Forecasting a Festive Season with a Broken Panel
Rupali Home Appliances must commit to a festive-season production plan for its 1.5-tonne inverter air conditioner by the end of June, four months before the peak. Last year's peak-month sales were 9,400 units. The company's historical forecasting has relied on a consumer panel that has just lost 40% of its members after a fieldwork contractor collapsed, and the remaining panel skews urban and higher-income. Trade partners submit optimistic indents because they can return unsold stock. Two new competitors entered in March with aggressive dealer incentives. The factory needs a single number to schedule components with a 10-week lead time, and component orders cannot be revised upward after July. The marketing head has proposed using last year's number plus 15% "because the market is growing".
1. Evaluate the marketing head's proposal and identify the specific biases in each available data source.
Read the worked answer
The "last year plus 15%" proposal is a naive extrapolation and fails on three counts. It ignores the loss of panel representativeness, it treats dealer indents as demand, and it assumes the competitive entry has no effect. It is also unfalsifiable: no one can say why 15% rather than 8% or 20%.
The available sources each carry a known bias:
- Consumer panel. Down 40% and skewed urban/higher-income, so it will overstate the premium end and understate rural and price-sensitive demand. It is now a biased sample, not a small random one; weighting cannot fully repair it.
- Dealer indents. Upward biased because unsold stock is returnable - dealers pay nothing for over-ordering. This is a classic rationing/returns distortion and will systematically overshoot true sell-through.
- Historical sales. Reflects last year's supply constraints and stock-outs, not last year's demand. If the brand was out of stock in peak weeks, the 9,400 figure understates demand.
- Competitor entry. New entrants with dealer incentives will take shelf space and may pull forward dealer stocking, making this year's early indents look strong for the wrong reason.
Any single-source forecast will be wrong in a predictable direction. The right approach is to triangulate and to state the bias direction of each input explicitly.
2. Design a forecasting approach that produces a defensible number for the component order, showing how you would combine the sources.
Read the worked answer
The constraint is asymmetric: component orders cannot be revised upward after July, but excess components are costly. So the forecast must be built for a decision, not for accuracy in the abstract.
Step 1 - Reconstruct last year's demand. Adjust 9,400 for stock-out weeks using weekly sell-out rates, producing a "demand, not sales" baseline.
Step 2 - De-bias the trade input. Ask dealers for sell-through expectations rather than indents, and apply a historical correction factor (compare past indents with actual sell-through). Where returns are permitted, discount indents by the observed overshoot ratio.
Step 3 - Repair or replace the panel. Re-weight the remaining panel and supplement it with a short dealer sell-through survey and any available category data. Treat panel output as directional only.
Step 4 - Build three scenarios (base, downside, upside) with explicit assumptions about competitor share loss and price effects, and assign rough probabilities.
Step 5 - Choose the order quantity against the cost of being wrong. Because upward revision is impossible, order at a level between the base and upside case for long-lead components, and hold a smaller buffer of short-lead components. Document the reasoning so the number is auditable.
This yields a number with a stated confidence range rather than a single unexplained percentage.
3. What monitoring would you put in place during the season, and what would you do if early sell-through diverges from plan?
Read the worked answer
Monitoring must be weekly and at the sell-through level, not the dispatch level, because dispatches to dealers can rise while consumer offtake falls - exactly the trap the returnable-stock system creates.
Set up:
- Weekly dealer sell-out reports for the top 60% of outlets by volume.
- A stock-cover tracker by region, flagging any dealer holding more than an agreed weeks-of-cover.
- A simple competitor price-and-incentive log in the top 20 cities.
- A pre-agreed trigger rule: if cumulative sell-through at week 4 of the season is more than 10% below the base case, escalate.
If divergence occurs, the response depends on direction. If sell-through is below plan, the danger is a channel stuffed with returnable stock; slow new dispatches, shift promotional support to consumer pull (finance schemes, exchange offers), and prepare to reallocate stock to stronger regions rather than discount nationally. If sell-through is above plan, ration stock to the highest-margin channels and prioritise service to dealers who sell through, rather than spreading limited supply thinly.
The key discipline is that the trigger and the response are agreed before the season, so the decision is not made under pressure with incomplete information.