Forecast Bias and Launch Planning at Meridian Home
Meridian Home sells seasonal outdoor furniture through 240 retail doors and e-commerce. The S&OP team forecasts at product-family level. Last year's forecast was 120,000 units; actual was 108,000. Monthly forecast errors ranged from -22% in April to +14% in July. The planner used a three-month moving average and raised the forecast after a competitor's stockout. Two new cushion colours launch in Q1 with no sales history. Fill rate fell to 91% in June, but leftover stock was 19% of year-end inventory. The CFO wants one accuracy metric for the scorecard and a rule for new launches. Sales wants to override forecasts freely; operations wants frozen periods.
1. Calculate last year's forecast bias and explain why a low MAPE can still hide a costly bias. Recommend the primary accuracy metric for the S&OP scorecard.
Read the worked answer
Bias = (Forecast - Actual) / Actual = (120,000 - 108,000) / 108,000 = +11.1%. The team overforecast by 12,000 units, or 11.1% of actual demand. Using forecast as the denominator gives 10.0%, so the convention must be stated.
MAPE averages absolute percentage errors. A forecast can have a moderate MAPE and still be consistently one-sided: overforecasting every month produces excess inventory and markdowns, while underforecasting produces stockouts. MAPE also treats over- and underforecasts symmetrically and becomes unstable when actuals are small, which matters for new colours.
Recommended scorecard: use signed forecast bias and weighted MAPE (WMAPE = sum of absolute errors / sum of actuals) as the primary pair. Add a tracking signal to detect persistent bias. Use MAPE only for comparing stable families. For Meridian, the +11.1% bias already explains the 19% leftover stock better than the monthly error range does; the scorecard should show bias by family and by month, not just one annual accuracy number.
2. Diagnose the likely causes of the June fill-rate failure and the 19% leftover stock. Propose three process changes, each with an owner and timing.
Read the worked answer
The symptoms point to a mix problem, not simply a total-volume problem. Family-level forecasting hides colour and SKU mix. The three-month moving average lags a seasonal turn, and the competitor stockout likely triggered an unvalidated uplift. If Sales overrode the plan freely, the overforecast went into the wrong items. Operations then faced late mix changes, while other colours accumulated.
Process changes:
- Introduce a frozen zone for weeks 0-4 and a slushy zone for weeks 5-8. Owner: Supply Planning. Timing: next S&OP cycle. Sales can override inside the frozen zone only with a documented cost approval.
- Create an override log with reason codes and track override accuracy. Owner: Demand Planning. Timing: immediately, reviewed monthly. This converts 'Sales wants to override freely' into a measured exception process.
- Forecast events separately from the baseline. Owner: Demand Planning and Sales. Timing: before the next season. Promotions, competitor stockouts, and weather should be event uplifts with a probability, not blended into a moving average.
Add a mix-level review for the top 20% of SKUs and a markdown trigger when weeks of supply exceed plan for two consecutive periods.
3. Design a launch forecast for the two new cushion colours. Show the logic, a reasonable initial range, and the reorder rule.
Read the worked answer
With no history, do not use a single-number forecast. Use an attribute-and-analog method:
- Map each new colour to the two closest existing colours by price, pattern, and channel. Use their first eight weeks of sales as the base curve.
- Adjust for distribution: 240 doors, but launch in the top 120 doors first. If analog colours sold 6,000 and 9,000 units in the first eight weeks at full distribution, a 50% door rollout implies a base range of about 3,000-4,500 units per colour, before any trend adjustment.
- Set an initial buy at the low end, for example 3,000-3,500 units per colour, with a supplier option for a fast second order. Do not buy the midpoint of a wide range.
- Define a reorder trigger: if sell-through reaches 55% of the initial buy by week 6, place the second order; if it is below 35%, hold and plan markdown. Review weekly for the first eight weeks.
The forecast should be a range with a review cadence, not a point estimate. Record the analog assumptions so the team can learn from the launch.