When it comes to getting value from Business Intelligence one of the capabilities often cited is that of being able to alert the business community about some significant business performance event; an alert being a directed proactive notification to one or more recipients, rather than just a status or conditional formatting displayed on a screen a business person chooses to visit. Unfortunately, not enough has been written about this important topic, or if it has then I haven’t been able to find it other than some good content from a niche consulting company in the US (Virginia/Georgia) at the site www.juiceanalytics.com.
Some of the points worthy of discussion on alerting are to do with identifying who you actually alert, how do you alert them, and, perhaps most importantly, when do you alert them?
Let’s deal with the last point first. When should the business community be alerted?
Perhaps strangely this is not as obvious as it might seem. The answer might appear to be in the opening sentence of this discussion “(when) some significant business performance event (occurs)”. The trick is to identify “significance”. Is something significant if there is a 500% increase (or decrease) in some measure, or if an “actual” varies from “plan” by 150%. The answer might be yes, but just as easily, the answer might be no. Take the case that some measure increased by 500%. If the baseline for that measure, which let us say measures the occurrence of some business activity, was that something that only happened once (or twice) and now it had happened 5 (or 10) times? Is it statistically a 500% increase? If so, that does not necessarily determine if in fact it is significant to the business.
If the event was something of expected low frequency, such as a natural disaster, or maybe a terrorist attack, then 500% is hugely significant. ,if the event is something like visits of the same child to a hospital, then it is necessary to consider other factors, such as whether the visits were for the same or different causes, the age and background of the child, visits by other children in the family, before determining if this pattern is truly an outlier worthy of alerting that a child is at risk. Or, if the event was tracking shopper traffic to some store, it may be necessary to identify other factors, such as how long the store has been opened, and thus whether this is just a “novelty” effect of a new store in a mall.
If an actual varies from plan, say by 150% this would seem to be significant. But is it really? How sound was the plan? It is not unusual when rolling out new BI/Performance Management solutions, which integrate elements of Planning, Budgeting and Forecasting, that the plan numbers may take time to “settle down”, especially if new planning processes have been recently introduced.
Business rules for defining alerts must consider more than just a mathematical statistical change. They must be founded on business understanding to ensure true significance.
If the business community is over-alerted because of poorly thought out alert criteria, which consider only simplistic statistical change, they will quickly lose confidence and come to ignore those alerts. Any value the alerts may have had will have been lost for a significant period, perhaps forever.
The alternative is to under-alert, at least to begin with. Although this sounds risky, it is likely more effective in the long-run. As a business community experiences real business scenarios where an alert could have enabled pre-emptive or responsive action, then the requirement, alert criteria, frequency, timing, target alert recipients, alerting mechanism will be clearer.
This brief discussion only touches on a few of the considerations for creating an effective approach to alerting. A well thought out approach for alerting is as important as a good approach for every other design element of a good BI, Analytics, Performance Management application.
Learn more about IBM Scorecarding solutions by clicking here.