Spark Inventory Blog

A SKU selling three times its normal daily volume is not automatically a growth signal. It may be a wholesale order posted through the wrong channel, a promotion that ends Friday, a marketplace listing error, or a genuine shift in customer demand. Unusual demand detection ecommerce teams can trust is the process of separating those cases before they become expensive reorder decisions.

For an inventory planner or operations leader, the question is not simply whether sales are up. The useful question is: should this change the per-SKU forecast, buying plan, and next purchase order? Getting that decision wrong can create a stockout after a real demand surge, or tie up working capital in inventory ordered for a one-off event.

What unusual demand detection should identify

Unusual demand detection compares recent demand against an expected range, not just a prior-week or prior-month average. A spike is unusual when it is materially different from the demand pattern the item would normally produce after accounting for seasonality, day-of-week behavior, channel mix, and known events.

That distinction matters because average sales can hide operational risk. A SKU that normally sells 10 units a day might sell 30 units on a promotion day without changing its long-term demand. Another SKU might move from 10 to 18 units daily for three weeks after gaining a retail account. Both patterns are above average, but only one may justify a higher reorder recommendation.

Good detection also finds downward exceptions. A sharp demand drop can leave incoming inventory exposed to excess stock, particularly when a supplier lead time is long or an open purchase order is already in transit. Teams often notice stockouts quickly because they interrupt revenue. Slow-moving inventory can be quieter, but it consumes warehouse capacity and cash just as surely.

Start with the inventory decision, not the alert

An alert that says demand is unusual is only useful if it leads to a decision. Before setting thresholds, establish what the planner needs to determine for each exception: whether demand is real, how long it is likely to last, and what action remains possible given incoming inventory and supplier lead time.

For example, suppose a product has 400 units on hand, 200 units arriving in 21 days, and demand has risen from 12 units per day to 22. At the old rate, on-hand inventory represented roughly 33 days of supply. At the new rate, it represents about 18 days. The incoming shipment may arrive after the SKU is already at risk, depending on safety stock and the reliability of that delivery date.

The right response is not always to place a larger order. First confirm whether the demand is sustained and whether the inventory position is complete. Then evaluate alternatives: expedite incoming supply, transfer inventory from another warehouse, limit a promotion, adjust allocation by channel, or increase the next order quantity. The buying plan should show the consequence of each choice rather than treating every spike as a purchase trigger.

Validate the signal before changing the forecast

Sales history is useful, but it is not self-explanatory. Before accepting an unusual demand signal, check the operational context behind it.

Check for demand that is real but temporary

Promotions, influencer activity, bundles, holidays, product launches, and a customer-specific wholesale order can all create legitimate sales. The issue is duration. A two-day promotion should usually be represented as an event, not folded into the baseline demand forecast as though it will repeat every week.

If the promotion continues or becomes part of the commercial plan, then the forecast needs a deliberate adjustment. The planner should record the assumption: expected uplift, start and end dates, applicable channels, and whether replenishment should support the higher rate after the event ends.

Check for data and fulfillment exceptions

A demand anomaly can be created by an operational error. Duplicate orders, delayed order imports, cancellations recorded late, inventory adjustments, or a bulk order allocated to the wrong SKU can distort sales history. If those transactions enter the forecast unexamined, the system may recommend inventory for demand that never existed.

This is why data validation matters as much as forecasting logic. A single source of truth should combine sales, on-hand inventory, incoming inventory, supplier lead times, and purchasing data, while making questionable inputs visible for review. The goal is not to eliminate every imperfect record. It is to keep known exceptions from silently becoming purchasing assumptions.

Check whether the change is channel-specific

Total sales can look stable while one channel is accelerating and another is declining. That matters if inventory is physically separated by warehouse, reserved for wholesale, or committed to marketplace fulfillment. A company may have enough units in aggregate and still face a stockout where demand is occurring.

Review the unusual demand by SKU, channel, and location when those dimensions affect fulfillment. This makes the response more precise. A transfer recommendation may protect service better than a new purchase order if inventory is already available elsewhere.

Use lead time to decide how much evidence you need

The longer the supplier lead time, the earlier you need to respond to a plausible shift in demand. But earlier action also means acting with less certainty. This is the central trade-off in unusual demand detection.

For a domestic supplier with a seven-day lead time, it may be reasonable to wait for several days of confirmation before increasing an order. For a supplier with a 70-day lead time, waiting for perfect proof can leave no recovery option. In that case, the planner may make a measured increase to the reorder recommendation, preserving some flexibility rather than making an all-or-nothing bet.

Supplier constraints shape the decision too. Minimum order quantities, case packs, order cutoffs, and available capacity can turn a small forecast adjustment into a large working-capital commitment. A useful buying plan makes those constraints explicit. It should show projected stock risk, days of supply, incoming inventory, and the point at which delay becomes costly.

Build an exception workflow your team can run

The best process is not a flood of alerts. It is a short, repeatable review of exceptions that could change inventory action. For most commerce operators, that means reviewing material upside and downside deviations at the SKU level alongside stock risk and upcoming reorder dates.

A practical review follows a simple sequence. First, identify the variance against the expected forecast. Next, validate the transactions and identify the business cause. Then decide whether the event is temporary, persistent, or unresolved. Finally, update the planning assumption and review the resulting reorder recommendation.

Documenting the reason matters. Without it, the next planner has no way to distinguish a calculated override from a guess. A note such as "two-week marketplace promotion, no baseline change" is enough to preserve context. So is "new wholesale account confirmed, demand uplift expected through Q4." The record should be concise and tied to the operational decision.

Keep automation governed by operator approval

Automation is valuable when it prepares the work, not when it hides the decision. A system can combine recent sales behavior with inventory policies, supplier lead times, open purchasing, and warehouse context to flag unusual demand and recalculate stock risk. It can then prepare a draft purchase order based on the revised need.

The operator should still review the demand assumption, quantity, supplier constraints, and timing before approval. That is especially important when the signal is new, the SKU is high value, or the order carries a large minimum. Drafts reduce manual work; they do not remove commercial judgment.

Spark Inventory applies this approach by turning connected or imported operational data into per-SKU forecasts, stock-risk reporting, reorder recommendations, and a buying plan. When a purchasing action is appropriate, it prepares a draft purchase order for review, adjustment, and approval. Nothing is sent or committed without the operator.

When an anomaly should not change the plan

Some exceptions deserve observation rather than action. A short-lived spike with ample days of supply may not justify any forecast override. A low-volume SKU can appear volatile because one or two orders represent a large percentage change. And an item near end of life may require controlled depletion rather than replenishment, even if demand briefly improves.

The discipline is to distinguish awareness from intervention. Flag the change, capture the likely cause, and decide when to revisit it. Not every unusual pattern needs an immediate purchase, but every material pattern should be visible before it surprises the buying plan.

A good demand-detection process gives your team more than a warning that sales changed. It gives them enough context to decide whether to hold, transfer, expedite, reorder, or simply watch the SKU for another week.