Takt time, cycle time and lead time: the three numbers people mix up

Takt time, cycle time and lead time get used interchangeably in conversation and they should not be. They are measured differently, they are owned by different people, and they answer three unrelated questions. Confusing them produces two characteristic failures: lines balanced to a rate nobody actually needs, and improvement projects that shorten a process by seconds while the customer waits exactly as long as before.

This guide defines each precisely, carries one example through all three, and then deals with the measurement problem that makes most of the arithmetic worthless if you skip it.

Three numbers, three different questions

• Takt time answers: how often must a unit come off the end of this process to satisfy demand? Set by the customer and your working calendar. No machine appears in the calculation.
• Cycle time answers: how often does a unit actually come off? Set by the process — by the slowest station, on a line. Measured with a stopwatch, not derived from a plan.
• Lead time answers: how long does the customer wait? Measured from request to satisfaction, and dominated by waiting rather than working.

The relationships are simple to state and easy to forget. Cycle time must be at or below takt or you cannot meet demand. Lead time is almost always vastly longer than cycle time, because most of it is queueing. And improving cycle time changes lead time only if cycle time was why things were queueing — which, on a line running below capacity, it usually is not.

Takt time: the rate demand sets

The formula, exactly as published:

• Takt time = available production time ÷ customer demand
• (both over the same period, in the same units)
• Here: Takt (seconds) = (available minutes per shift × shifts per day × 60) ÷ demand per day

The second line is the one people skip and the one that causes the errors. Available time and demand must cover the same period. A weekly demand figure divided by a daily available time gives a number with no meaning, and because it still looks like a plausible quantity of seconds, nobody catches it until the line is built.

Available time is not shift time

Available production time is time the process could actually be producing. Everything that is scheduled not to produce comes out before the division:

• Breaks and meal breaks, when the line genuinely stops rather than being covered.
• Shift start-up, handover and end-of-shift cleandown.
• Planned maintenance windows and scheduled changeovers.
• Team meetings, training and any other standing commitment on the calendar.

What does not come out is anything unplanned. Breakdowns, material shortages and quality stops are losses to be measured and reduced, not deductions from available time. Removing them makes takt look longer, which makes the line look adequate, which is precisely the reassurance you do not want. A takt time built on optimistic available time will be met on paper and missed every week.

Demand must be real customer demand

The other half of the division is just as easy to corrupt. Real customer demand is what customers ordered over the period, levelled across the period they want it in. It is not what the plan says, not what the forecast hopes, not what the previous line was capable of, and not last month's output — using output as demand guarantees takt matches current capability and tells you nothing.

Two adjustments are legitimate. If demand is seasonal, calculate takt for each period you plan to hold a stable rate; a single annual average describes a year nobody works. And if you build a buffer or a scrap allowance, add it to demand explicitly and label it, so the next person can see what customers wanted and what you chose to make.

A worked example, carried through

Step 1 — takt time

An assembly line runs two shifts a day. Each shift is 480 minutes long, with 30 minutes of breaks, a 15-minute start-up and handover, and a 15-minute end-of-shift cleandown — so 60 minutes come out and available production time is 420 minutes per shift. Levelled customer demand is 800 units per day.

Takt (seconds) = (available minutes per shift × shifts per day × 60) ÷ demand per day = (420 × 2 × 60) ÷ 800 = 50,400 ÷ 800 = 63 seconds.

One unit must come off the line every 63 seconds. That number now governs every design decision downstream: how many stations, how the work is split, and how much spare capacity you have if demand rises.

Step 2 — measuring cycle time honestly

How you measure cycle time decides whether everything built on top of it is real. Time several consecutive cycles at each station — ten is a workable minimum — and record every one, including the interrupted ones. Then use the repeatable time: what the station takes when the work is done the normal way, which for a well-behaved station is close to the mean.

Here are ten consecutive readings at station 1, in seconds: 58, 60, 57, 62, 59, 61, 58, 60, 63, 62. They sum to 600, so the mean is 60.0 seconds. The deviations from the mean are −2, 0, −3, 2, −1, 1, −2, 0, 3, 2; their squares sum to 36. Dividing by n − 1 gives a variance of 36 ÷ 9 = 4.0, so the standard deviation is exactly 2.0 seconds.

The repeatability test is the coefficient of variation, published as: CV % = (standard deviation ÷ mean) × 100. For station 1 that is 2.0 ÷ 60 × 100 = 3.33%. This station does the same thing every time, and 60 seconds is a number you can build a line on.

Now station 4, ten readings taken the same way: 48, 72, 55, 90, 61, 52, 78, 64, 58, 82. They sum to 660, so the mean is 66.0 seconds. The squared deviations sum to 1,746, so the variance is 1,746 ÷ 9 = 194.0 and the standard deviation is about 13.93 seconds. CV % = 13.93 ÷ 66 × 100 = 21.1%.

Station 4 has a mean above takt, which is a problem on its own. But the more important finding is the spread. Five of the ten observed cycles took longer than the 63-second takt and five took less. The station does not have a cycle time in any useful sense; it has a distribution, and the distribution is wide enough that no single number describes it.

Step 3 — what a high CV means before you balance anything

A coefficient of variation above roughly 15% is the signal to stop and fix the variation before doing any balancing arithmetic. The reason is structural rather than a matter of taste. Balancing distributes work so that every station finishes within takt. If a station's actual time swings between 48 and 90 seconds, then whatever number you assign it will be wrong most cycles: assign the mean and it overruns half the time, assign the worst case and you have built idle time into every station on the line to protect against one of them.

Variation also propagates. On a connected line a station that overruns starves everything downstream and blocks everything upstream, so one unrepeatable station degrades stations that are themselves perfectly stable.

The causes are usually mundane and findable by standing at the station for an hour: two operators using different methods with no agreed standard, parts that sometimes need persuading, a fixture that occasionally has to be reset, a check performed only when someone remembers. Fix the cause, re-time, and confirm the CV has come down before going near a balance chart. Here, once the fixture on station 4 was reset at the start of each run, the station settled to a repeatable 60 seconds.

Step 4 — balancing against takt

With repeatable times in hand, takt becomes the design constraint. The published formulas are:

• Line efficiency = Σ element time ÷ (stations × longest station time) × 100
• Theoretical minimum stations = ⌈Σ element time ÷ takt⌉
• Balance loss = 100 − line efficiency

Total work content on this line is 268 seconds. Theoretical minimum stations = ⌈268 ÷ 63⌉ = ⌈4.25⌉ = 5. The ceiling matters: 4.25 stations is not available, and four stations would need each to average 67 seconds, above takt. Five is the floor, and it is a floor rather than a plan — perfect division of work between stations is rarely possible because elements cannot always be split or reordered.

The actual balance across five stations comes out at 60, 58, 55, 52 and 43 seconds, which sums to 268. The longest station is 60 seconds, so line efficiency = 268 ÷ (5 × 60) × 100 = 268 ÷ 300 × 100 = 89.3%, and balance loss = 100 − 89.3 = 10.7%. That 10.7% is the waiting time built into the line by the uneven split — roughly 32 seconds of idle time across the five stations every cycle.

The line's cycle time is its slowest station: 60 seconds, inside the 63-second takt. Over 50,400 available seconds a day at 60 seconds per unit the line can deliver 840 units against demand of 800. That margin is thin — one station drifting three seconds slower leaves none at all, which is why the 43-second station is where you look first when demand rises.

Step 5 — lead time

Lead time is the customer's number, and it is measured on a calendar rather than a stopwatch. The published definitions, stated in flow terms, are: Lead time = date finished − date raised (the customer's experience); Cycle time = date finished − date started (your process).

Follow one order through. It arrives on day 0 and is scheduled into the next available slot. Material is picked and staged on day 3. The unit is built on day 4, taking 60 seconds at the constraint station and a little over four minutes of total work content. It then sits in finished goods until day 8, when the consolidated shipment leaves, and reaches the customer on day 11. Lead time is 11 days. Total process time is 268 seconds — under five minutes.

That ratio is the lesson, and there is a published measure for it: PCE = total process time ÷ total lead time × 100. Almost all of the eleven days is queueing — waiting for a slot, waiting for material, waiting for a shipment. If you cut the constraint station from 60 seconds to 50 seconds, an excellent piece of work, the customer's lead time goes from 11 days to 11 days. The wait was never caused by the work.

Shortening lead time means attacking the queues: scheduling more often, releasing work in smaller batches, shipping more frequently, removing the approval that runs once a week. None of that appears on a cycle time study, which is why lead time needs measuring separately, by someone looking at dates rather than seconds.

What each number does and does not tell you

Takt time

Tells you the rate you must sustain, and gives every design decision one reference point. Does not tell you whether you can achieve it or how much capacity is in reserve. Takt is an instruction, not a measurement — if it comes out shorter than your best station can run, no amount of recalculation fixes that.

Cycle time

Tells you what the process actually does, and identifies the constraint: the slowest station sets the line's cycle time and is the only station where an improvement changes output. Does not tell you whether that output is enough, does not describe the spread unless you report the CV alongside it, and says nothing about how long a customer waits. A cycle time quoted without a sample size and a CV is an anecdote.

Lead time

Tells you the customer's experience and exposes how much of your process is waiting rather than working. Does not tell you where the waiting happens, and does not respond to cycle time improvements unless the constraint is genuinely the cause.

The short version

Calculate takt from real demand and honestly available time. Measure cycle time by timing several cycles and reporting the mean with its coefficient of variation, never the fastest you saw. Fix any station above roughly 15% CV before balancing. Then balance to takt, and measure lead time separately on a calendar — the number your customer cares about is mostly waiting, and no stopwatch will find it.