Case studyAnonymised · measured off the live system
A distributor that runs its
back office on AI.
A Western Australian wholesale food distributor runs day-to-day operations on a custom AI system. Not a chatbot bolted onto a website. An operational layer on top of the ERP and Microsoft 365 they already had.
01The numbers
Measured,
not promised.
This is not a pilot and not a demo environment. The system runs the distributor’s working week: pricing, ordering, reconciliation, claims and reporting, against the live ERP, every day.
The figures alongside were measured off the live system over a representative 14-day window. The client stays anonymous. The numbers do not need to.
- 44scheduled jobs running unattended
- 99.7%success rate over a 14-day window
- 34packaged workflows doing real jobs
- ~40 hrsof admin recovered each week
- Anonymised WA distributor · live system · 14-day window
02The shape of it
Two halves, on top of the
systems they already had.
A workforce that does the recurring jobs, and a workplace where the team directs it. Nothing was ripped out to make room for it.
- The workforce: 34 packaged workflowsReal recurring jobs, packaged and repeatable: apply supplier cost changes across staggered price tiers, process emailed orders into the ERP, reconcile every supplier invoice line by line against its purchase order, draft the monthly rebate claims, forecast demand, build per-store reports.
- The workplace: a team dashboardRoughly 49 pages the whole team logs into, with five-tier role-based access, two-factor auth and an AI assistant on every screen. The team approves price batches, activates promotions, works the retention list and runs the warehouse pick queue from it.
03Before and after
Same team. Same ERP.
A different working week.
Before
- PricingMaintained by hand across wholesale tiers and RRPs, drifting in spreadsheets.
- OrdersArriving as forwarded emails, PDFs and spreadsheets, each retyped into the ERP.
- InvoicesSupplier invoices eyeballed against purchase orders, or skipped when time ran out.
- ReportingWeekly and monthly reports assembled by hand.
- RebatesClaims filed late or not at all. Unfiled claims are margin left on the table.
After
- On schedulePrice lists, cost changes, reports, statements and claims run on their own schedule.
- 3,361 to 103,361 successful runs against 10 failures over a representative fortnight.
- Checked, every timeEvery invoice diffed line by line. Every claim drafted on cadence.
- Early, not lateChurn risk, stock problems and cost drift surface early instead of late.
- People decideThe team approves, the system does the typing.
The biggest wins were leakage, not speed.
The most valuable jobs were the ones quietly being skipped because they were too tedious to do consistently: checking every supplier invoice, filing every rebate claim, keeping related products on the same margin logic. Automating those recovers money, not just hours.
On top of that sits capability that did not exist before at all: per-store analysis, churn scoring, demand forecasting and competitor intelligence that nobody had the hours to do, now running on a cadence.
04Why it is safe
The hard part
was never the AI.
An AI system only earns a place in a real back office if it cannot quietly do damage. So the boundary comes first, on every write, on every send, with no exceptions for convenience.
The genuinely difficult engineering was this boundary layer, not the AI. That ordering is why it works in production instead of being a demo.
- No direct writesThe AI never writes to the live database. It works through purpose-built capability layers that expose narrow, read-only-by-default operations.
- Dry run firstEvery write to live data goes through a reviewed process that runs dry-run first and shows its output before anything moves.
- Explicit human approvalWrites execute only when a person has seen the output and said yes.
- Everything loggedEvery action lands in an audit trail, so you can see what ran, when, and why.
Honest limitations
- Supervised, not autonomousA human approves every consequential write and every external email.
- 0.3% of runs still failThey need a person to notice and fix them.
- Placeholders remainA few dashboard areas are still placeholders, marked as such.
- An estimate, not a stopwatchThe time-saved figure is the operator’s estimate, not a stopwatch study.
05The pattern that generalises
The domain changes.
The shape does not.
- WrapWrap the systems you already have. No rip-and-replace, no migration project.
- ScheduleFind the recurring admin and put it on a schedule.
- HuntHunt the leakage, not just the time.
- GateGate every write, log everything, dry-run first.
- PackagePackage it so a non-technical team can use it.
Recognise the shape?
If your business runs on an ERP, has a mailbox full of orders and invoices, and a person spending afternoons on admin that follows the same steps every time, this pattern applies to you.