Lead time

Your supplier already told you the lead time. In receipts.

Not in the quote — in the dates on the last ten purchase orders you closed. Those dates say something the quote does not: how much the number moves. For this article, supplier lead time means sent-to-supplier through sellable; the linked piece explains every leg inside that interval. The ten receipts below are invented so the arithmetic can be checked.

A replenishment calculation has a lead-time field in it somewhere: reorder point, days of cover, the date the next order has to go out. They all need to know how long stock takes to arrive. If yours was filled in once, from a number a supplier said out loud, and never touched again, this post is about what should be in it instead.

That number is a promise. Your receipts are a record. When the two disagree, the record is what actually happened, and it is sitting in your purchase order history waiting to be counted.

This post is only about extracting it. What lead time should include — production, QC, freight, customs, the receiving lag at the far end — is a different question, worked through in the post on what lead time actually covers. Get that boundary settled first, because the arithmetic below is only as good as the two dates you feed it.

Step 1 — Pull the pairs

For every closed purchase order with one supplier, you need two dates:

• The day the order went to the supplier. Not the day you created the draft, not the day you decided to order. The day they received it.
• The day the stock became sellable. Not the day the container docked, and not the day the invoice was reconciled.

Subtract, and you have one observation. Do it across the orders you have closed with that supplier and you have a distribution instead of a single figure.

Two boundary rules stop this from going wrong:

Your own delay is not the supplier’s lead time — but it is still your lead time. If a purchase order sat in someone’s drafts for six days before it was sent, those six days do not belong in the supplier’s performance figure. They absolutely belong in the number you plan replenishment with, because they are six real days between deciding and receiving. Keep both figures. One tells you which supplier to trust; the other tells you when to reorder. Conflating them is how a supplier gets blamed for an internal approval queue — which is the subject of the last post in this series.

A split delivery forces a choice, and the choice depends on the question. If 700 units arrived in 41 days and the remaining 300 took 68, recording one order that took 68 days throws the 41 away, and averaging the two into a single 54 describes a delivery that never happened. Counting them as two observations keeps both dates, but it also lets 300 units carry the same weight as 700, and those are not equal exposure. Delivery-level observations answer “how long does a shipment from this supplier take”; a quantity-weighted measure answers “how long does the stock I actually sell take to arrive”. Decide which question you are planning against, then apply the same rule to every order so the observations stay comparable. Short shipments are one of the places lead-time data gets quietly corrupted.

Step 2 — Sort them and read the spread

Here are ten closed orders with one supplier, in days from sent to sellable:

Sorted, those are 38, 39, 40, 41, 41, 43, 44, 47, 52, 65. Which gives:

The supplier quoted 40 days. That quote may describe their target or their standard case, but your history shows what actually happened: three of these ten orders landed at 40 days or better, and seven did not.

Half your orders beat the median. That is what a median is. The mean makes no such promise — here it sits three days higher, because one order took 65.

What the choice costs, in units

Take a SKU selling 20 units a day. Lead-time demand is days times rate, and that is the floor your reorder point is built on:

Read the last column, because it is the whole decision. Planning at the supplier’s quoted 40 days would have left you short on seven of the last ten orders. Planning at the 45-day mean would have left you short on three. Planning at 52 leaves you short on one. And notice what the median row does and does not do: 42 days splits these ten evenly, which is exactly the job of a median, and it is also five orders you would have run past.

The gap, and what to call it

Look at the gap between the median (42) and the observation that covered nine of these ten (52). Ten days, or 200 units on this SKU.

Those 200 units are lead-time buffer, and you derived them from evidence rather than assuming them. They are what stands between planning on a typical delivery and planning on a slow one, under an empirical rule of covering nine of the last ten. That is intuitive and it is useful. It is worth naming carefully, though, because it is not the safety-stock calculation. Service-level safety stock combines demand variability, lead-time variability and a chosen service-level factor; what you have here is one of those terms, measured from your own receipts, with the others left out.

A safety-stock method that treats lead time as fixed misses this source of variability. Look at where the 27 days of spread in this sample came from: demand did not move at all — the SKU sold a flat 20 a day throughout — and the supplier still delivered in 38 days once and in 65 days once. Every day of that spread is on the supply side.

If you carry safety stock for demand risk but hold your lead time constant at a quoted figure, you have protected against demand variability while leaving this demonstrated supply variability untreated.

Step 3 — Pick a number, write it down, revisit it

There is no percentile that is correct for every SKU. The level you plan to is a policy input you choose, not a default you inherit, and choosing it means buying one thing with another:

1. Choose the historical coverage level you want to plan to first, and know what it costs. Higher means more of your past orders would have been covered and more units standing on the shelf waiting. Lower means less stock and more orders that would have run past the plan. The table above prices every choice in both directions at once — 840 units and five orders covered at the median, 1,040 units and nine orders covered at 52 days, 1,300 units and ten at the slowest. Pick the row you can defend, then read across to the days.
2. Move up for SKUs where a stockout is expensive. A product with contracted availability, or one whose listing or subscription base you would rather not interrupt. For these, a more conservative point — including the slow end of your observed history — may be defensible.
3. Move down for SKUs where holding is expensive. Short shelf life, seasonal, bulky, capital-intensive. Here the median plus a small margin may beat carrying 1,300 units of something that expires.

What ten observations can carry. Ten deliveries describe ten deliveries. The second slowest covered nine of these ten, and that is a reasonable place to plan from, but it is an empirical planning point drawn from this history — not a settled ninetieth percentile for the process behind it. Ten more orders could move it in either direction. Treat the figure as the best reading of the history you have, and expect it to move as the history grows.

Then set a review cadence and keep it. Recompute on a regular cadence, and whenever there is evidence the supplier process has changed — a different factory, a different freight forwarder, a season they handle differently. A lead time measured in March is a claim about March.

In SKU Compass, lead time is held per SKU per supplier: the manufacturer’s SKU table carries Lead as a column beside MOQ, Case and Last PO, and the panel offers a paste control for bringing cost, MOQ and lead-time columns in from a spreadsheet.

Which is the right place to end this, because it is the shape of the problem. The field is one number per SKU per supplier. Everything above is about what that one number should be — and the honest answer is that it should be a point you chose from your own history, deliberately, rather than the figure on the quote.

When you do not have much history yet

This method needs closed orders, and a new supplier has few of them. There is no count at which the number flips from unreliable to reliable; it is simply better evidence with each delivery. Until you have enough closed orders to see a spread, the quoted figure is what you have. Record every delivery as it closes, and recompute as the history grows.

Older observations are evidence about older conditions. If the process materially changed — a different factory, a different route, a supplier who moved production — analyze the post-change orders separately rather than letting obsolete history dominate the number. That is a judgement about the supplier in front of you, not a number of observations you wait for.