Delivery performance cannot be measured until someone defines what counts. Is a shipment on time if it leaves the factory on the agreed date, or if it arrives at the destination on the agreed date? Is a partial shipment on time, or does it count against the order? Is a shipment that is ready but held for a missing document on time?
Most disputes about delivery performance are disputes about definitions, not about facts. The factory says it shipped on time. The buyer says it arrived late. Both are correct under different definitions, and neither definition was written down.
The fix is unglamorous: agree the definition at the start, in one sentence, and use it consistently. Everything else follows from that. The range of fabricated parts this applies to is listed under other parts.
Once goods leave the factory, the factory controls very little. Port congestion, customs holds, carrier schedule changes, and weather are outside its authority. A delivery metric that includes those events measures the factory's ability to forecast shipping, not its ability to produce.
That does not mean the factory has no responsibility. It means the responsibility should be split at a clear boundary. Readiness is the factory's metric. Transit is a shared metric, because the factory chooses the carrier and the routing even if it does not drive the truck.
Splitting the metric removes the most common excuse and the most common unfairness at the same time. A factory can then be measured on what it actually does, and logistics can be measured separately.
Readiness means the order is complete, inspected, packed, documented, and available for collection on the agreed date. It is a state the factory fully controls, and it is the fairest basis for comparison between suppliers.
Measuring readiness requires a record of the date the order became ready and the date it was promised. The difference is the performance. Nothing else is needed, and nothing else should be allowed into the calculation.
A factory that reports readiness honestly will sometimes report badly, which is the point. A metric that is always good is a metric that is not being measured.
Partial shipments distort every delivery metric, and they are common in mixed work because different components finish at different times. If the metric counts only complete orders, a factory can improve its score by holding a finished portion until the last item is ready, which is worse for the buyer.
A better approach is to measure line-level readiness: each line of the order has its own promised date and its own ready date. The order-level figure is then a summary rather than the primary number.
This also makes the conversation more useful. Instead of arguing about whether the order was late, both sides can see which lines were late and why, which is the information needed to prevent a repeat.
None of these require software. A shared sheet with two dates per line is enough to start, and it is more honest than a dashboard built on unclear definitions.
An average delivery figure conceals the shape of the distribution. A factory that is on time for most orders and very late for a few can show a respectable average while being unusable for a buyer who depends on the late ones.
The tail is what matters in supply chains, because a late critical component stops an assembly regardless of how many other parts arrived early. Measuring the worst decile, or simply counting the number of lines more than a defined period late, is more informative than any mean.
This is also where blame creeps in. A review that opens with an average invites a defence. A review that opens with three specific late lines invites a fix.
Buyers affect delivery performance as much as suppliers, usually through late releases, changed specifications, and delayed approvals. A measurement system that ignores these inputs will attribute everything to the factory and will therefore be wrong.
The honest approach is to record the date the factory could have started, which is the date the last input arrived. Performance is then measured from that date, and the buyer's own contribution becomes visible.
This is not a way of excusing lateness. It is a way of separating two different problems that need two different fixes. A factory that is late from a late start needs capacity. A factory that is late from an early start needs process.
A delivery metric survives only if it is cheap to maintain. If it requires a weekly exercise to reconstruct dates, it will be abandoned in a busy quarter and revived only after a failure.
Two dates per line, recorded when they occur, maintained in a shared file, reviewed monthly with exceptions named: that is enough. It will not produce a sophisticated score, and it will produce something better, which is a shared factual basis for a conversation.
For buyers of fabricated parts, the practical test of a supplier is whether it can tell you, without preparation, which of last month's lines were late and why. A shop that can answer is measuring. A shop that cannot is estimating, and estimating is what produces the arguments that delivery metrics were supposed to prevent. The company background, the customisation process, the stainless exhaust systems and titanium components, and the dealership terms are worth checking before that test is set.
