The failed Risk Management practice of the ubiquitous risk matrix will finally be laid to rest in the 2020s. Vague subjective estimation of likelihoods and consequences will be replaced with Predictive Analytics objective predictions, based historical patterns and current trends, leading to informed risk based decision making.
This is Part 2 on Predictive Analytics in the series on the Top 10 Disruptive Technologies that will transform Risk Management in the 2020s. So far I have covered:
10. Scenario Analysis – to provide operational management with possible paths and outcomes for risk events
9. Big Data – to provide the collateral for analytics & risk based decision making
8. Neural Networking – to identify and map the drivers & influences on risk events
7. Predictive Analytics – Part 1 – 5 Primary Reasons for failure of Predictive Analytics projects
This week on a more positive footing, I look at how Predictive Analytics can be employed in Risk Management to provide meaningful forecasts of risk likelihood and consequences.
Misunderstanding the capabilities of Predictive Analytics
Before delving into what Predictive Analytics can be used for, it is important to truly understand the real capabilities of Predictive Analytics. Executives and managers must understand that predictive analytics involves probabilities and correlation. It should be looked on as just one input into the decision making process. Its results are the probability of an event happening under similar conditions and must in no way be considered as replacing decision making nor abdicate management responsibility for evaluating all relevant information in deciding on action and/or direction.
The Predictive Analytics Family
Although Machine Learning based analytics are the flavour of the month, as covered in my last article 5 Primary Reasons for the Failure of Predictive Analytics, until you have established an effective framework for using Predictive Analytics, you should firstly consider some more traditional Predictive Analytics methods including:
- Regression Analysis: graphing history or trends to enable projection of future values given known factors
- Predictive modelling: yields the probabilities of event occurrences based on previous event occurrences
- Bayesian Inference: the mathematics of using the influence of contributing factor to re-calculate the probability of an event/outcome as more evidence or information becomes available (one of my favourites).
- Cluster Analysis: method for classification and regression that predicts an object’s values or class membership based on the k-closest training example.
- Partial least squares. Identifies relationships between inputs and outputs and looks for factors that explain variations.
- Plus others…
As mentioned in the previous article, although unlikely to produce the serendipitous results of “black-box” machine learning methods, they have the practical advantage of their results fitting within operational management “gut-feel”, and unless accepted and used at the operational level your Predictive Analytics project will fail. Remember, in either case, they should be only one of the considerations in operational decision making.
All that said, let’s look at some practical applications of Predictive Analytics in risk management.
Threat Analysis
This use of Predictive Analytics can be extended from its obvious uses in Fraud and Cyber Risk detection, to any area of threat. A threat is a direct actor on a risk and generally, like fraud, can be identified by monitoring “normal” behaviour (via a process metric) and ascertaining “unusual” behaviour (outside normal process variation). Using Neural Networks to identify event influences and drivers; Big Data to source benchmark process metric data; and Scenario Analysis to list possible outcomes, any or a combination of the Predictive Analytics methods listed above can present management with likely outcomes of each course of action.
Compliance Surveillance Targeting
Audits, inspections, and reviews, together with monitoring CARs, mitigation, controls and improvements, can all be better targeted to minimize compliance activity time and cost. Predictive Analytics can be used to identify areas of greatest risk or most likely to fail. Using both internal history in combination with industry averages from Big Data, Predictive Analytics can also calculate confidence levels of success of mitigation or corrective activities as well as allowing management to target individual people or activities for closer surveillance.
Reduce Failure Rates
In PDCA is NOT Best Practice I detailed the use and benefits of Realization and Optimization mathematical techniques for identifying corrective action and process improvements. Conversely, these Predictive Analytics techniques can be used to identify circumstances and/or timing of likely failure. Used heavily in setting preventive maintenance programs, the same techniques can easily be applied to any process which might fail under stress (human or volume) or deterioration (complacency or elapse time). Note, both project and customer confidence success rates deteriorate as delivery/services times increase.
Other Applications of Predictive Analytics
There are as many other application of Predictive Analytics in risk management as there are points of uncertainty. Some other key areas to consider are:
- Using credit scoring techniques to assess staff suitability, supplier capability, and/or customer resource allocation.
- Sensitivity Analysis to rate the effect of risk drivers and influences on events
- Identify changes in behavioural patterns in human dependant customer or financial critical services
- Identify validity or optimum values of key assumptions in satisfying business outcomes
- Debt collection targeting those vulnerable to current economic, market, & individual factors
- Any likelihood, consequences, sensitivity, outcome, behaviour, etc., etc.
The Payoff
IBM research into firms who have used Predictive Analytics reveals impressive returns. It found that such companies reported an average of 24% greater cost efficiencies through modelling risk scenarios and optimizing processes, 21% improvement in growth as a result of reduced interruptions to product/service delivery and customer satisfaction, and better brand reputational equity.
Regardless to the specific outcomes, or which techniques you use, for a Predictive Analytics project to be effective and to add value ensure:
- They are selected by operational management to solve operational problems
- Are developed in conjunction with scenario analysis with multiple outcomes
- Use multiple sources of input metrics from a large historic dataset
- Consider producing the same analytic from 2 or more different methods to verify predictions
- Only use as collateral to operational management for human based decision making.
Next week I will look at No. 6 – “Using IoT (Intelligent Things) technology to monitor Risk and Threats in real-time”, in more detail.
Bio:
Greg Carroll - Founder & Technical Director, Fast Track Australia Pty Ltd. Greg Carroll has 30 years’ experience addressing risk management systems in life-and-death environments like the Australian Department of Defence and the Victorian Infectious Diseases Laboratories among others. He has also worked for decades with top tier multinationals like Motorola, Fosters and Serco.
In 1981 he founded Fast Track (www.fasttrack365.com) which specialises in regulatory compliance and enterprise risk management for medium and large organisations. The company deploys enterprise-wide solutions for Quality, Risk, Environmental, OHS, Supplier, and Innovation Management.
His book “Mastering 21st Century Risk Management” is available from the www.fasttrack365.com website.