When the Forecast Lied: A Demand-Planning Post-Mortem
A regional equipment distributor we'll call Northline came into the spring of 2021 confident. Their sales team had promised a strong rebound, the planner punched the numbers into the spreadsheet, and purchasing committed to a year of inventory. By August, two-thirds of the warehouse was tied up in product nobody wanted, while the three SKUs customers actually called about were on six-week backorder. Nothing about that outcome was bad luck. It was a forecast built on wishful thinking, and the post-mortem is worth walking through.
How the wish crept in
The first failure was treating a sales target as a forecast. A target is what you hope to sell; a forecast is your honest best estimate of what will actually happen. Northline's planner started from the revenue number leadership wanted, then worked backwards to the units that would justify it. That inverts the entire exercise. A forecast should be built from demand signals and then compared against the target, so the gap is visible and someone has to own it.
The second failure was ignoring the obvious distortion in the history. Their 2020 sales data was shaped by lockdowns, panic buying, and stock-outs — not normal demand. Feeding that raw history into a model produced a curve that looked precise and meant nothing. Pandemic recovery made this trap especially common: every planner in 2021 was staring at a baseline that no longer described the world they were ordering for.
What an honest forecast looks like
Rebuilding Northline's process did not require fancy software. It required discipline applied in the right order.
Clean the history before you model it. Flag and adjust the months distorted by stock-outs or one-time events. A demand you couldn't fulfil is not the demand that existed — reconstruct it from lost-sales notes and quote data.
Separate the baseline from the bets. Use a simple statistical baseline for the bulk of stable SKUs, and layer human judgement only where you have a real reason — a known contract, a discontinued competitor line, a confirmed project.
Forecast at the level you can defend. Forecasting every SKU at every location invites noise. Aggregate to a level where the pattern is stable, then disaggregate down. Accuracy lives in the aggregate; volatility lives in the detail.
Measure the error every cycle. Track bias (are you consistently high or low?) and accuracy (MAPE or similar) by product family. A forecast you never grade is a forecast you never improve.
Make it a conversation, not a spreadsheet
The fix that mattered most at Northline was social, not technical. They stood up a monthly sales-and-operations planning meeting where sales, finance, and supply met over one shared set of numbers. Sales had to commit to their additions in writing. Finance had to reconcile the forecast against the plan. Supply showed what each scenario meant for cash tied up in inventory. Once the forecast had owners in the room, the wishful padding evaporated — because someone now had to defend it.
Hold a forecast review on a fixed cadence, with the same people each time.
Make every manual override carry a name and a reason.
Report forecast accuracy back to the people who set the numbers — visibility changes behaviour.
Treat the safety stock you carry as the price of forecast error, and shrink it by improving the forecast, not by guessing.
Six months later, Northline's inventory was down a quarter and their fill rate on the fast movers was back above target. The forecast was no more 'accurate' in some magical sense — it was simply honest, graded, and owned. That is usually the whole game.
Getting demand planning, sourcing, and supplier commitments to line up is exactly the kind of work that pays for itself; if your forecasts and your purchasing have drifted apart, XNM's procurement, sourcing & contract management can help you rebuild the discipline.