Procurement Cycle Time and the Orders Still in Transit

Why this matters

Procurement cycle time is the average number of days from placing an order to receiving it, and it is computed over the orders that have actually arrived. Read that sentence again with a supply problem in mind. The order stuck at a supplier with no date has no receipt date, so it is not in the average. The order holding up a job that has now been rescheduled twice is not in the average. Every order the figure exists to warn you about is structurally absent from it, which is why cycle time can sit flat for five straight months while a shop's material situation falls apart underneath it. This is ordinary survivor bias, an average over completed cases that excludes the open ones, and the open ones are the bad ones. It has its own article; what follows is what it does specifically to a purchasing number, and what to put beside that number so it stops lying.

What the figure is made of

• Numerator: days elapsed, order placed to order received, summed.
• Denominator: the count of orders RECEIVED in the window.
• Silently absent: every order still open, at any age.
• Silently present: counter pickups and will-call collections, which complete the same day and enter the average as zero.

Two of those four lines do real damage and they pull in the same direction. The absent orders are the slow ones, so removing them pulls the average down. The counter pickups are instant by construction, so adding them pulls the average down too. A shop that is buying more at the counter because its supplier has stopped delivering will watch its procurement cycle time IMPROVE through exactly that transition.

Five months of a supply problem

Across all five months the mean cycle time held inside a 0.4 day band, 3.9 to 4.3, with no trend. Over the same five months the count of open orders rose from 3 to 19 and the age of the oldest open order rose from 6 days to 41.

The orders-received column is the one nobody watches and it is the tell. Orders received fell from 28 to 17, which is a 39 percent drop in the monthly receipt count while the shop's workload did not change. The mean is being computed over a shrinking population, and the orders leaving that population are leaving it because they did not arrive.

What the figure is at minimum, rather than what it averages

You cannot compute the true cycle time of an order that has not arrived, but you can compute a floor for it: an open order's eventual cycle time is at least its age today. That turns the unusable half of the population into a bound, and a bound is worth far more than an average that excludes it.

Month 5, using the same 36 orders in play:

• Received, 17 orders at a mean of 4.2 days. That is 17 times 4.2, or 71.4 order-days.
• Open, 19 orders. Twelve are under 5 days and average 2.5, which is 30 order-days. Three sit between 8 and 15 days and average 11, which is 33 order-days. Four are past 30 days, averaging 35, which is 140 order-days. Those three groups are 12 plus 3 plus 4, which is the full 19, and 30 plus 33 plus 140 is 203 order-days.
• All orders in play: 71.4 plus 203 is 274.4 order-days over 36 orders, which is 7.62 days.

So the all-orders figure is at or above 7.6 days, against a reported 4.2. The floor alone is 1.8 times the number the shop has been reading, and it can only rise as those four aged orders finally land.

Now strip out the counter pickups, because they are a different transaction. Six of the 17 received orders were collected at a counter the same day, contributing 0 days each. The 71.4 order-days therefore belong to 11 delivered orders, which is 6.49 days per delivered order rather than 4.2. Reworking the bound over delivered orders only, 274.4 order-days over 30 orders is 9.15 days, so the honest floor on a delivered order in month 5 is at or above 9.1 days.

Three figures, three populations, and each one has to carry its population in the same breath: 4.2 days as a mean over all received orders, 6.5 days as a mean over delivered orders received, 9.1 days as a floor over delivered orders in play. They are not three estimates of one quantity and averaging them together would be meaningless.

The two dates, and the interval that goes negative

The figure is a difference between two timestamps, so it inherits every problem either one has.

Measure from when the order was TRANSMITTED to the supplier, not from when the record was created. Those diverge in both directions in a real shop. A buyer who phones an order in at seven in the morning and keys it at four in the afternoon produces a record that understates the wait by most of a day, every time. A buyer catching up on a backlog of paperwork keys three orders on Friday that went out on Tuesday, and those three overstate it.

Reject any interval below zero rather than averaging it in. A receipt keyed against an order whose date was later backdated produces a negative number that no amount of reading will notice, because the total still looks reasonable. Size it: on a 17-order month with a 4.2 day mean, one order whose true cycle was 6 days entering the record as minus 1 moves the mean by 7 divided by 17, which is 0.41 days, or roughly a tenth of the reported figure. One row in seventeen, and it moves the headline by more than the entire five-month spread in the table above.

Treat a zero-day interval as a category, not a value. Counter pickups are genuinely zero and they are genuinely not procurement, so they get their own count and stay out of the delivered-order average, as above.

The interval is four waits, and the last one is entirely yours

Order-to-receipt is not one duration. It is four stacked in sequence, and knowing which one is moving is the difference between a supplier conversation and a shop conversation.

1. The supplier picking and packing your order. Genuinely theirs, and the only part that responds to being a good account.
2. The wait for the next departure on their route or freight schedule. Structural. A supplier running your area on Tuesdays and Thursdays cannot deliver a Wednesday order before Thursday however fast they pick, so part of your measured cycle time is a function of WHEN you order, not of how fast they are.
3. Transit. Theirs, and mostly fixed by distance.
4. The gap between the truck arriving and somebody keying the receipt. Entirely yours, entirely invisible in the number, and charged to the supplier rather than to the shop that caused it.

That fourth wait is worth sizing, because shops consistently assume it rounds away. A shop that keys receipts the next working day adds about 1 day to every order, which against the 6.5 day delivered mean is about 15 percent of the whole figure. Worse, it is not even. A delivery that lands Friday afternoon and gets keyed Monday carries a three-day lag against the one day an identical Tuesday delivery carries, so it reads two days slower, and a supplier who runs Friday routes will look like the slow one forever.

Two fixes, and the second is the real one. Key receipts the day the material lands, and where that is not going to happen reliably, take the date off the packing slip rather than off the moment of keying. Then read item 2 and order into the route: the cheapest day you will ever take off a lead time is the one you give away by ordering the morning after their truck ran.

Cut it by supplier, because that is where the decision is

A shop-wide cycle time is a blend of a working channel and a broken one, and the blend is not a thing you can act on. Same month 5, split:

• Supplier X: 5 delivered orders at a mean of 9.0 days, which is 45.0 order-days. X also holds all four of the open orders past 30 days.
• Everyone else: 6 delivered orders totalling 26.4 order-days, a mean of 4.4 days. The two groups sum to 71.4 order-days over 11 delivered orders, which is the 6.49 day figure above.

X takes just over twice as long as the rest on the orders it delivered, 9.0 against 4.4, and both of those are means over delivered orders received in the same month, so they compare directly. But the ranking understates X badly, because X also owns the entire aged tail that neither figure contains.

The decision this changes is not primarily sourcing, it is what the shop PROMISES. A special-order job on an X part cannot honestly be scheduled at nine days out. The truthful read is nine days typically, with a real chance of a month, and a shop has two defensible answers to that: quote the tail rather than the typical, or do not put the job on the calendar until the part is physically received. The second is usually cheaper, because a rescheduled install costs a mobilisation and a customer conversation, and the first costs only a longer quoted date.

Three more notes on how to report it, so nobody has to remember all of this:

• Report the MEDIAN days for delivered orders and read it as the headline. A right tail with a few long orders in it makes a mean describe nobody, and this population always has a right tail. The figures earlier in this card are means because the order-day bound is an addition and only means add; the bound is a separate calculation from the number you publish.
• Print the count and age of open orders beside it, always, in the same line. The median answers what a normal order does. The open count and oldest age answer whether normal is still the right word.
• Name the datum. The age used here is days since the order was PLACED, which is the same datum as the cycle time it sits beside. That is deliberately not the datum used for a chase trigger, which is days past the EXPECTED receipt date and belongs to a different job. Neither replaces the other and they are not two versions of one number.

Reading the pair

The bottom row is worth sitting with, because it is the shape that gets an owner congratulating a buyer. Cycle time improving while receipt volume falls is almost never procurement getting better. It is the difficult orders leaving the population by one of the two doors this card is about, still open at month end or collected in person, and the number cannot see either. Cancellation is a third door and the fulfilment-rate card owns it.

References

• See related: Time to Invoice Only Counts the Invoices You Sent, for averages computed over completed cases only
• See related: Open Purchase Order Value Is a Commitment, Not a Cost, for the chase trigger measured from the expected receipt date
• See related: Purchase Order Fulfilment Rate and the Orders It Never Counts, for where cancelled orders go
• See related: Choosing a Primary Supplier, for the decision a per-supplier cut feeds