The standard guidance for setting up a control chart is to collect 25 batches of data before calculating your control limits. This number comes from statistical theory: with 25 subgroups, the estimate of process variation is stable enough to produce reliable limits.

But for a small manufacturer running 10–15 batches per month, waiting for 25 batches means waiting two months before you can start monitoring your process. And if your process is already drifting, two months is a long time to wait.

So what is the practical answer?

What control limits actually represent

Before addressing the sample size question, it helps to be clear about what control limits are and what they are not.

Control limits are calculated from your process data. They represent the range of variation your process naturally produces when it is operating consistently. A point outside the control limits is a signal that something has changed — not that the product is out of specification.

This is the most important distinction in statistical process control. Control limits are about process consistency. Specification limits are about product acceptability. A process can be in control (consistent) but not capable (unable to meet specification). It can also be out of control (inconsistent) but currently producing acceptable product — a situation that will eventually resolve itself in the wrong direction.

The 25-batch rule: where it comes from and when it applies

The 25-batch recommendation assumes you are setting up a Shewhart control chart (an X-bar and R chart, or an individuals chart) and that you want your control limits to be statistically reliable from the start.

With fewer than 25 subgroups, the estimate of process standard deviation is less precise, which means your control limits are wider than they should be — and you will miss signals that a properly calibrated chart would catch.

This matters most when:

• Your process is already well-understood and you are setting up a permanent monitoring system
• The cost of a false alarm (investigating a point that turns out to be within normal variation) is high
• You have the luxury of waiting for sufficient data

When you cannot wait for 25 batches

In practice, small manufacturers often cannot wait. You may be:

• Setting up monitoring for a new product or process
• Responding to a customer complaint that requires immediate action
• Trying to demonstrate process control to a regulatory body or customer audit

In these situations, you have options.

Start with provisional limits. Calculate control limits from whatever data you have — even 10–15 batches — and treat them as provisional. Plot new batches as they come in, and recalculate the limits after every 5 additional batches until you reach 25. This gives you a working chart immediately while acknowledging that the limits will improve over time.

Use a wider detection threshold. With fewer data points, your control limits are less precise. You can compensate by using additional Western Electric rules (runs of points on one side of the centreline, trends of consecutive points in one direction) rather than relying solely on points outside the 3-sigma limits. These rules are more sensitive to sustained shifts and trends, which are the most common failure modes in small-batch manufacturing.

Focus on the centreline first. Before worrying about control limits, establish whether your process average is stable. A simple run chart — plotting each batch result in time order — will show you whether the process is trending, cycling, or jumping. This requires no statistical calculation and can be done with any data you have.

What to do when your process is not stable

The 25-batch rule assumes your process is in a state of statistical control when you collect the baseline data. If your process is drifting or unstable, the control limits you calculate will be too wide — they will incorporate the variation caused by the instability, and the resulting chart will be insensitive to future problems.

Before setting up a control chart, you need to understand whether your process is stable. If it is not, the first priority is to identify and remove the sources of instability — not to set up a chart that will mask them.

This is where batch variation analysis (discussed in our previous post) is useful. Identifying the root causes of variation and addressing them before setting up statistical process control gives you a much more sensitive and reliable monitoring system.

How many batches for a useful chart: a practical guide

Here is a practical framework based on process maturity and risk:

New process, high risk (pharmaceutical, food safety, regulated manufacturing): Wait for 25 batches before finalising control limits. Use provisional limits in the interim and treat any out-of-control signal as requiring investigation regardless of whether the product is in specification.

Established process, moderate risk (industrial coatings, adhesives, plastics): 15–20 batches is sufficient to establish working control limits. Recalculate after 25 batches and document the revision.

Established process, lower risk (bulk chemicals, non-critical industrial products): 10–15 batches can provide a useful starting point. Be explicit about the provisional nature of the limits and plan to revise them.

Very small batch count (fewer than 10): A control chart is not yet meaningful. Use a run chart to monitor trends and focus on understanding the process before attempting statistical control.

The limits of control charts

Control charts are a monitoring tool, not a diagnostic tool. They tell you when something has changed; they do not tell you what changed or why.

When a point falls outside the control limits, the chart has done its job — it has flagged a signal. The investigation that follows is where the real work happens. That investigation requires process knowledge, not statistics.

This is why control charts work best when they are part of a broader quality system that includes documented process parameters, raw material specifications, and a structured approach to out-of-specification investigation. The chart is the alarm. The quality system is the response.

Getting started

If you are setting up control charts for the first time, start simple. An individuals chart (plotting each batch result individually) is easier to implement and interpret than an X-bar and R chart, and is appropriate for most small-batch manufacturing situations where you have one measurement per batch.

Collect your baseline data, calculate your provisional limits, and start plotting. The chart will tell you things about your process that you did not know — and that knowledge is the foundation for systematic improvement.