Turning a Covid supply shock into $3.8m of recovered revenue

QuantSpark's optimisation work recovered $3.8m in net revenue over six months for a premium global footwear and lifestyle brand, by solving what looked like two separate inventory problems as one. That is the headline. Everything else in this case study explains how a single mathematical model, not a bigger warehouse or a faster freight contract, got there.

The problem: stocked and starved at the same time

In 2021 the brand's supply chain was hit from both ends. Worldwide shortages lengthened freight lead times, and power cuts reduced capacity at overseas factories. The result was not a simple shortage. It was a business simultaneously over-stocked and under-stocked at SKU level: roughly 384,500 units sitting in excess while 175,800 units were in shortfall elsewhere in the range.

That split matters. Excess stock ties up capital; shortfall stock means priority lines go unfulfilled and revenue is lost. Treated as separate problems, the natural response is to discount the excess and expedite the shortfall, an expensive fix that does nothing to stop the imbalance recurring. QuantSpark treated them as one problem instead: a misallocation of capacity, freight and purchasing decisions across roughly 1.3 million items, all pulling against the same constrained production and logistics network.

The method: one model, four moving parts

QuantSpark built a single mathematical optimisation model, powered by machine learning demand forecasting, that tied excess stock directly to fulfilment rather than managing each in isolation. Four elements did the work.

First, machine learning forecasts projected excess stock against outstanding purchase orders at SKU level, giving the model a clear view of where the imbalance actually sat. Second, the model reallocated available production capacity toward the SKUs in shortfall, using product priority rankings so that high-value lines were protected first. Third, it balanced ocean freight against air freight, the cheaper-but-slower option against the faster-but-costlier one, to protect availability without letting air freight quietly erode margin. Fourth, it re-optimised the entire purchasing plan across the full 1.3 million-item base, within the real constraints of production and logistics capacity.

Crucially, the model did not hand back one fixed answer. It ran multiple scenarios, including 15% and 20% air-freight budget caps, so leadership could see the explicit trade-off between cost and availability and choose a position rather than inherit one by default.

How it ran: from stock data to a purchasing plan

The workflow moved in one direction: SKU-level stock, purchase order and logistics constraint data went in, and a re-optimised purchasing plan came out. Machine learning forecasting sat first in the chain, projecting the excess-versus-shortfall picture; the optimisation engine sat next, reallocating capacity and freight mode by product priority; a scenario-testing layer sat alongside it, stress-testing freight-budget choices before the plan reached execution. The categories at work, a demand-forecasting layer, an optimisation engine and a scenario-testing layer, are generic components of a supply chain analytics stack rather than named proprietary tools.

What it delivered

Over six months, the engagement delivered a $3.8m net revenue uplift from increased order fulfilment, the direct result of shortfall SKUs being served that would otherwise have gone unfulfilled. Alongside that headline figure, more than 64,000 units of excess stock were removed and a 30% increase in available capacity was freed, adding resilience for future seasons rather than a one-off fix. In the primary scenario, the final stock position fell by 19%, and shortfall across the top three priority rankings dropped by around 39%.

The right way to read the $3.8m is as the value of sales made possible once capacity was reallocated to the SKUs that needed it, measured over the six-month window and set against the freight-mix trade-off the model exposed rather than hid. That is what makes it a recovery figure with a clear driver, rather than a headline number floating free of its mechanics.