What does AI actually do for a mining procurement team?
It does the checking and searching that takes buyers away from buying. It compares, counts and finds, then explains what it found. It never approves an order or awards a contract.
| Job | What it does |
|---|---|
| PO vs contract checking | Compares every PO line against the agreed rate, quantity, unit and contract dates |
| Contract price book | One table of agreed rates and last paid prices, built from your own contract history |
| Spend analysis | Counts only approved, real spend, with clear rules everyone can see |
| Tender scoring | Scores bids against your criteria, calibrated against your own panel |
| Commercial assistant | Answers questions in Microsoft Teams from your purchasing, receipt, payment and stock data |
| Stock levels | Recalculates min/max for every line from real usage and lead times |
Each piece can stand alone, but they share the same cleaned data, so the numbers agree with each other.
Is every PO being paid at the contract rate?
Usually not all of them, and most teams can't say which ones. A PO checker compares every line against the contract and flags any gap before it's paid.
| Check | What gets flagged |
|---|---|
| Price | Unit price above the agreed rate for that item and supplier |
| Unit of measure | PO unit differs from the contract unit, so prices are converted before comparing |
| Contract dates | Order placed after the contract expired or before the new rate started |
| Supplier | Contracted item bought from a supplier with no agreement |
| Extras | Freight or surcharges the contract doesn't allow |
Example with made-up figures: the contract rate is $88.00 per unit. The PO comes through at $95.00 for 120 units. That's $7.00 x 120 = $840.00 above the agreed rate on one line. The buyer sees the flag, the contract it breaches and the source lines, then decides what to do.
More detail is on the PO vs contract rate checking page and the contract price book page.
Can you trust your spend numbers?
Only if everyone agrees what counts as spend, and that rule is applied the same way in every report. Disputes about spend figures often come from rules nobody wrote down.
Working with an international tier 1 mining operator, we found a whole class of wrong spend numbers. Quotes and unapproved orders were being counted as spend. One rule fixed it. A spend figure should only include orders that were approved and actually placed, and that rule now sits in the code, not in someone's head.
That's why every project starts by auditing the data, not by building. The audit surfaces problems the business didn't know it had, before anyone makes a decision on bad numbers. See the AI data audit case study.
How can AI help with tender evaluation?
It scores every bid against the same criteria, the same way, and shows its working. Your evaluation panel still makes the call.
For an international tier 1 mining operator, a tender scorer was calibrated against a real evaluation panel. It matched the panel exactly on one bid and came within a point on another. That level of agreement means the panel can use it as a consistent first pass, then spend its time on the bids that are close.
- Load your evaluation criteria and weightings.
- Run the scorer over past bids your panel has already scored.
- Compare the scores and adjust until it agrees with the panel.
- Use it on new bids as a first pass, with the reasoning shown for every score.
- The panel reviews, changes what it disagrees with, and decides.
Read the tender scorer case study.
What is a commercial assistant in Microsoft Teams?
It's a chat assistant inside Teams that answers commercial questions from your own data. You ask in plain English, it runs the query, and it shows the answer with the rows behind it.
For an international tier 1 mining operator, the assistant answers from millions of purchasing, receipt, payment and stock rows in about a minute. Commercial leadership use it without being asked to. That's the real test of any internal tool.
Typical questions look like this:
- "What did we spend with this supplier last quarter, and how much was off contract?"
- "Which open POs are above the contract rate?"
- "When did we last buy this part, and what did we pay?"
See the Teams commercial agent case study.
How do you know the AI's answers are right?
You test them, every time. Every answer the assistant gives passes 511 automated checks, and every mistake a user catches becomes a permanent test, so it can't happen again unnoticed.
Four principles keep it honest:
- Deterministic maths. Totals, rates and stock levels come from plain code. The AI only explains them.
- Every number links back. Any figure can be traced to the line it came from.
- Read-only access. It reads your source data and never writes to your ERP.
- A person approves. The AI recommends. It never decides.
It's built inside your own data platform, such as Azure, Databricks or Snowflake. The AI verification method explains the testing in detail.
What does it cost, and what's the first step?
The first step is a free 15 minute call with Tristan, or a quick enquiry. No obligation, no pitch. If there is something worth fixing, you get a clear scope and price for the build.
- 15 minute call. Free, no obligation, no pitch. Tristan asks what you run and what is costing you, and tells you straight whether there is something worth fixing and roughly what it would cost.
- A look at a sample export. Only if it's useful, and read-only.
- A clear scope and price. What gets built, how many days it takes and what it costs, in writing before any work starts.
- Build inside your systems. In your own Azure, Databricks or Snowflake, with read-only access to source data.
- Run, check and support. Monthly re-runs, new sites and fixes. A person approves every change.
| Item | How it's priced |
|---|---|
| 15 minute call | Free |
| Build: stock level engine, PO vs contract checker, contract price book, Teams assistant | Consulting day rate, agreed after a 15 minute call |
| Support: monthly re-runs, new sites, fixes | Consulting day rate, agreed with the build |
Work is charged at a consulting day rate. The price depends on your sites, systems and data, so we talk first, then Tristan sets out what he'd build, how many days it takes and what it costs. What moves the number is how many sites you run, which systems the data sits in, how clean that data is, and how many of the tools you need.
See the full AI for mining supply chain offer.
