AI Solutions Locations All Services Website Design Perth SEO Perth Google Ads Perth Facebook Ads Perth Content & Video Real Estate Videography Shopify Perth AI Services Perth Case Studies Free Strategy Call
AI Case Study. Multi-Site Mining Operator

A min/max inventory engine matched to the client's own expert

AI inventory management with its working shown. A min/max engine proven to 99.8 percent agreement with the client's own inventory expert.

Start a Similar Project All Case Studies →
Client
Multi-site mining operator
Industry
Mining
Scope
Tens of thousands of items
Output
Review workbooks per site and department
Engine
Deterministic rules
99.8%
Agreement With The Expert
Day Lockout On Human Edits
Top Items Per Department
10,000s
Items Covered
01
Before

The problem

Stock levels were set by hand, and rarely revisited once set.

Some items tied up cash in stock that would never be needed. Others were running dry. Both problems sat in the same lists, invisible at scale.

Nobody could review tens of thousands of items by hand. So the levels stayed where they were.

Cash sitting still
Some items held stock that would never be needed, tying up cash for no return.
Items running dry
Other items were heading for empty under levels set by hand.
Too many to check
Tens of thousands of items. No person can review a list that size item by item.
02
The build

What we built

A rules based engine. Not AI guessing. It reads usage, stock and lead time data and recommends new minimum and maximum levels.

The reasoning is stated for every item, so a reviewer never has to wonder why the engine wants a level changed.

Output arrives as review workbooks, one per site and per department. Each team reviews its own items.

Deterministic by design
The same inputs always produce the same recommendation. Nothing in the numbers is guessed.
Reasoning on every line
Every recommended level carries its reasoning, stated in plain terms.
Shaped for review
Workbooks arrive per site and per department, built for a human to work through and sign off.
03
Under the hood

How it works

Every rule in the engine was settled against evidence, not opinion.

  • Usage is counted at warehouse level, stock transfers are excluded, and returns are netted off. Each rule was settled by comparing every candidate treatment against the expert's own workbook as the referee.
  • Refuses to re-recommend any item a human changed in the last 30 days.
  • Reviewers can dismiss items, and dismissals expire, so nothing stays hidden forever.
  • Ranks items by impact, so each department sees its top 25, not 1,200 rows.
  • Every workbook carries a stamp saying exactly what has and has not been verified.
The referee
Where two treatments of the data disagreed, the expert's own workbook decided. Every rule had to earn its place against it.
Humans outrank the engine
An item changed by a human in the last 30 days is left alone. The engine waits its turn.
04
Verification

Results and proof

The engine reached 99.8 percent agreement with the expert's own workbook. The remaining gaps were not defects. They were fixes he asked for himself.

The platform's data collection was then machine checked against the verified engine. It came back identical on usage.

The workbooks identify cash release and running-dry items for every department.

Agreement
99.8%
Match with the inventory expert's own workbook. The remaining gaps were fixes he asked for himself.
Data collection
Identical
The platform's collection was machine checked against the verified engine and came back identical on usage.
Coverage
Every dept
Cash release and running-dry items identified for every department.
05
After

What changed

Stock levels are now reviewed every cycle by a machine that shows its reasoning, and humans decide.