The problem
A multi-store retailer and distributor — around thirty staff, roughly A$5M turnover, based in regional Victoria — was ordering by feel. One buyer, working across four store locations and a small trade counter, decided what to reorder and how much, mostly from memory and a weekly walk of the shelves.
The result was the pattern every small retailer knows. Some lines were consistently over-ordered — stock sat for months, tied up cash, and eventually got discounted to clear. Other lines ran dry at the worst times, and customers walked or waited. The buyer was good at their job; they simply could not hold the demand curve of several thousand SKUs in their head, and the point-of-sale reports they had didn't tell them what to do next — only what had already happened.
The owner didn't want to hand ordering to a machine. They'd seen "automatic reordering" pitched before and didn't trust it near their cash. What they wanted was a second opinion the buyer could sanity-check in minutes, and a reason to believe it.
What we built
Eight weeks: two weeks of scoping (we sat with the buyer through two ordering cycles and read eighteen months of sales history), four weeks of build, two weeks of pilot on a single high-turnover category before widening.
The deployed system:
- Pulled daily sales history per SKU per location from the existing point-of-sale system — no new hardware, no rip-and-replace.
- Modelled demand per SKU, accounting for seasonality, day-of-week patterns, recent trend, and known promotions.
- Joined in each supplier's lead time and minimum order quantity, so a suggestion accounted for how long stock actually takes to arrive.
- Produced a weekly reorder worksheet: for each SKU, a suggested order quantity, the forecast behind it, and a plain-English reason.
Crucially, the system suggests. It never orders. Every line lands on the buyer's worksheet for review, and nothing becomes a purchase order until the buyer approves it.
How it works
The buyer opens a weekly worksheet, ranked by urgency. Each row reads in plain terms: "Suggest ordering 24 units. Forecast demand over the next 3 weeks is ~30 units; you have 11 in stock; this supplier takes 10 days. Demand is up ~15% on last month, in line with the usual winter lift for this line."
That explanation is the point. The buyer can see whether the forecast matches what they know — a supplier price rise coming, a local event, a line being discontinued — and adjust or reject a suggestion on the spot. When they override, the system records it. Those overrides feed back so the forecast learns the buyer's local knowledge over time.
It augments the buyer's judgement rather than replacing it. The buyer stays the decision-maker; the system does the arithmetic across thousands of SKUs they could never track by hand, and shows its working so the decision stays theirs.
The results
These are pilot estimates, measured against the same category's prior-year performance and extrapolated cautiously across the range.
| Before | After (pilot) | |
|---|---|---|
| Stockouts on tracked lines | baseline | ~20–30% fewer (est.) |
| Excess / slow-moving inventory | baseline | ~15–25% lower (est.) |
| Buyer time on manual ordering | ~15 hrs/month | ~3 hrs/month |
| Working capital tied up in overstock | baseline | modestly freed |
The clearest early win was working capital. Trimming the consistently over-ordered lines freed cash that had been sitting on shelves — a modest but real amount for a business this size — without the shelves looking any barer to customers. The buyer's time on the weekly ordering grind roughly halved, and the fewer stockouts meant fewer of the small lost sales that never show up in any report.
How the team owns it
The buyer owns the worksheet, and that was the design goal from the first week. There is no autonomous ordering to switch off, no black box to trust blindly — the buyer reviews, adjusts, and approves, exactly as before, but with the arithmetic done and the reasoning laid out.
We ran two supervised cycles with the buyer before handing over, wrote a one-page runbook for reading a suggestion and its explanation, and set a simple monthly check: stockout rate, overstock value, and how often the buyer overrode a suggestion. A rising override rate is the signal to look at the forecast, not to argue with the buyer.
If you're ordering by gut across more SKUs than one person can hold in their head, this is the kind of engagement we'd scope. A short readiness audit tells us whether your sales history is clean enough to forecast from, and our AI automation work covers the build. Talk to us about your range.