Supply Chain Manager interview questions

I start with a clean statistical baseline, then add only documented business intelligence that changes future demand.

• I segment history by SKU and location, remove one-off stockout and data errors, and preserve real seasonality and promotions.
• I compare a few suitable models on holdout periods rather than choosing the model with the best fit to historical data.
• Sales and marketing overrides need an owner, reason, expected uplift, and expiry date so I can measure whether they added value.

I forecast at the lowest level that is both decision-relevant and statistically reliable, then reconcile it with higher-level plans.

• Procurement may need SKU-location-week detail, while finance may plan at category-month level.
• Very sparse combinations create noisy forecasts, so I may forecast at product family level and allocate down using recent mix.
• I check that the detailed forecast sums back to the approved aggregate plan before releasing it.

I keep the system baseline and every manual override as separate versions so their accuracy can be compared.

• The baseline captures repeatable demand patterns without unverified optimism from sales targets.
• Each override records the event, quantity, period, requester, and evidence such as a signed customer commitment.
• After actuals arrive, I calculate forecast value added by override type and stop accepting overrides that consistently make the plan worse.

I forecast promotions as baseline demand plus an incremental uplift tied to comparable events.

• I match prior promotions on discount depth, channel, duration, placement, and season instead of using one average uplift.
• I adjust for cannibalization of related SKUs and for demand pulled forward from the following period.
• I compare sell-through and remaining stock with the uplift assumption after the event, then update the promotion library.

I use an analogous product or category curve, then manage the launch through frequent assumption reviews.

• I choose analogs based on price, channel, customer need, launch timing, and expected distribution rather than product name alone.
• I create low, base, and high cases with explicit assumptions for trial, repeat purchase, and cannibalization.
• I buy initial stock in stages where lead time allows and replace assumptions with actual sell-through as soon as data arrives.

Forecast bias shows whether forecasts systematically overstate or understate demand rather than merely being inaccurate.

• I calculate signed error consistently, for example forecast minus actual, so positive bias always means overforecasting.
• I review bias by product family, planner, horizon, and customer because an aggregate total can hide offsetting errors.
• Persistent positive bias usually raises excess stock, while negative bias raises stockouts, so I assign corrective action to the source of the pattern.

I usually use WAPE for portfolio reporting and add bias because no single metric tells the full story.

• WAPE divides total absolute error by total actual demand, so high-volume items receive appropriate weight.
• MAPE becomes unstable when actual demand is zero or very small and can overemphasize low-volume SKUs.
• At SKU level I may use MAE or scaled error, but I keep the metric definition and forecast horizon fixed for fair comparisons.

I use a tracking signal to flag sustained forecast bias that ordinary monthly error may hide.

• A common version divides cumulative signed error by mean absolute deviation, using one documented sign convention.
• I set alert limits based on portfolio behavior and investigate repeated breaches rather than automatically changing the model.
• The review checks structural causes such as lost distribution, a price change, or an incorrect seasonal profile.

Forecast value added shows whether each planning step improves the forecast compared with the prior step.

• I compare a naive benchmark, statistical baseline, planner adjustment, sales override, and final consensus at the same horizon.
• A step with repeatedly negative value added is simplified, retrained, or removed rather than protected as tradition.
• I evaluate both accuracy and bias because an adjustment can reduce absolute error while introducing systematic overforecasting.

I create a demand history that reflects customer need rather than blindly forecasting from shipment records.

• I correct duplicates, unit conversions, returns, and item or location mapping errors before modeling.
• I flag periods constrained by stockouts because shipments may understate unconstrained demand.
• Promotions and one-time orders remain identifiable so the model can treat them as events instead of normal recurring demand.

ABC ranks items by business value, while XYZ separates stable demand from volatile or intermittent demand.

• An AX item deserves tight review and high data quality because it is valuable and predictable.
• A CZ item should not receive the same planning effort or service target as an AX item because it is low-value and erratic.
• I use the combined segments to set review frequency, forecast method, counting priority, and inventory policy.

I set safety stock from the target service level and uncertainty during replenishment lead time.

• For stable independent demand, the calculation uses demand variability, lead-time variability, and the service factor rather than a fixed number of weeks.
• I use clean historical variability and exclude distortions such as one-time promotions or data errors.
• I recalculate when lead time, demand pattern, or service target changes and validate the result against actual stockouts and excess.

The reorder point is expected demand during replenishment lead time plus safety stock.

• Demand and lead time must use compatible units, such as daily demand multiplied by calendar-day lead time.
• I base the lead time on actual receipt performance, including internal approval and receiving time, not only the supplier quote.
• I review exceptions when orders trigger too early or too late because stale master data makes a correct formula useless.

EOQ estimates the order quantity that balances ordering and holding costs, while MOQ is a commercial constraint that may force a different quantity.

• I calculate EOQ with realistic order administration, freight, capital, storage, and obsolescence costs.
• If MOQ is far above economic demand, I negotiate pack size, order frequency, blanket releases, or price breaks instead of accepting excess stock silently.
• The final policy also respects shelf life, capacity, and service requirements that the basic EOQ formula does not include.

I set differentiated service targets based on customer impact, margin, substitutability, and replenishment risk.

• A critical high-margin item with no substitute may justify a 98 or 99 percent target and more safety stock.
• A slow, low-margin item with an acceptable substitute may use make-to-order or a lower target.
• I document the policy by segment and quantify the inventory investment required before business owners approve it.

I identify excess against future usable demand, not simply by applying one age threshold to every item.

• The review combines on-hand stock, open orders, forecast, shelf life, supersession, and contractual return rights.
• I stop or reschedule supply first, then pursue transfer, substitution, promotion, supplier return, or controlled disposal.
• I record the root cause, such as forecast bias, MOQ, or engineering change, and change the responsible planning rule.

The decoupling point determines where the chain holds forecast-driven stock and where it responds to a real customer order.

• Upstream of the point, I plan common materials or modules against aggregate demand and replenishment lead time.
• Downstream, I postpone final configuration until demand is known, reducing finished-goods variety and obsolescence.
• I place the point by balancing response time, customization, process lead time, and the cost of holding each inventory form.

Both metrics describe inventory efficiency, but I interpret them with service and product mix rather than in isolation.

• Inventory turns are annual cost of goods sold divided by average inventory at cost.
• Days of inventory convert the same relationship into days, making it easier to compare with lead time and policy.
• A higher turn rate is not an improvement if it comes from repeated stockouts, while low turns may be justified for seasonal or long-lead items.

I allocate constrained stock using transparent business priorities rather than rewarding the location that escalates first.

• I consider confirmed demand, customer service commitments, lost-margin risk, available substitutes, and each location's remaining cover.
• I reserve some supply for high-priority future orders when consuming everything today would create a larger shortage tomorrow.
• I publish the allocation rule and track overrides so commercial teams can challenge assumptions without bypassing the process.

I keep the scorecard focused on measures that drive sourcing and corrective-action decisions.

• Core measures include on-time in-full delivery, lead-time adherence, defect or rejection rate, and responsiveness to issues.
• Commercial measures include price variance, invoice accuracy, and delivered cost where the supplier controls freight.
• I define each metric, source it from agreed ERP or quality records, and review trends with the supplier at a fixed cadence.