Lead Time: Why Your Estimate Is Always Too Low

The symptom

"A standard order takes us about eight days." I hear some version of that sentence in first meetings, said with confidence and confirmed by everyone in the room.

Then someone measures, and the number comes out far higher. Nobody was lying. People know their own work precisely. What they can't see is the waiting between one person's work and the next, because waiting belongs to nobody.

Why the estimate misses

Asked for a lead time, people add up work in their heads. An hour of work preparation, a few hours on the machines, some assembly, dispatch, then a generous buffer on top. Each item is right. The sum leaves out that the order sits between steps, and sits far longer than anyone works on it.

Nobody notices, because waiting is never booked anywhere. Each person sees the order at the moment they work on it, and in that moment they are fast. Between those moments the order is in a queue, and no system reports queues.

The lead time formula

Manufacturing lead time runs from the moment the order arrives to the moment the carrier takes the goods. Every hour in between is one of four kinds.

Lead time = processing time + setup time + queue time + move time

• Processing time. A person or a machine is working on the order. You will also see it called run time.
• Setup time. The station is being prepared for this order. Someone clamps the fixture, loads the program, changes the tool.
• Queue time. The order waits for a machine, a release, a part or an answer.
• Move time. The order is travelling to the next station.

One ratio tells you how much of the lead time was work.

Process cycle efficiency = processing time ÷ lead time

It also goes by the name value-added ratio. In this article both mean processing time divided by lead time, as a percentage. Turned over, lead time ÷ processing time says how many times longer the order was in the building than it was worked on. Before you compare your figure with somebody else's, check which way round they divided.

One order, six stations

The shop, the order and every figure below are made up for this example. They are chosen so that you can check each line with a pocket calculator.

A fabrication shop on a single shift builds one machine frame to a customer drawing. The order arrives on a Monday at 7 a.m. The shop works Monday to Friday, 7 a.m. to 3 p.m., and we ignore breaks so that a working day is a round 8 hours. All times in the table are working hours.

Move time is booked to the station the order leaves. Of the 46 queue hours in work preparation, 40 belong to a query to the customer. One dimension on the drawing is unclear, and the answer takes five working days.

Add it up

Sum the four column totals:

8 h processing + 2 h setup + 148 h queue + 2 h move = 160 h

Then sum the station totals as a cross-check:

47 h + 23.5 h + 34 h + 39 h + 6 h + 10.5 h = 160 h

The two sums have to agree. If they don't, an hour is missing somewhere or has been counted twice.

160 h ÷ 8 h per working day = 20 working days

Work preparation releases the order on Monday of week two at 2 p.m. The frame leaves the building on Friday of week four at 3 p.m.

Process cycle efficiency

8 h ÷ 160 h = 0.05 = 5.0%

Queue time takes 148 h ÷ 160 h = 92.5%. Setup and move time take 1.25% each. Turn the ratio over and you get 160 h ÷ 8 h = 20, so the frame spent twenty times as long in the building as anyone spent working on it.

That split is more use to you than the 20 days. It says faster machines would barely move this order. Halve the processing time and you save 4 hours out of 160.

Five ways to get the number wrong

Each one is shown on the same order.

1. Adding up the work and calling it lead time

This is the eight-day answer from the first meeting. Processing alone is 8 hours, one working day. Add setup and moves and you have 12 hours, a day and a half. Round that up to eight days and it feels cautious.

Eight working days are 64 hours, 40% of the 160 that were measured. Queue time alone, at 148 hours or 18.5 working days, is more than twice the whole estimate. A buffer sized on the work can't cover that, because queue time doesn't grow with the work. It grows with queues and unanswered questions.

2. Starting the clock late and stopping it early

Take lead time from shop-floor bookings and the clock starts with setup at the saw and stops when assembly books the job complete. On this order that window is 80 hours, ten working days.

The other 80 hours fall outside it. Before the first booking there are 47 hours in work preparation and 22 hours queuing at the saw. After the last there is half an hour of moving and 10.5 hours in dispatch.

Inside the window sit 6.5 processing hours. Process cycle efficiency then looks like 6.5 h ÷ 80 h, which rounds to 8.1%, when the order's real figure is 5.0%. The customer query, the largest single item on the whole order, never shows up.

Your customer counts from the day they ordered. Put the first timestamp on order receipt and the last on handover to the carrier.

3. Mixing working days and calendar days

160 working hours are 20 working days. On the calendar, Monday at 7 a.m. to Friday of week four at 3 p.m. is 25 days and 8 hours. That is 608 hours, or 25.3 calendar days after rounding.

Quote "20 days" and the customer counts calendar days. Those run out on Sunday of week three. The frame ships the Friday after.

The ratio shifts too. 8 processing hours over 608 calendar hours rounds to 1.3%, against 5.0% on working time. You can defend either basis, but only compare figures that share one. Label every lead time as working days or calendar days, and give the customer a date.

4. Leaving "processing" undefined

Use machine occupancy as processing time and setup gets counted in. Here that is 10 hours instead of 8, and the ratio rises to 10 h ÷ 160 h = 6.25%.

A stricter reading goes the other way. Work preparation and packing don't change the frame, so a strict value-added count drops both. That leaves 6.5 hours, and 6.5 h ÷ 160 h rounds to 4.1%.

The same order scores 4.1%, 5.0% or 6.25% depending on the rule. Write the rule down once and keep it. Otherwise a change from one month to the next may only mean that somebody counted differently.

5. Quoting from the average

One order tells you little about the next. Suppose the shop's last ten completed orders took this many working days:

12, 13, 14, 14, 15, 16, 17, 20, 24, 45

The frame is the 20. The 45 was stuck on a query nobody chased.

Mean = 190 ÷ 10 = 19 working days

Median = (15 + 16) ÷ 2 = 15.5 working days

People remember the typical order, and the typical order is the median. If sales quotes three weeks from experience, which is 15 working days, five of the ten orders are late. Quoting the mean doesn't rescue it. At 19 days, three of the ten still miss.

The figure to quote is the one that nine out of ten orders meet. Here that is 24 working days, just under five weeks.

What the gap costs

An underestimated lead time is paid for in sales first. With a three-week quote, every second order in the example is late. Depending on the contract that means penalties, or else a discount on the next order. To measure the on-time share month by month, how to calculate on-time delivery works the formula through sixteen order lines.

The usual response is a safety buffer in the quote, and that costs as well. Quote a blanket eight weeks to be safe and you lose orders to faster competitors, although the frame holds one working day of actual work.

The fix

Measure first. You don't need a new system, because the handover points already exist: order receipt, release, the bookings for each operation, handover to the carrier. What is missing is a report that puts them side by side. One timestamp per handover is enough. Queue time is the gap between two timestamps, less the work booked in between.

Then go after the largest item. On this order it is the customer query at 40 hours, a quarter of the lead time. Two rules shrink it. Queries are bundled and raised within 24 hours of the technical review instead of one at a time. And every open query gets a date and an owner. After three days without an answer, somebody picks up the phone.

Suppose the answer had come after one working day instead of five. Then 32 queue hours disappear. Lead time falls from 160 to 128 hours, from 20 to 16 working days, and process cycle efficiency rises from 5.0% to 6.25%. No machine ran any faster.

Next on the list are the 36 hours queuing at the milling machine. Whether that machine is really full, or losing its time to changeovers and breakdowns, is a question for its OEE. How to calculate OEE works one shift through. Whether its cycle time keeps up with what customers call off is a separate check against takt, which how to calculate takt time works out on one shift.

Last, move the delivery promise from experience to the distribution. Promise the figure that nine out of ten orders meet. On the example that stretches the quote from three weeks to just under five, which is uncomfortable at first. But five weeks kept is worth more than three weeks missed. Once queue time comes down the quote can shorten again, this time with cover behind it.

This separation of processing from waiting is the core of a process analysis: measuring an order's path end to end with real timestamps, rather than adding up remembered durations. Fixed scope, fixed duration, fixed price.

Frequently asked questions

What is included in manufacturing lead time?

Every hour from order receipt to handover to the carrier, split into processing, setup, queue and move time. For the frame in the example that comes to 8 + 2 + 148 + 2 = 160 working hours. Queue time is the part estimates leave out, and here it is 148 of the 160. A lead time taken from shop-floor bookings starts at the saw and stops at assembly, which cuts this order to 80 hours and drops the customer query entirely.

What is the difference between lead time and cycle time?

Lead time is how long one order spends in the building. Cycle time is how often a single workstation hands on a finished part. You time cycle time with a stopwatch at one station, and you take lead time from the timestamps at order receipt and dispatch. A short cycle time tells you nothing about the queue in front of the station. The frame got 1.5 hours of milling and waited 36 hours for it. How to calculate takt time sets cycle time against takt on one shift.

Does lead time include weekends?

A lead time in calendar days includes them, one in working days doesn't, so always say which you mean. The frame took 20 working days. On the calendar, from Monday 7 a.m. to Friday of week four at 3 p.m., that is 25.3 days. A customer who hears "20 days" counts calendar days and expects the frame on the Sunday of week three, five days before it ships. Quote a date and the question goes away.

How do you calculate lead time in value stream mapping?

Count the inventory in front of each station on one day, divide each count by daily demand, and add up the days. That gives lead time in working days without following an order through. Processing time goes on the map as well, but next to weeks of waiting it barely registers. The worked value stream mapping example arrives at 30 working days for a machined flange, against 430 seconds of processing. Following one order with timestamps, as this article does with the frame, measures the same thing directly.

How do you reduce manufacturing lead time?

Find the largest block of queue time and shorten that first. On the frame, halving every processing time would save 4 hours out of 160. The customer query in work preparation held the order for 40 hours. Had the answer come after one working day instead of five, 32 hours would have gone and lead time would have dropped from 160 to 128 hours. Two rules get there. Raise all queries together within 24 hours of the technical review, and give every open query a date and an owner.

The next step

Take your last twenty completed orders and count calendar days from order entry to dispatch. Put the median, the mean and the figure that 18 of the 20 meet next to what you quote. You don't need a project for this, just an hour and your own data.

Then follow one order through its stations the way the example does, and fill in the four columns.

If you want to know at which station the waiting builds up, this worked value stream mapping example shows how to count the inventory between stations and convert it into days. The same sum for the warehouse, stock divided by consumption per day, is in how to calculate inventory turnover.