Analyse

you have unknown unknowns

If you need to take a step back and look at how the processes that you already have are functioning, a good place to start would be to create a Linkage of Process (LOP) map for the part of the organisation that concerns you. This is a high level view of the main processes, the linkages between them and where appropriate, the information that flows between them. The next stage would then be to rate each of the processes for their current condition and add the impact each has on the business. This simple exercise can indicate which processes need immediate attention and which can be left until later.

Even where you may have good data, such as control charts or defect counts, an LOP is still useful for understanding the bigger picture and how one process may affect others.

An LOP doesn't take long to create, but allows you to discover how your system of processes are linked, so that when you change an individual process, you can maintain those linkages and not break the whole system. For clarity, they are often divided into three sections of Driver, Core and Support, although this is not always necessary. What matters is that the result is simple to understand and reflects reality.

Below is an example of an LOP for a sales/service organisation, where various processes have been rated for condition and impact.

The rating is typically subjective, unless data is available to provide some guidance, but it's a good place to start. For each process, a scale of 1-5 is applied to both criteria, and the sum calculated.

A high score indicates a process in poor condition (for example, poorly defined and with variable outcomes) together with a high impact (for example, critical to organisation’s objectives), meaning it should be top of the list of potential improvements.

Following this initial rating, it might also be useful to test against other criteria such as changeability, for example if a process is owned by another department and cannot easily be changed, and customer impact, if there is a direct impact to the product or service being supplied to the customer.

Another factor to consider is the volume of traffic through a process. Other factors being equal, there is usually more benefit to be found by making a small improvement to a high volume process than in making a larger improvement to a low volume process. Similarly, looking at time and resources required at each process might be more appropriate, especially where there is a resource constraint in the organisation.

Another example of an LOP is for manufacturing, where no rating has yet been applied. In this instance each of the numbered blocks is based on a standard framework, with each containing a number of sub processes.

The exact format and content of an LOP will depend on your organisation and on what aspect of it you want to focus. It should just include what is relevant.

You suspect A specific process

Before embarking on a formal improvement process, it may be worthwhile obtaining some appropriate metrics to see if your suspicions are correct. To some extent, what these are will depend on the type of process, but you can’t go wrong by first seeing if your process is stable and under control. This is important as a process must be stable and under control before any reliable improvements to performance can be made.

A word on variation might be useful at this point. All processes produce a degree of variation in their output. Nothing is ever perfect. This is called common cause variation. When something exceptional occurs, which is not caused by normal operation, this is called special cause variation. Not very original naming but there it is.

Stability can be checked with a control chart, which is simply a plot of your specific process parameter over a time period with an overlay of some statistically important limits and rules, which can indicate instability, if it is present. A control chart can be created with native spreadsheet software, but it’s hard work and best done with a Statistical Process Control (SPC) software application. As an example, the following control chart (one of a pair) shows instability with red/diamond plot points. Some of these may not look erratic but there are a number of statistical rules that the data points must follow, in order to be stable. In the example shown, two rules have been broken, one for the string of 8 consecutive points below the centre line and the other for a single point above the Upper Control Limit (UCL). For more information on control charts, have a look in the glossary.

If the source of this instability can be easily traced to some event, such as a particular batch of raw material, a faulty machine or a software change for example, and which has now past, a new chart with current data may indicate a stable process. If not, then further analysis will be needed to discover the underlying cause. Similarly, if you have a stable process which is not producing the output that you require, with either too much variation or a too high/low mean, then further analysis will be needed.

When you have determined where improvements to your processes should focus or gaps need to be filled, then it may be time for some process improvement projects.

Statistical Analysis

Whilst the control chart uses statistical tests, there are a number of additional tests which can be used to explore process data, often obtained from research or from designed experiments. They aim to draw inferences, usually from samples, as to various aspects of the data such as means, variances and potential relationships. An example of use in manufacturing would be to compare different quantities and grades of raw materials to determine if they produce the same quality output, and in services to test the efficacy of different queue structures in a call centre. I have added a summary of some of these tests and how they can be interpreted, in the glossary.