header page

Introduction to Generation and Selection Key Performance Indicator (KPI) Sets

Image of a graph This book presents an approach to generating and selecting Key Performance Indicator (KPI) sets suitable for managing business processes. The "set" part of the process recognizes that a typical business process can rarely be appropriately managed by a single KPI; rather a balanced set of KPIs needs to be carefully selected to ensure appropriate long-term management of the business process. The approach to selection of a balanced set owes a considerable debt to work done in the 1990's by Kaplan and Norton, in particular their work on the "Balanced Scorecard".

In the course of the synthesis, we will build a set of simple tools for generating and selecting a KPI set.

[...pages omitted...]

KPI Thresholds, Targets & Benchmarks

A KPI in itself is not sufficient; boundaries for performance against those KPIs need to be established. That is to say, a KPI itself is just the expression of the measure which will be used, such as "%age of enquiries responded to within two working days"; it does not, in itself, express whether a result of 75% is good, bad or indifferent; to decide the quality of the measurement, we need to consider thresholds, targets & benchmarks.


In discussing General Characteristics of KPIs, we explored the unpredictable variance of KPIs over time, with that variance measuring the extent to which factors outside of the span of control of the process can impact on its measurement.

Generally, a process should operate within a performance band which makes allowance for variability inherent in the process and its measurement.

Process performance will vary due to factors beyond our ability to reasonably predict; a week of unseasonable heavy rain may suppress the month's sales figures; industrial action in a supplier's factory may limit production of a certain line of product. To allow for these variations, the performance should be managed within an expected band. Depending on the frequency of measurement and the degree of variation imposed by the operating environment, a band of about +/- 5% may be reasonable, meaning that results of 95 to 105 units against a target of 100 units/month might be accepted without requiring investigation or explanation.

Naturally, repeated results over, say, three measuring periods may require investigation to ascertain whether they are unconnected environmental effects or indications of a trend. The relative magnitude of the band of acceptable operation depends on the degree to which the process is impacted by environmental factors outside the process span of control..

The following graphic illustrates unpredictable variance:

random variation

Consider steady state operation of a process, i.e. a situation in which the process is generally considered to be operating satisfactorily. If the desired KPI value is set at, say, 100, we can see from the preceding graphic that every month the value will probably be something other than 100. In order to avoid chasing shadows (or, to be precise, trying to control that which we have already defined as outside the span of control of the process), we need to set a reasonable range of operation, establishing upper and lower thresholds of the values which form an acceptable operating range. In the example of the graphic above, the illustrated thresholds would provide an acceptable operating range of 90 110.

This process variability performance band can initially be set by reference to the accumulated knowledge of the process variability of those who are responsible for the process. However, there is strong argument for continually reviewing this tolerance and in particular for questioning the extent to which these environmental factors are truly beyond our control. We may have no control over rainfall within a month, but it is likely that we can exert control over the extent to which rainfall dampens sales. The pursuit of reduced process variability has been extensively analysed in works on quality management and assurance, in particular the stream known as Six Sigma. For our purposes it is enough to say that the initial process variability performance band should be continually reviewed to determine whether it can be tightened.


The thresholds described in the preceding section are used to measure whether a process is performing satisfactorily in steady state operation. In practice, most organisations are seeking to improve performance of their business processes and this involves the setting of targets for improvement.

Targets for business process improvement have two equally important aspects; the new target per se and the qualifying time constraint associated with that new target. Thus a suitable target might be, "sustained performance at 110 units or greater by end Q3". Both aspects of the target need to be realistically achievable by the efforts of the Accountable party for the business process.

The following graphic illustrates the establishment of a new target, together with its target implementation date:


As can be seen in this graphic, both the new target (a level of 70) and the target implementation date (August) are made explicit. New upper and lower thresholds will need to be established; the existing thresholds are 90 and 110. A good initial point for upper and lower thresholds would be 50 and 70, in the absence of evidence that the magnitude of the unpredictable variation will decrease as a consequence of the new, lower target.

Considering first the new target itself, it must be possible for the business process' Accountable party to deliver on that target, whether through their own direct efforts or through influencing better operating conditions for the process. To illustrate, consider a common target within service departments of billable hours. Let's say the new target for a small graphic design department's billable hours is "70% of department total hours billed to customers within one year", up from the current 50%. This seems reasonable at first glance, but it merits some analysis. Allowing for four weeks annual leave, a week of holidays, say one week of sick leave, there are only 88% of total hours potentially billable. If 2 people in the department of 10 are not billable (overheads such as manager and receptionist) then only 70.4% of the department's total hours are even theoretically billable. A reasonable allowance for non-billable time spent on the pursuit of new accounts is around 10% - this leaves 63% of total hours as theoretically billable, assuming no downtime between engagements. It is clear that the basic target is not within the reach of the department's manager to achieve.

To address the issues the process owner would need to alter such basics as annual leave entitlement or policy with respect to the charging of hours spent in pursuit; even then, it is unlikely to have been the original intent of the target. If we restructure the target to more directly reflect the underlying corporate goal, we might end up with something more like "70% of department total payroll billed to customers within one year". This target value may be much more within the process owner's control. In attempting to achieve it, the process owner might set the hourly rate to incorporate the departmental overheads and entitlement costs. He or she might look at the blend of permanent staff to contract staff to see if some improvement might be achievable there. In short, expressing the target in a manner as closely reflective of the real corporate goal (cost recovery instead of hours) helps to incite consistent action, but, even then, the target needs to be carefully analysed to ensure that it is sensibly within the scope of the process owner to deliver.

Equally, the time constraint element of a performance improvement target must be reasonably achievable by the process owner's efforts.

As well as being aligned with corporate goals, KPI improvement targets need to make a material contribution to the corporate goal. Thus, where a corporate goal is, say, a 33% increase in market share over the next 2 years, it is clearly inadequate for a departmental KPI improvement target, for example, "customer satisfaction rating after 30 days" to have a 5% / annum target improvement rate; even if met, the contribution will not be material to the extent required by the corporate goal.


As well as determining thresholds and performance improvement targets, comparison of the performance of a business process with similar processes in other organisations can be helpful in deciding whether the performance level is in the right range. Benchmarks can be determined to allow comparison of local performance with industry averages or the known performance of principal competitors. For standard processes, benchmarks for acceptable or even 'world class' performance may be available either from consultancies or from publicly available data stores.

All of the caveats that apply to performance comparisons with competitors and industry averages apply here but doubly so; except in special cases, managing by KPIs is not about trying to "beat" one specific target but to drive the process so that it contributes to corporate goals on all levels. It is also important in KPI benchmarking to avoid potential pitfalls, such as.:

[...pages omitted...]

Where Next?

The complete series of Business Process Management e-books is as follows:

The series is available at the following website:

Modulus provides tools, applications and services to consultancies and website developers. For more information contact Peter Hill, peter.hill@modulus.com.au.

[...pages omitted...]