Most supply chain software makes a quiet assumption: that you can buy more of what sells. Place a bigger order, accept a longer lead time, pay a premium — but supply, ultimately, is something you can turn up. Demand is the uncertain part, and forecasting exists to reduce that uncertainty.
Refurbished electronics do not work that way. You cannot manufacture a two-year-old smartphone in good condition. It has to be traded in, collected, graded, repaired and tested first — and how many of them arrive next month is only partly within anyone's control. Supply is a forecast in its own right.
That is the problem refurbed, the Vienna-based re-commerce platform for refurbished electronics, set out to solve with numi. This article looks at why recommerce planning is structurally harder than conventional retail planning, and what changes when AI-powered forecasting is applied to both sides of the equation.
Why refurbished electronics break conventional planning
Three characteristics make this category unusually difficult to plan.
Supply is discovered, not ordered. Volumes depend on trade-in activity, corporate fleet refreshes, carrier upgrade cycles and the flow of returns. A new flagship launch releases a wave of previous-generation devices into the market months later. Planning has to anticipate that wave rather than react to it.
Condition is a dimension of its own. A single device model fragments into dozens of tradeable units once storage size, colour and condition grade are combined. These variants are not interchangeable from a customer's perspective, and they do not sell at the same rate or the same margin. Forecasting at the model level averages away exactly the signal a buyer needs.
Demand and price are tightly coupled. In recommerce, the competing product is often the new device, and the gap between the two prices moves constantly. Demand for a refurbished unit can shift sharply when the new model is discounted — without anything about the refurbished unit itself having changed.
Add multiple national markets, each with its own demand profile and regulatory context, and the planning surface expands quickly. Spreadsheets and rule-based reorder points struggle at that resolution. They tend to fail in one of two directions: capital tied up in variants that are not moving, or missed sales in the variants that are.
Forecasting at the level the business actually trades at
The first step was moving demand forecasting down to the level where decisions are made. Rather than forecasting a product family, numi forecasts each tradeable variant — model, configuration and condition grade — and learns the pattern each one follows.
That matters because variants of the same device diverge. Higher storage configurations often behave more like a separate product than a variation of the same one. Lower condition grades attract a more price-sensitive buyer whose demand responds to different signals. The forecast engine also handles predecessor-successor relationships, so when a generation is phased out its demand history informs the successor instead of being discarded — a recurring problem in categories with fast product turnover.
Safety stock follows from forecast accuracy rather than from a fixed rule — the core idea behind inventory optimization in numi. Variants that are predictable carry less buffer; volatile ones carry more. Over a large catalogue, that reallocation is where much of the working capital improvement comes from.
Automated replenishment and order creation
Forecasts only create value when they reach a purchase order. numi generates daily replenishment proposals across locations, ranked so buyers see the decisions that matter most first rather than working through a flat list.
Proposals are reviewed and released with one click into the ERP system, which removes the re-keying step where errors typically enter.
Supplier and partner performance made measurable
In a marketplace and partner model, supply reliability is a supplier-management question as much as a forecasting one. numi tracks partner performance through a full KPI system: promised versus actual lead times, fill rates, quality and grading consistency, and responsiveness.
The value is less in any single metric than in having one shared, fact-based view. Conversations with partners shift from anecdote to evidence, and the sourcing mix can be steered towards the partners who actually deliver.
Alerts instead of reports
Reports describe the past. Dynamic alerts flag the exceptions worth acting on today: a variant whose sell-through has broken from its forecast, a partner slipping against lead time, a category where availability is tightening ahead of a demand peak. Teams work from a short list of exceptions rather than reading dashboards to find them.
Results
numi was integrated into operational purchasing and planning processes within two months. refurbed now reaches 66% forecast accuracy through AI forecasting methods and the integration of external data sources, and reports 64% less lost revenue, achieved by reducing stockouts and using automated order recommendations.
numi brought a new level of precision and control to our purchasing operations. The tool helped us move from reactive supply chain decisions to a significantly more proactive and data-driven approach to demand and inventory optimization.
Tushita Sethi, Operations Manager · refurbed
What other recommerce businesses can take from this
The specifics are refurbed's, but the pattern generalises to any business where supply is uncertain rather than merely delayed — recommerce, second-hand and rental, spare parts, agricultural and natural-material trade.
Forecast at the level you actually trade at, not the level your catalogue is organised at. Treat supply availability as something to be predicted rather than assumed. Let safety stock follow forecast accuracy rather than a blanket rule. And make sure the output of planning is a decision a buyer can execute in one click, not a report someone has to interpret.
Do those four things and the planning problem stops being a matter of working harder through spreadsheets. It becomes a question of which decisions deserve human judgement — which, in a market that moves this fast, is where the advantage actually sits.
