
The Person Question
This question never gets asked out loud. But it shapes every conversation about the system. The office manager does not ask whether she is automating herself out of a job. She either forwards the page to the owner or she says nothing, and the difference depends entirely on whether she thinks the answer is no.
The work that automates is re-entry: taking a figure that exists in the daily and entering it into two more places, then tracing why three versions of the same number diverge by a fraction. That is the work that takes the most hours and produces the least recognizable value.
The work that does not automate is the judgment she applies to those numbers: noticing that a reported blast sequence is out of order with what the geologist said last week, knowing which contact at the mine will resolve a discrepancy quickly and which one will not, recognizing that three consecutive dailies showing zero advance means the heading is stuck and not that the supervisor stopped reporting.
A system that eliminates re-entry does not eliminate the person. It eliminates the part of the day that is least satisfying and returns that time to the work that is actually hard. That is the honest answer to the question she is not asking out loud.
Common questions
Is this AI, or is it just automation?
For daily production reporting, it is primarily document reading and structured extraction: the system takes whatever format the supervisor submits, pulls the key production fields, checks them for consistency, and routes them to the right destinations. The AI layer handles variable formats and handwriting. The work of deciding what the numbers mean is still yours.
AI hallucinates. I cannot have it making up meters.
The system does not generate numbers. It reads numbers that already exist in the field submission and flags the ones inconsistent with the round plan or prior dailies. When a field is missing, it surfaces a gap, not an estimate. The source document is attached to every extracted figure, so you can verify any value against the original in two clicks.
What does this not fix?
It does not fix verbal-only reporting, an inconsistent field submission process, or formats that change every few months. It does not resolve contract disputes, make ground condition calls, or replace the judgment your superintendent applies before calling a round safe to shoot. It preserves the record. Everything requiring a read of the ground or the room is still yours.
Is this going to replace my office manager?
The work it replaces is re-entry: taking a number that exists in the daily and entering it into two more places. That is the work consuming the most hours for the least recognizable value. The work it does not replace is the judgment she applies: noticing inconsistencies, knowing who to call about a discrepancy, and catching what a system cannot catch because it is not underground.
Does my data train your model?
No. The AI is Claude on Amazon Bedrock, it runs in your own AWS account, and your data does not train it. It reads your codes, your rate schedule, your supervisor's handwriting and your exception patterns where they sit. You own the account and can lock us out with one click. Your data already lives in cloud systems today. What changes is that something can finally read it consistently and give you a uniform record.
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.
