Takt Time, Cycle Time and Lead Time Are Three Different Clocks

Takt is demand wearing a stopwatch

Takt time is available production time divided by the number of units the customer requires in that time. The subtlety is entirely in the numerator. Available time is not the length of the shift; it is the shift minus everything the plant has already agreed not to produce during. An 8-hour shift is 480 minutes, but if 30 minutes go to breaks and 20 to a planned changeover, the line has 430 minutes, or 25,800 seconds. Divide by a requirement of 860 units and takt is 30 seconds: one unit must leave the line every half minute for demand to be met.

Two things follow that are easy to say and hard to internalise. Takt is not something a team can be asked to achieve faster; it is arithmetic on demand and on hours, and the only ways to change it are to change the order book or to change how long the plant is open. And takt has nothing to say about whether the process can meet it — that is the second clock's job. A takt of 30 seconds is a target, not a measurement.

Cycle time is capability, and the bottleneck owns it

Cycle time is how long the process actually takes to produce one unit. On a single machine that is the machine's time. On a line it is the time of the slowest station, because a unit cannot leave the line faster than the slowest step lets it through. Our six stations run at 26, 28, 34, 25, 29 and 27 seconds. The total work content is 169 seconds — that is how much labour a unit absorbs — but the line's cycle time is 34 seconds, the time of station three.

Compare the two clocks and the situation resolves itself. Cycle time 34 seconds against a takt of 30 seconds means the line yields 25,800 ÷ 34 = 758 units in the shift, 102 short of the 860 required, which is 11.8% of demand unmet. No exhortation closes that gap, because it is not an effort problem: the physical process cannot pass units through faster than its slowest station. The rule is short. If cycle time exceeds takt you cannot meet demand, whatever anyone does. If cycle time is far below takt you have spare capacity, and if you keep the line running anyway you are building stock nobody has ordered.

Quantify the second half so it stops being a slogan. If the same line ran a 24-second cycle against a 30-second takt, it would produce 1,075 units in the shift against 860 demanded — 215 units of overproduction, 25% more than anyone asked for, made with the line idle 20% of the time it is manned. Those 215 units are cash converted into stock, plus the space, handling and obsolescence that follow them around. Running below takt is not free efficiency; it is production you have to store.

Balancing beats speeding up a non-bottleneck

Station four takes 25 seconds, the fastest on the line. Suppose a project team spends a month cutting it to 15 seconds — a 40% improvement, presentable in any steering committee. The line still delivers 758 units, exactly as before, because the bottleneck is still 34 seconds. The gain is zero. All the project has done is let station four wait longer between units.

Now move five seconds of work off station three onto station four instead. The stations become 26, 28, 29, 30, 29 and 27 seconds, the same 169 seconds of total work redistributed, and the cycle time falls to 30 seconds. Output rises to 25,800 ÷ 30 = 860 units — demand met exactly, a gain of 13.3% over the old 758, achieved without buying anything or asking anyone to work faster. The line balance efficiency, work content divided by stations times cycle time, rises from 169 ÷ 204 = 82.8% to 169 ÷ 180 = 93.9%.

The work content also tells you the floor. Divide 169 seconds by the 30-second takt and you get 5.63, so six stations is the theoretical minimum for this product at this demand; five would be arithmetically impossible. That is a useful sanity check before anyone proposes a redesign: if the plan needs fewer stations than work content ÷ takt, the plan is wrong, not ambitious.

Little's Law is the bridge to lead time

Little's Law says that in any stable system, the average number of items inside it equals the average arrival rate times the average time each item spends inside: L = λ × W. It is startlingly general — it holds for a production line, a queue at a counter, an inbox or a hospital ward — and it needs no assumption about the order in which things are served or how variable the times are, only that the system is stable over the observation window.

Apply it to the balanced line. Throughput is 860 units in 430 minutes, so λ = 2 units a minute. If 240 units are on the line and in the buffers between stations, then W = 240 ÷ 2 = 120 minutes: a unit entering now leaves two hours later. Halve the work in progress to 120 units and W falls to 60 minutes. Throughput has not moved — it is still capped at 2 units a minute by the bottleneck — and nobody is working faster. Lead time halved because there is half as much queue in front of each unit.

The floor is instructive. A six-station line with exactly one unit at each station has L = 6, so W = 6 ÷ 2 = 3 minutes — which is also six stations at 30 seconds each, the two derivations agreeing as they must. Against that floor, 240 units of work in progress means 40 units queued at every station and a lead time forty times the theoretical minimum. Put differently: of the 120 minutes a unit spends on the line, only 169 seconds — 2.8 minutes, or 2.35% — is work being done to it. At 120 units of work in progress that ratio doubles to 4.69%. Every improvement project that cuts queues rather than motions is chasing the other 97%.

The lead time the customer actually feels

The 120 minutes on the line is manufacturing lead time. What the customer measures starts when the order is placed and ends when the goods arrive, and the line is usually a small slice of it. Take a plausible chain: half a day to confirm and schedule the order, three days waiting for material, two days queued in front of the line, a quarter of a day on the line and packing, and two days in transit. That is eight days end to end, of which the line itself is a quarter of a day — 3.1%.

That proportion decides where improvement money should go. Halving the time on the line — an enormous engineering achievement — removes an eighth of a day and shortens the customer's wait by 1.6%. Halving the two-day queue in front of the line removes a full day and shortens it by 12.5%, and Little's Law says you do that by admitting less work into the queue, not by working harder inside it. First improvement projects fail here more often than anywhere else: they attack the visible motion at the workstation because that is what a stopwatch sees, while the waiting that dominates the customer's experience sits in the white space between the boxes on the process map.

One last connection worth drawing. Because output is capped by the bottleneck, the bottleneck is the only place where an hour lost is an hour lost for the whole plant, and the only place where an hour gained is worth its full value in units. That is why capacity utilisation should be measured at the constraint and nowhere else, and why the profit a shift earns is best read per bottleneck hour rather than per worked hour. A busy non-bottleneck station is not producing value; it is producing queue.

Frequently asked questions

Should cycle time equal takt time?

It should sit slightly below it. Setting cycle time exactly equal to takt leaves no room for a jam, a missing part or a slow changeover, and any of those turns immediately into a missed day. Most lines are planned at some deliberate margin under takt — the size of that margin is a judgement about how reliable the process is, not a universal constant. What must never happen is cycle time above takt, which guarantees a shortfall no matter how the shift goes, and what should not persist is cycle time far below takt, which produces stock nobody ordered.

Does takt time change when demand changes?

Yes, and that is the point — takt is the only one of the three clocks that belongs to the customer. If demand rises from 860 to 1,075 units on the same 430 minutes, takt falls from 30 to 24 seconds and a process that was comfortable is suddenly incapable. If demand falls, takt lengthens and the honest response is to run fewer hours or fewer stations, not to keep the line at its old speed and store the difference. Recompute takt whenever the forecast moves materially, and treat a change in takt as a change in the design brief for the line.

Does Little's Law apply outside a factory?

Yes, and that is its main practical value. L = λ × W holds for any stable queue: tickets in a support backlog, cases in a claims department, patients in a ward, features in a development pipeline. If a team closes 20 tickets a week and the backlog holds 200, the average ticket waits ten weeks, and it will keep waiting ten weeks however hard anyone works until either the closure rate rises or fewer tickets are admitted. That is the same lever as cutting work in progress on a line, dressed in different clothes.

How do I find the bottleneck if I do not have station times?

Walk the line and look for the inventory. Work in progress piles up immediately before the constraint and is scarce immediately after it, so the biggest queue is standing in front of the answer. Two other signals confirm it: the station that is almost never idle while its neighbours wait, and the station whose stoppages show up in the day's output within minutes. Measure that station properly, then measure it again after any change, because relieving one bottleneck usually promotes another somewhere else on the line.

Is takt time useful in an office or a service?

It is, provided the work is repetitive enough to have a unit. Claims processed, invoices posted, blood samples analysed, meals served — each has a countable output and a defined shift, so available time divided by required units gives a takt exactly as on a line. Where it breaks down is on genuinely one-off work, because there is no comparable unit to divide by. In those cases keep Little's Law, which needs only a stable throughput and a count of what is in progress, and drop takt.