Heijunka / Production Levelling: Size the Gain Before You Commit to a Levelled Week

Heijunka levels workload volume and mix over time to smooth flow and reduce variability in the future state. That is the whole job. You take the demand pattern you actually face and you ask what the week would look like if the work arrived in even portions instead of lumps.

The tool sits in the Improve phase of RDMAICS. By the time you open it, you have already recognised a problem, defined scope, measured the baseline and analysed the cause. Heijunka is not where you discover that Monday is overloaded. It is where you decide what to do about it and how much the fix is worth.

Reach for it when your analysis points at arrival pattern rather than capability. If the process is capable but the queue swings, levelling is a candidate. If the process cannot do the work correctly at any volume, levelling will only spread the defects more evenly. Check that first.

The output is not a plan. It is a sized change. You leave with a target volume per bucket, a mix rule, and a per bucket movement figure you can argue about with the people who will run it.

Everything below uses the sample data that ships with the tool. The case is fictional. It exists so you can see the mechanics without needing your own numbers in front of you.

The picture is a five day week in a single work stream. Monday carries 140 units. Tuesday drops to 60. Wednesday sits at 95, Thursday at 70, Friday at 85. Five buckets, one row each, and a spread of 80 units between the busiest and the quietest day.

That shape will be familiar. Work piles up at the start of the week, the team burns through it, then the back half runs light while people wait. Overtime on Monday and idle capacity on Tuesday are two symptoms of the same cause.

The levelled target in the sample is 90 units for every day. The mix pattern for every bucket is recorded as standard plus exceptions, meaning a repeating base sequence with a defined allowance for work that does not fit the standard.

The worksheet is editable and each row is one time bucket. In the sample the buckets are weekdays. They could be shifts, half days or weeks. Pick the smallest bucket you can actually schedule to, because a levelled plan you cannot act on is just a chart.

Current Volume is the observed load per bucket. Use real counts from the Measure phase, not an average and not a forecast. The value of the tool comes from the gap between what happens and what you intend, so the left hand column has to be honest.

Levelled Volume is your proposed target. In the sample it is 90 across all five days. Mix Pattern records how the work inside the bucket will be sequenced. Standard plus exceptions is the entry used on all five rows here, which keeps the rule identical across the week.

Variability Reduction is the per bucket movement between current and levelled. The tool shows minus 35 per cent on Monday, plus 50 per cent on Tuesday, minus 5 per cent on Wednesday, plus 29 per cent on Thursday and plus 6 per cent on Friday. Blanks are excluded from the parse, so a row you are unsure about can be left empty rather than guessed.

The result is a comparison of Current Volume against Levelled Volume across the five buckets, built directly from the worksheet. The chart axis runs from 0 to 140, which is set by the Monday peak. The flat levelled line at 90 cuts straight across the sawtooth of the current pattern.

Read the shape before the numbers. Monday towers above the line. Tuesday sits well below it. Wednesday almost touches it. Thursday falls short again and Friday sits just under. That is a two day problem dressed up as a five day problem.

Now read the per bucket movements. Monday at minus 35 per cent and Tuesday at plus 50 per cent carry almost all the change. Wednesday at minus 5 per cent and Friday at plus 6 per cent are close to noise. Thursday at plus 29 per cent sits in between.

The tool also offers an AI review that explains the result in plain language. Use it as a second opinion on your reading, not as a substitute for walking the chart yourself.

The first decision is whether to pilot at all. You now know the size of the move. Monday has to shed 35 per cent of its load and Tuesday has to absorb 50 per cent more. If neither of those is feasible with current staffing and equipment, the pilot does not start and you go back to Analyse.

The second decision is scope. Because Wednesday and Friday move by 5 and 6 per cent, you can run a narrower pilot that touches Monday, Tuesday and Thursday only. That is less disruption for most of the movement. A narrow pilot is also easier to reverse if it fails.

The third decision is the mix rule. Standard plus exceptions is recorded on all five rows, which means the exception path has to work every day, not just on the heavy day. If exceptions currently get absorbed by Monday overtime, levelling removes that absorber and the rule has to stand on its own.

The fourth decision is what you will measure in Control. The levelled volume of 90 is your target. The per bucket movements are your leading indicators. If Monday is still running above 90 after two weeks, the pull mechanism is not holding and you have a sustainment problem rather than a design problem.

The most common error is setting the levelled volume by feel. In the sample it is 90 on every day, which is a deliberate flat target. If you pick a number that does not reconcile with the total work the week has to clear, you have built a plan that fails arithmetic before it fails in practice.

The second error is treating every per bucket percentage as equally important. A minus 5 per cent on Wednesday is not a change programme. Teams that chase all five numbers spread effort thin and lose the two days that matter.

The third error is levelling volume and ignoring mix. The Mix Pattern column exists because an even unit count made up of wildly different work is not level at all. Standard plus exceptions is a rule. Write down what counts as an exception before the pilot starts.

The fourth error is changing the worksheet mid pilot. If you adjust the current volumes because a week ran oddly, you lose the baseline you are measuring against. Record the odd week as a note and leave the numbers alone.

It will not tell you whether 90 is the right target. The flat 90 in the sample is an input you supply, not an output the tool derives. The chart will draw whatever line you type.

It will not tell you how to move the work. Shifting 35 per cent off Monday and adding 50 per cent to Tuesday requires a mechanism, which might be scheduling, order release, a buffer or a conversation with whoever sends the work. None of that appears in the worksheet.

It will not price the change. There are no cost, hours or savings figures here. The per bucket percentages tell you how far the load moves, not what the movement is worth. If you need a business case, you build it alongside this, using the movement as the input.

It will not prove the gain is real. The comparison is between an observed pattern and a proposed one. Only a pilot produces evidence. Treat the output as a sized hypothesis that is worth testing, and carry it into Control with the measures already named.

Finally, it will not fix a capability problem. If the process produces defects at 85 units a day, it will produce them at 90. Levelling smooths arrival. It does not change what happens once the work arrives.