Halden Home: High Inventory, Poor Promo Fill Rate
Halden Home is a $310M housewares brand. Its 12-month plan totals 1.2M units, built bottom-up from last year's shipments plus a 6% growth factor applied uniformly to every SKU. Last year's results: MAPE 28% at SKU-week level but 11% at product-family level, with a persistent bias of +9% because sales consistently over-forecasts. Eight promotional weeks a year carry a 3x lift and 45% forecast error. Overall fill rate is 96%, but it is 84% on promoted items. Finished goods sit at 71 days of supply and clearance/obsolete stock is 9% of the total. Weekly demand averages 23,000 units with a standard deviation of 6,000; replenishment lead time from Halden's own plant is two weeks. The CFO has committed to a 20% inventory reduction without losing the 95% cycle service level on core items, and the new S&OP lead must present a plan to the executive committee next month.
1. Explain how Halden can carry 71 days of inventory and still fill only 84% of promoted demand.
Read the worked answer
The two symptoms have one root cause: inventory is not where the variance is.
The bias is uniform, so it misallocates. A +9% uplift applied to every SKU over-builds slow and non-promoted items - visible in the 9% clearance share - while doing nothing for the promotional SKUs where error is 45%. Aggregate bias near zero is entirely compatible with severe SKU-level error; the 28% SKU-week MAPE versus 11% at family level proves the error lives at the item level, exactly where buy decisions are made.
Safety stock is sized on the wrong distribution. At 95% cycle service level with a two-week lead time, SS = 1.65 × 6,000 × √2 ≈ 14,000 units - a number calibrated to an average week. In a promotional week demand triples and the standard deviation roughly triples too, so the same policy under-protects promos by several multiples while over-protecting ordinary weeks. Fill rate on promos collapses for structural reasons, not bad luck.
The metrics hide it. A 96% blended fill rate averages a 99% non-promo performance with 84% on promos, and 71 days of supply averages excess core stock against shortages in the items customers actually want during the eight weeks that drive margin.
Diagnosis: Halden is running a single forecast and a single inventory policy across two very different demand patterns. The fix is structural separation - baseline versus promotional demand - not a better single forecast.
2. Design the forecast improvement program. What would you change first, and how would you prove it worked?
Read the worked answer
Sequence the work by value, not by tooling.
- Separate the demand streams. Build a statistical baseline forecast (shipments plus point-of-sale where available, cleaned of promotions and stockouts), then layer a promotional plan that uses a lift library by SKU family, mechanic, and discount depth, with cannibalization and pull-forward built in. Two errors tracked separately can be fixed; a blended error cannot.
- Forecast at the right level of aggregation. Statistical forecasting at product-family level (11% MAPE) with disaggregation to SKU using recent mix shares will beat item-level forecasts. Keep SKU-level only for A items where the planner adds genuine judgment.
- De-bias with governance, not adjustment. Publish bias by planner and by account monthly. Apply a documented correction factor to the sales overlay rather than letting each planner quietly shade numbers. Target ±3% bias at family level within four quarters.
- Measure forecast value added. Compare naive forecast, statistical forecast, and the consensus forecast each cycle; where the human overlay is not adding value, remove it. This is the only reliable evidence that the process is improving rather than merely changing.
- Tie the S&OP agenda to decisions. The meeting should resolve volume, mix, and promo capacity conflicts, not review accuracy slides.
Success criteria for the executive committee: bias within ±3%, FVA positive for the consensus forecast on A items, SKU-week MAPE down 5 points, and promotional fill rate at 95% - all reported monthly against the same baseline so improvement is demonstrable.
3. How would you deliver the CFO's 20% inventory reduction without breaching 95% cycle service level on core items?
Read the worked answer
Twenty percent of 71 days is roughly 14 days, or about 47,000 units on a 1.2M-unit plan. A blanket cut would breach service; the reduction has to come from identified structural excess.
- Remove the bias buffer. The +9% systematic over-forecast is embedded in open orders and pipeline. Correcting it removes roughly two weeks of inventory with zero service impact, because it is inventory covering demand that was never real.
- Pool and then disaggregate. Size safety stock at family level and allocate to SKUs by share of variance, rather than setting each SKU independently. Pooling typically yields 10-20% lower total safety stock at the same service level.
- Segment the service policy. 95% is a core-item commitment, not a universal one. Moving C items to 90% service (z of 1.28 versus 1.65) cuts their safety stock by about 22% while leaving A items untouched - the visible promise stays intact.
- Attack lead time, not just stock. Safety stock scales with the square root of lead time. Cutting plant replenishment from 14 days to 10 days reduces safety stock by roughly 15% for the same service level, and it is a permanent structural gain.
- Delay differentiation. The slow-moving fashion SKUs differ mainly by finish; holding semi-finished goods and labeling late removes finished-goods weeks without touching core availability.
Sequence: bias correction and service segmentation first (no capital), lead-time work and postponement second. Publish days of supply, fill rate, and obsolete share together so the committee sees that the reduction came from structure, not from starving the core.