Key performance indicators (KPIs) first appeared in the 1990s with the aura of mystery that accompanies any new idea. The mystery has gone, and now we all know now that KPIs have no rigorous definition, but are just metrics that summarize the state of a business as well as possible. They can sometimes be more of a hinderance than a help, particularly if we have too many of them, and so some skill is needed to create a set of KPIs that give the broadest possible picture using the fewest possible metrics.
Traditional KPIs are historical measures of performance, that often have some level of lag associated with them. This was the best we could do using traditional BI platforms. Sometimes the lag is not an issues, and sometimes it is. The new generation of BI platforms offer wholly new possibilities for KPIs, and specifically the ability to predict future values, find correlations between KPIs, and create near real-time dashboards displaying KPIs.
Predictive KPIs estimate the value of a KPI for the current and possibly the next period. Any number of methods can be used to do this, the simplest being linear regression – although in most instances linear regression will be too simplistic. Various BI suppliers have added forecasting tools, so that predictive KPIs can be created.
Correlated KPIs can give valuable insights, but are most useful when the correlation indicates that one of the KPIs is leading, and is in some way predictive. A good example might be a customer sentiment KPI leading a KPI measuring sales. Care is needed when looking for correlations because correlation does not mean causation. In other words two metrics can be correlated without any causal relationship existing at all. Usually the cause is ‘lurking’ in the background, and unseen until additional analysis is performed.
Finally it should be obvious in most businesses that the timeliness of KPIs is of increasing importance. A weekly update just might not cut it any more, and with the Internet of Things appearing on the horizon, many businesses will need to create ‘real-time’ KPIs – the amount of traffic on your web site for example.
Data exploration and discovery is key to uncovering meaningful KPIs, and the primary method for finding correlated KPIs and those which might have predictive value. This is often called discovery analytics, as opposed to the descriptive analytics commonly associated with traditional BI. Using data exploration and discovery we answer questions even when we do not know what we do not know. This is in contrast to traditional BI where we only seek to know what we know we don’t know.
The dumb KPIs that are characterized by lag and very little added value, will soon be replaced by smart KPIs that are timely and contain information on related KPIs and future performance.