Operational risk management and decision latency
B-eye network hosted a webinar this week that was interesting: Risky Decisions Translate to Risky Business - The Negative Impact of Latency in Your Data Value Chain. In it Claudia Imhoff gave a nice overview of operational risk and the role of BI in addressing it before diving into the issues of latency. One of her first points was to discuss the difference between High Frequency Low Impact (HFLI) risks and Low Frequency High Impact (LFHI) risks. Claudia argues that LFHI risks are much more serious but here I have to disagree with her. It is not that these low frequency risks are not significant or dramatic as things like Hurricane Katrina certainly are both of these, but that the cumulative impact of high frequency risks can be much more serious. If my risk models are off and I am writing policies at a lower price than I should be I am building up an ever-increasing exposure a small amount at a time. If I cannot detect small frauds then I could be racking up a huge fraud bill without realizing it. If I am over-paying on claims then I could rapidly run into problems. And unlike large unusual events I probably can't get insurance against this kind of risk the way I can against a larger, but less common, risk. Just because the amount of each risk is low does not mean that the cumulative impact is not high. You must use the standard risk approach of assessing the likelihood of risk, the amount of risk and the number of occasions on which the risk could occur. Now I do agree with Claudia Agree that BI is most useful for LFHI risks but I think HFLI risks are also important and Enterprise Decision Management (EDM) approaches are ideal for HFLI risks.
Later on Claudia discusses the need for detailed, timely, trustworthy data. This is true not justfor the LFHI risks she was focused on but also for the kind of HFLI risks I am focused on. Clearly there are differences, especially in the issue of timeliness, but these are good requirements regardless. Decision Latency is also even more of an issue - like Claudia I like Richard Hackathorn's approach. When you are focusing on eliminating latency, especially decision latency, automation is key for HFLI risks. Not only automation of alerts and notifications but automation of actions in response also - sense, predict, respond. Indeed this gives you a way to move the results of BI back into operational systems - you can automate the operational action required based on your BI results and so reduce decision latency. Actionable insight does not always mean making it usable by a person (reports etc), sometimes it means delivering it to software in the form of executable predictive analytic models and using business rules to act on the results.
Claudia also noted that one of the challenges of operational risk is the embedding of "BI" results into operational processes and systems with all the performance and other implications. Personally I don't think it's helpful to think about this in terms of embedding BI into operational systems as this makes people think about displaying reports and graphs in their transactional systems. I think instead you should worry about how to make your operational systems smarter (in analytic terms) and use technology like business rules and embedded predictive analytic models to do so (see this article for instance). The added advantage of this is that you don't have to train a whole bunch of people who use the operational system, instead you make the system smarter.