Supplier lead time distribution: bu ne anlama geliyor
Your ERP holds one lead time per supplier: 15 days. In reality that supplier has delivered anywhere between 9 and 31 days over the past year.
Using the average means, by definition, being late in about half of cases.
The shape of the distribution
Lead times are not symmetrically distributed. They are bounded on the left — no supplier delivers in negative days — but have a long right tail: customs, a production stoppage, a holiday period, capacity pressure.
In this right-skewed distribution the mean exceeds the median, and the real risk sits in the tail.
Which percentile to plan against
Use percentiles rather than the average:
– P50 (median) — the typical case, useful for communication
– P85 — reasonable for normal planning
– P95 — for critical and A-class items
The gap between P95 and P50 is that supplier’s real uncertainty. A supplier averaging 15 days with a P95 of 31 carries 16 days of uncertainty.
What to measure
Elapsed time from order date to actual receipt, per order, over at least a year. Not the date the supplier committed to — the date that happened.
Delivery date consistency is a separate metric. A supplier who takes 20 days but always exactly 20 is more plannable than one averaging 15 with a range of 9 to 31.
Effect on safety stock
Lead time variability produces a larger term in the safety stock formula than demand variability does. If you are using the simple form that only includes demand volatility, your safety stock is systematically insufficient.
Building a supplier scorecard
Lead time distribution is one part of a broader performance measure. A meaningful scorecard has four dimensions:
On-time delivery rate — percentage delivered on the committed date. Early delivery is also a deviation: goods arriving early consume unplanned space and capital.
Lead time variability — P95 minus P50. A supplier with a low spread is plannable even with a higher average.
Quantity accuracy — difference between ordered and received.
Quality acceptance rate — percentage of arriving goods accepted. A rejected batch counts as undelivered.
Going into the conversation with data
Most supplier performance conversations rest on impressions: “you are always late”. With distribution data the conversation changes: “of 47 orders in the last 12 months, 11 exceeded 25 days, and all of them fell in these three-week windows”.
That may reveal a pattern in the supplier’s own capacity planning, and turns into a solvable problem.
The dual sourcing decision
Dependence on a single high-variability supplier is compensated with safety stock, and the annual cost of that compensation is calculable. If the setup cost of a second supplier is below that annual carrying cost, dual sourcing is economically correct.
This comparison is rarely made, because the portion of safety stock attributable to supplier variability is rarely separated out.
Computing the distribution is a day of work
Receipt date minus order date, grouped by supplier and material, over two years. Every ERP holds both fields.
What comes back is rarely a tidy curve: a tight cluster around the contractual date and a scattering of arrivals one to three weeks late. The tail is the useful part; the average produces a respectable number that describes none of the actual behaviour.
What a bimodal shape tells you
Two behaviours under one supplier code is worth investigating: two plants, two shipping modes, or a material sometimes made to stock and sometimes to order.
Splitting the code is worth more than any modelling work, and the data is already there.
Where the cost currently lands
Not in the planning function that set the lead time. In expedited freight, in substitution scrap, in the line that stopped, and in the service credit given to a customer.
None of it is attributed back to the assumption that produced it, which is why an obviously wrong lead time field can sit unchanged for years.