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. International Tier 1 Mining Operator

A commercial AI agent that answers in Teams, with its working shown

Procurement and commercial staff ask questions in Microsoft Teams and get verified answers from millions of purchasing, receipt, payment and stock rows in about a minute.

Client
International tier 1 mining operator
Industry
Mining
Footprint
Multi-site
Interface
Microsoft Teams
Deployment
Inside the client's own data platform
Automated Behaviour Checks
Checks At User Testing
~1 min
Typical Time To Answer
Millions
Transaction Rows
01
Before

The problem

How much did we spend with this supplier. How many orders are awaiting approval. Simple questions, asked all the time.

Answering one meant an analyst, a spreadsheet, and days of waiting. Every question became a task in someone else's queue.

Off the shelf chatbots were not an option. They were not trusted with real commercial data, because they cannot show their working. A figure with no trail behind it is a figure nobody will act on.

The old path
A question went to an analyst. The analyst built a spreadsheet. Days later an answer came back, with no easy way to check how it was produced.
Why generic chatbots failed
They answer confidently but cannot show their working. On real commercial data, that is worse than no answer at all.
02
The build

What we built

A governed agent that runs entirely inside the client's own data platform. The agent goes to where the data lives, not the other way around.

It works under two layers of written rules. One layer governs how SQL is written. The other governs what may be claimed in the answer. They are separate on purpose. A correct query can still produce a misleading sentence, so the sentence gets its own rules.

The agent is wired into Microsoft Teams. People ask where they already work, and the answer lands in the same chat.

Rules for the query
A written layer of rules governs how the agent writes and runs SQL against the platform.
Rules for the claim
A second layer governs what may be stated in the answer, and what must carry a caveat.
Asked in Teams
Questions are asked and answered inside Microsoft Teams. No new tool, no new habit to build.
03
Under the hood

How it works

Every question runs through the same steps, whoever asks it.

  • Resolves supplier names into system codes before it queries anything.
  • Writes and runs its own SQL across millions of transaction rows.
  • States the time window, currency basis and caveats on every answer.
  • Refuses when the data cannot support a claim, and says exactly what is missing.
  • Serves several people at once, with the same rules for everyone.
A refusal is an answer
When the data cannot support a claim, the agent says so and names what is missing. That is what makes the rest of its answers worth trusting.
Every answer carries its basis
Time window, currency basis and caveats are stated on the answer itself, not left for the reader to guess.

The agent's behaviour is tested, not assumed. 379 automated checks run against it, grown from 140 during user testing. Each check pins down something the agent must do, or must refuse to do.

Every figure in every answer traces back to the rows it came from.

The clearest signal came from the users themselves. The client's team adopted the agent unprompted within days of going live.

Behaviour checks
379
Automated checks run against the agent's behaviour, grown from 140 during user testing.
Traceability
Row level
Every figure traces back to the rows it came from.
Adoption
Days
Adopted unprompted by the client's team within days of going live.
A question that used to take days now takes a minute, and the answer arrives with its sources.