CASE STUDY
The shelves were empty.
The warehouse was full.
They asked us to build a better demand-forecasting model. The real problem: a smarter forecast still died in a two-day manual gap. Nobody owned the decision — and the system had quietly outgrown its design.
Real Problem
"Build us a better demand-forecasting model. Ours keeps getting it wrong."
Scope
A better forecast still died in a two day manual gap. The issue wasn't prediction — it was that nobody owned the decision.
The Challenge
A distributor moving 2,400+ SKUs daily across 340 retail accounts in Gauteng and the Western Cape faced a paradox: stockouts and overstock at once — slow stock had frozen [R_m] in working capital while hero SKUs bled repeat sales. A diagnostic found three compounding causes.
- Forecasting on four-week averages. Missed seasonal spikes, promo lifts, and Gauteng-vs-Cape demand gaps.
- Manual replenishment, two-day lag. In a fast-moving category, two days is expensive.
- Slotting by arrival date, not turnover. Fast movers sat in the least accessible locations.
The Solution
- Predictive forecasting: We fused sales, promotions, weather, and regional signals into daily replenishment calls — per SKU, per region.
- A conversational interface: We gave buying managers a plain-language way to ask, and get ranked, confidence-scored answers. The AI surfaces; they decide.
- AI-led slotting: We moved the top 20% of SKUs by turnover into primary pick locations — fulfillment speed up immediately.
The Impact
40%
Fewer stockouts
Real-time signals replaced rolling averages.
28%
Less working capital tied up
Slow-moving overstock freed from warehouse.
30%
Faster order fulfilment
Tighter replenishment cycles plus AI-led slotting.
The win wasn't a better forecast. It was a decision that finally had an owner.