Line balance
Our demand swings by half. Does takt time still apply?
Takt assumes steady demand and almost nobody has it. The answer is not to abandon it, and it is not to average the year. It is to pick the window you are willing to be wrong in.
Takt time divides the available working time by the demand in that time, and it gives the beat a line has to run at to keep up. It is one of the most useful numbers in operations, and it is derived from an assumption that is almost never true: that demand in the period is a number rather than a range.
So the question comes up constantly, and it usually gets one of two bad answers.
The two bad answers
Average the year. Demand between four hundred and eleven hundred a month, so call it seven hundred and fifty and set takt from that. The line is then wrong in both directions: it cannot meet the peak, so it misses; and it is over-resourced in the trough, so it costs. It is a number that is correct on average and wrong every single month, which is the general failure described in the average was met and a quarter of the orders were late.
Abandon it. Demand varies, so takt does not apply, so we will staff by feel. This throws away the one number that connects customer demand to line design, and replaces it with an argument.
Neither is necessary, because the real question is not what demand is. It is what window you are prepared to be wrong in.
Recomputing on a window
Takt is not a constant of the process. It is a ratio between available time and demand, and both sides can be recalculated as often as you can act on the answer.
| Available time, a month | 9,600 min | |
| Annual average demand, a month | 750 | |
| This month, actual demand | 1,100 | |
| Takt at the annual average | 12.8 min | |
| Takt this month, at real demand | 8.7 min |
Illustrative figures. A line built to a beat of 12.8 minutes cannot meet a month that needs 8.7. The gap is not a performance problem and no amount of pressure closes it: the line was designed for a different number.
The useful move is to compute takt for the shortest window you can actually respond to. If the roster can change monthly, compute it monthly. If it can change weekly, weekly. Computing it for a window shorter than you can act in produces a number that is accurate and useless.
The three ways to absorb the swing
Once takt is a range rather than a number, the design question becomes which lever moves. There are only three, and every operation uses some mixture.
Move the available time. Overtime in the peak, short weeks in the trough. Takt is a ratio and the top of it is a real lever, not just the bottom. This is usually the cheapest response to a swing that is genuinely temporary.
Move the resource. Staff the line to the peak takt and accept idle time in the trough, or staff to the trough and accept missing the peak. Most operations do a version of this without deciding it, which is how you end up with a line that is neither.
Move the demand. Build to stock ahead of a known peak, quote longer lead times when full, or schedule discretionary work into the trough. This is the most powerful of the three and the one furthest outside the operation’s own control, which is why it is raised last and should be raised first.
When the swing is a pattern rather than noise
Before designing for variability, check whether it is variability at all. A demand that is high every March and low every August is not variation, it is seasonality, and it is forecastable. Takt should be recomputed per season and the line planned against a calendar.
The test is simple: plot two or three years by month on the same axes. If the shapes line up, you have a pattern and should plan for it. If they do not, you have genuine variation and should design for a range. Treating a pattern as noise means being surprised annually by the same event.
What to do with it
State the window every time you quote a takt. “Takt is 8.7 minutes” is incomplete. “Takt is 8.7 minutes for this month at 1,100 units” is a number somebody can check and act on.
Compute takt at the peak and at the trough, and design knowing both. One number hides the range that the design actually has to survive.
And check for seasonality before designing for variability. They look identical in a single year of data and they need completely different responses.
Takt still applies. It is just not one number, and it never really was.