Ask most inventory planners what their lead time is for a given supplier, and they'll give you a number. "Fourteen days." "Three weeks." That number usually comes from the supplier's stated lead time, possibly updated by the planner's experience. It's a single figure, and it lives in the ERP as a fixed parameter.
That single figure is almost certainly wrong — not because the planner is careless, but because lead time isn't a fixed value. It's a distribution. And the distribution's variance, not its mean, is what determines whether your safety stock protects you or leaves you exposed.
This is one of the places where we see planning teams systematically under-invest. They've got sophisticated demand forecasting setups, careful ABC classification, attention to promotional lift and seasonality. Then they use a supplier's average lead time — often one that was set when the purchasing relationship started — in their safety stock formula, and wonder why stockouts keep happening on items where the demand forecast was accurate.
How Lead Time Variance Breaks the Safety Stock Formula
The standard safety stock formula — in its most common form — is: Safety Stock = Z × σ_demand × √LT, where Z is the service level Z-score, σ_demand is the standard deviation of daily demand, and LT is lead time in days. More complete versions incorporate lead time variance explicitly, but many planning tools and spreadsheet implementations don't.
The complete formula accounting for both demand variance and lead time variance is:
Safety Stock = Z × √(LT × σ²_demand + D²_avg × σ²_LT)
where D_avg is average daily demand and σ_LT is the standard deviation of lead time in days. When σ_LT is nonzero — which it always is in practice — ignoring it systematically understates the safety stock you need to hit your target service level.
The magnitude of the error depends on how variable your lead time actually is. A supplier with a 14-day average lead time and ±1-day standard deviation is essentially fixed — the simplified formula works fine. A supplier with a 14-day average and ±6-day standard deviation — not unusual for overseas suppliers, or domestic suppliers during peak logistics seasons — has a completely different safety stock requirement at the same service level target.
Plug realistic numbers in: 14-day average, 6-day standard deviation, average daily demand of 20 units, demand standard deviation of 8 units, 98% service level (Z = 2.05). The simplified formula gives safety stock of about 69 units. The full variance-adjusted formula gives 119 units. The difference — 50 units, or 2.5 days of average demand — is inventory you need but aren't holding if you're using the simplified formula with the average lead time.
Where Lead Time Variance Actually Comes From
Understanding the sources of lead time variance is useful because different sources have different predictability — and different implications for how you should model them.
Supplier production queue variation
Most suppliers don't produce your item on demand; they batch it with similar orders in production runs. The timing of those runs, relative to when your purchase order arrives, creates baseline queue variance. This is structural and largely unpredictable at the order level, though it can sometimes be estimated from order volume seasonality at the supplier's end.
Logistics and transit time variation
Transit times vary based on carrier availability, port congestion, customs delays (for imported goods), and weather. The 2021-2023 period showed how dramatically this can move — some importers saw lead times double or triple at peak congestion. But even in normal conditions, transit time variance of ±3-5 days on a 2-week lane is common. This component is partially predictable from port congestion indices and carrier performance data.
Inbound receiving and quality inspection variation
Even after goods physically arrive, the time from arrival to available inventory depends on your receiving capacity, any required quality inspection, and system processing. High-volume periods — before major promotional events, at fiscal year-end — often compress inbound throughput, adding 1-3 days to effective lead time even when the shipment arrives on schedule. This is entirely internal and entirely trackable if you're logging receipt-to-available timestamps.
Measuring the Variance You Actually Have
The first step, which many planning teams haven't done, is building a lead time distribution for each active supplier from historical purchase order data. You need: the date each PO was placed, the confirmed ship date, and the date the goods were received and available in inventory. From those three timestamps, you can calculate realized lead time per order and build a distribution.
Minimum useful history: 20-30 purchase orders per supplier. With fewer than that, your variance estimate is too noisy to be reliable. For suppliers you order from infrequently, you may need 12-18 months of history to get enough data points.
What you'll typically find: lead time distributions are right-skewed, not normal. The minimum lead time (things going perfectly) is a harder floor than the maximum (things going wrong). A supplier with 14-day average might have a distribution that runs from 10 days to 28 days, with the tail pulled right by the occasional delayed shipment. Using the normal distribution assumption in your safety stock formula when the actual distribution is right-skewed understates the frequency of worst-case delays.
A more robust approach than σ_LT in the formula: use the 85th or 90th percentile lead time as your planning lead time, rather than the mean. If your 90th percentile lead time is 21 days but your average is 14 days, plan replenishment timing against 21 days for A-tier items where a stockout is most costly. Accept that you'll sometimes hold an extra week of inventory; that's the cost of protecting service level against a genuine worst-case distribution tail.
Supplier-Level vs. SKU-Level Lead Time
One nuance that matters operationally: lead time variance often isn't uniform across all SKUs from a given supplier. Fast-moving commodity items from a supplier may have more stable lead times than slow-moving specialty items from the same supplier, because the commodity items are running through a standard production and logistics process while the specialty items are made-to-order with more queue uncertainty.
If you're using a single lead time parameter per supplier across all SKUs — which is the default in many ERPs — you're averaging over this variation. The practical fix is to segment SKUs from each supplier into lead-time-stable and lead-time-variable classes and apply different safety stock multipliers. It's more administrative overhead, but for high-value SKUs where stockout cost is significant, the differentiation is worth it.
Dynamic Lead Time Monitoring in Practice
Supplier lead times aren't static. They drift over time with changes in production capacity, logistics contracts, and broader supply chain conditions. A lead time parameter set two years ago may be significantly off from current reality — either in the mean or, more commonly, in the variance.
We built lead time monitoring into Automcore specifically because we saw this pattern repeatedly: planning parameters calibrated to one market environment becoming quietly stale as conditions changed. The approach we use is a rolling 90-day realized lead time calculation per supplier that updates the model's lead time distribution weekly. When the rolling average or variance drifts more than 15% from the longer-term baseline, the system flags the supplier for review.
That flag doesn't automatically update safety stock — we put a planner review gate on it because lead time changes sometimes reflect short-term disruptions that will self-correct, and sometimes reflect permanent changes in the supplier relationship that require a full replenishment strategy update. The system's job is to surface the signal; the planner's judgment determines the response.
We're not saying you need sophisticated monitoring to manage lead time variance. A quarterly manual review of realized vs. assumed lead times per supplier, using PO history you already have in your ERP, is a good starting point. What we are saying is that treating lead time as a fixed number set by the supplier's stated terms — and never updating it from actual performance data — is a structural accuracy problem that no amount of demand forecasting improvement will fix.
The Compounding Effect of Demand and Lead Time Variance
The reason lead time variance is so damaging isn't just its direct effect on safety stock calculations. It's the compounding: demand variance and lead time variance combine multiplicatively in the full safety stock formula. High demand variance on a SKU with stable lead time is a manageable problem. That same demand variance on a SKU with high lead time variance requires safety stock that may feel disproportionately large.
The planning implication: segment your attention by the combination of demand variability and lead time variability. SKUs with both high demand variance and high lead time variance are your highest-risk items — the ones most likely to cause stockouts when your plan doesn't account for both. Those items deserve explicit safety stock modeling, not just a standard safety days parameter applied uniformly across the catalog.
If we had to pick one thing that most often explains persistent service level underperformance on items with good demand forecasts, it's this: the supply-side variance isn't in the model. The forecast team focuses on demand, the procurement team focuses on cost and relationships, and nobody is formally tracking the distribution of realized lead times and feeding it back into the planning parameters. Closing that loop is one of the highest-leverage improvements a planning team can make.