Merchandise Planner Interview Questions for AI Training Work

I use comparable products with similar characteristics as a baseline rather than forecasting from nothing, adjusting for known differences like price point or seasonality, and I build in more buffer for uncertainty than I would for an established product with a solid sales history.

I allocate based on each location's actual historical sell-through rather than distributing evenly, since even distribution overstocks slow locations and understocks fast ones. I revisit allocation regularly rather than setting it once at the start of a season, since demand patterns can shift.

I compare actual sell-through against the forecast at defined checkpoints rather than waiting until the season ends to evaluate accuracy, and I revise when the deviation is significant enough to be a real trend rather than normal week-to-week variation. Reacting to every small fluctuation causes more disruption than it solves.

I weigh the cost of a stockout, like lost sales and customer dissatisfaction, against the cost of excess inventory, like markdowns, and lean toward the side with the higher expected cost for that specific product category, rather than applying the same buffer to every item regardless of its margin and demand volatility.

I look at leading indicators, like sell-through rate acceleration in early weeks or a shift in which variants are selling, rather than waiting for the aggregate trend to become obvious, since by the time it's obvious in the top-line number, the window to react proactively has often passed.

I'd place an expedited reorder as soon as the sell-through trend is clearly outpacing forecast rather than waiting for more confirmation, since the cost of a stockout on a strong seller usually outweighs the cost of ordering slightly early. I'd also flag the forecast miss so future planning for similar products accounts for it.

I'd rely more heavily on scenario planning and shorter reorder cycles rather than trusting a single point forecast based on historical data alone, since normal patterns are less reliable under unusual conditions. I'd build in more flexibility to adjust orders as real data comes in rather than committing fully upfront.

I'd shorten the planning and reorder cycle for that category rather than committing to a long-range forecast with low confidence, since more frequent, smaller decisions reduce the cost of being wrong compared to one large upfront commitment. I'd also track which specific factors drive the volatility to see if any of them are predictable enough to build into the model.

I forecasted a steady sales pattern for a product that ended up having a much sharper seasonal spike than historical data suggested. Once early sell-through showed the deviation, I adjusted the forecast and expedited additional inventory, and I incorporated a more detailed seasonality analysis into how I forecast that category going forward.

I initially assumed a product would perform similarly to a comparable item from the prior season, but early data showed a different customer segment driving sales than expected. Adjusting the allocation and marketing focus based on the actual data, rather than sticking with the original comparable-product assumption, improved sell-through for the rest of the season.