Measurement and indicators

Lead time

In French: Lead time (délai de traversée)

Lead time is the time between the customer’s request and its fulfilment, waits included. It is the delay the customer feels, and it almost never bears any relation to the sum of the working times.

What the customer measures, and what the workshop measures

Lead time is counted from the moment a request enters the process to the moment it leaves it fulfilled. It therefore includes everything: the actual work, but also the queues, the transfers between departments, the approvals, the rework and the periods when the file does not move.

That is what separates it from added-up working time. A file that requires two hours of human intervention can take three weeks to cross an organisation, and the gap is not an anomaly: it is the rule. In most administrative processes, the time when something happens is a very small fraction of the total lead time.

That difference explains why speeding up the gestures shortens almost nothing. If the work weighs two hours over three weeks, halving the duration of the gestures saves one hour on a lead time that stays at three weeks — a result invisible to the customer, obtained at the price of real effort at the stations.

What really lengthens it

Work in progress. A request that arrives behind forty others waits for the forty to go through. It is the first lever, and the least intuitive: reducing the number of files open at the same time shortens the lead time of each without anyone working faster.

Handovers. Every change of hands creates a wait, because the next person never picks up the file at the instant the previous one puts it down. A process that crosses six departments accumulates six queues, none of which appears in the working times.

Batches. Handling in bundles of fifty forces the first item to wait for the other forty-nine, then to wait again for the whole batch to be handled before moving on. Batch size is an organisational choice, and it is paid for entirely in lead time.

Rework. A file sent back for missing information does not start again where it stopped: it goes back into a queue, often the first one. Poor input quality lengthens the lead time far beyond the time of the correction.

How to record it

Lead time is measured on real files, not on the theoretical process. A sample of already closed requests is taken, the entry date and the exit date are recorded, and the distribution is looked at — not only the mean, which systematically hides the long cases.

The median and a high percentile say far more than the mean. A customer does not judge on the usual lead time: they judge on the time they waited three times longer than expected. Announcing a lead time met nine times out of ten is a verifiable commitment; announcing a mean is not.

Value stream mapping gives the same information in the form of a drawing, separating processing time and waiting time on two lines. It is the representation that makes the gap indisputable in a meeting, because it becomes visual before being quantified.

Little’s law

lead time = work in progress ÷ throughput

work in progress : number of requests present in the process

throughput : number of requests leaving per unit of time

This relationship assumes nothing about the order of processing or the regularity of arrivals; it only requires that the system be observed over a period during which it neither fills up nor empties lastingly. Its practical consequence is direct: at constant throughput, halving the number of open files halves the lead time felt.

The common mistakes

Confusing lead time with processing time. It is the most frequent and the most costly confusion: it leads to timing stations and announcing gains the customer will never see, since most of the lead time is decided between the stations and not on them.

Steering on the mean. It is pulled down by the simple files and hides the tail of the distribution, that is, exactly the cases that produce the complaints.

Trying to shorten the lead time by launching more files in parallel. It is the opposite that works: opening more simultaneous work lengthens the queue, hence the lead time of each, while giving the impression of being more responsive.

On the shop floor — Telecoms customer service

On a cancellation request, the record of twenty closed files gave a mean of nine days and a maximum of thirty-four. The cumulative working time was under an hour. Most of it played out on two approval waits, neither of which was steered, because no indicator measured the end-to-end lead time: each department tracked its own and met it.

The other tools illustrated by this trade: services, it and engineering.

Frequently asked questions

What is the difference between lead time and cycle time?

Cycle time is the interval between two successive outputs of the process — it describes the production rate. Lead time is the duration of the journey of a given request, waits included. A process can turn out one part every two minutes while taking six weeks to handle the order that arrived today, if the queue in front of it is long.

How do you reduce a lead time without hiring?

By acting on what is not work: limit the number of files open at the same time, reduce batch sizes, remove handovers and make the input information reliable to avoid back-and-forth. These levers require no additional resource and produce the fastest gains, because they tackle the majority share of the lead time.

Does lead time have to be measured on every request?

No, a sample of actually closed files is enough to establish a starting point, provided it covers the simple cases and the complicated ones in their real proportions. What must be systematic is the timestamping of entry and exit: without those two dates, no retrospective measurement is possible, and that is the most common gap.

Does lead time include the time waiting for the customer?

It includes all the waits located inside the process, including those due to a missing document the customer has to provide. Many organisations deduct that time from their internal measurement, which is legitimate for steering their own teams but distorts the reading: the customer, for their part, counts from their first request.