When a SKU goes out of stock during an active selling period, the first instinct is usually to add more buffer. Order more, carry more, keep more safety stock across the board. It feels like a fix because it's simple and immediate. But if you're managing 800 SKUs and you blanket-increase safety stock by 30%, you've just added a significant inventory carrying cost — and you've probably still stocked out on the specific SKUs that matter most, while simultaneously overstocking slow movers.
The real problem isn't too little stock. It's wrong stock in the wrong quantities at the wrong time. Addressing stockouts precisely requires understanding where they come from — and most of them come from three identifiable, correctable causes.
The three root causes of most stockouts
1. Inaccurate demand forecasts on high-velocity SKUs
Your top 20% of SKUs by velocity are where stockouts hurt most. A hero product that sells 80 units per week gets a stockout event costing real revenue. Ironically, these are also the SKUs where spreadsheet forecasting fails most visibly — because when sales spike due to a promotion, a social moment, or a channel algorithm change, the manual process doesn't catch the signal fast enough to trigger a reorder before the stockout window opens.
High-velocity SKUs need demand models that are updated at least weekly, and ideally incorporate promotion calendars and channel-specific signals. If your Shopify store has a weekend flash sale and your Amazon Seller Central listing gets a "Best Seller" badge simultaneously, you're looking at a multiplicative demand event that a static reorder point model will miss entirely.
2. Lead time variability you're not modeling
Most mid-market brands set their reorder points using their supplier's nominal lead time — the number on the purchase order or in the supplier relationship discussion. But actual lead times have variance. An overseas supplier with a 60-day nominal lead time might deliver anywhere between 45 and 85 days depending on season, port congestion, and your priority in their production queue.
If your safety stock calculation uses 60 days but the actual delivery arrives on day 83, you've burned through your buffer. The fix isn't to use 83 days as your new nominal — it's to model the lead time distribution and set safety stock based on the percentile you want to cover. A standard formula uses z × σ_LT × d̄, where z is your service-level factor, σ_LT is the standard deviation of lead time, and d̄ is average daily demand. At a 95% service level, z ≈ 1.65.
3. Misallocated inventory across channels or locations
A brand running Shopify DTC, Amazon Seller Central, and two retail store locations can stock out on Shopify while the same SKU sits in a retail backroom. Inventory technically exists — it's just in the wrong place. This is a channel allocation problem, not a procurement problem, and no amount of additional safety stock solves it. What solves it is demand forecasting at the channel level, so you allocate inbound stock proportionally to where it's going to sell.
Where safety stock actually helps — and where it doesn't
We're not saying safety stock is the wrong tool. Safety stock is the right tool for demand uncertainty and lead time variability — it's the buffer that protects you against statistical variance in both. Where it fails is when it's applied as a flat buffer across a full catalog without distinguishing between SKUs by velocity, demand pattern, lead time variance, or margin.
A practical approach segments your SKU catalog before sizing safety stock:
- A-tier SKUs (top 20% by revenue contribution): model demand statistically, use formula-based safety stock, review weekly. These warrant higher service levels — a 95–98% target is reasonable.
- B-tier SKUs (next 30%): use exponential smoothing-based forecasts, monthly review cadence, 90–95% service level target.
- C-tier SKUs (bottom 50%): many of these have intermittent demand. Apply Croston's method or a similar intermittent-demand model. Don't size safety stock using continuous-demand formulas on SKUs that sell three units a month — you'll chronically overstock them.
This isn't a new idea — ABC analysis as an inventory classification framework has been documented since the 1950s. What's changed is the tooling to operationalize it. Doing ABC segmentation in a spreadsheet and then running three different safety stock formulas across 1,200 SKUs every month is a two-day job. It should be automated.
The replenishment signal problem
Many stockouts happen not because safety stock was wrong but because the reorder signal fired too late. Your reorder point is supposed to represent the quantity at which you trigger a new purchase order so the inbound stock arrives before you hit zero. If your demand increases between reorder cycles and you're not recalculating reorder points dynamically, you'll hit zero before the reorder point triggers.
Take a mid-market home goods brand selling on both WooCommerce and Amazon. One of their top-selling kitchen accessory SKUs averaged 35 units per week in Q3. In early Q4, a home improvement media publication featured the product and weekly velocity jumped to 80 units. Their reorder point was set at 280 units (35 × 8 weeks lead time). At 80 units/week, they should have reordered at 640 units. They didn't — the reorder point hadn't been updated since July. They stocked out within six weeks, right as Q4 holiday demand was building.
Static reorder points are a structural vulnerability for growing brands. Demand-driven replenishment — where reorder points are recalculated based on current demand velocity rather than set-and-forget formulas — closes this gap. See how replenishment planning differs from static reorder point models.
Forecasting error as a leading indicator
One of the more practical things a demand forecasting system gives you is visibility into forecast error before stockouts happen. If your MAPE on a specific SKU has jumped from 15% to 45% over the last four weeks, something has changed in the demand pattern — a new competitor, a change in ad spend, a channel algorithm shift. That's a signal to increase your safety stock buffer on that SKU, or to review the model assumptions, before you run out of stock.
Forecast error monitoring isn't just a retrospective quality check. It's a prospective risk signal. Brands that track MAPE and RMSE at the SKU level have a mechanism to identify demand pattern changes before they translate into inventory problems — rather than discovering them when the stockout alert fires.
What to measure to know if it's working
Stockout rate (the percentage of order requests that couldn't be fulfilled from available stock) is the headline metric. Track it by SKU tier, by channel, and by time period — not just as a total. A 2% total stockout rate that's driven by 8% stockout rate on your top-20 revenue SKUs is a very different problem than a 2% rate distributed evenly across the catalog.
Complement it with days of cover by SKU — how many days of current demand velocity does your on-hand inventory cover? A days-of-cover drop below your lead time window on any A-tier SKU is a reorder trigger, regardless of whether the reorder point formula has fired. If your supplier lead time is 45 days and your days of cover on a hero SKU hits 38, you're inside your buffer.
Inventory carrying cost as a percentage of average inventory value rounds out the picture. If you reduce stockouts by genuinely improving forecast accuracy, carrying cost stays flat or drops. If you reduce stockouts by padding safety stock bluntly, carrying cost goes up — and you'll see it clearly in the ratio. That's the test of whether the intervention worked or just transferred the problem from one balance sheet line to another.
Stockorlo tracks demand velocity per SKU and surfaces reorder recommendations before you hit critical stock levels — without requiring you to maintain separate safety stock formulas per ABC tier. See how replenishment recommendations are generated, or view pricing for your SKU count.