Benchmarking Dashboard.png

Situational Analyses

At times, need may arise to assess the potential impact of, or exposure to, non-recurring or otherwise atypical risks. In some situations, such informational needs may be adequately served with currently available, in-house resources - in other instances, a more specialized set of skills might be required. To that end, we offer fully customized, ad hoc data analytical services, where we leverage our data management and modeling skills and computing assets, and our risk/insurance industry knowledge to devise the best possible solution to a question at hand. Shown below are a few examples of such risk-related custom research projects.

Predicting the Likelihood of Gas Line Rapture

Natural gas is typically distributed via delivery lines that are buried in the ground - in North America, Europe, Australia, and other developed parts of the world the majority of human-inhabited areas are covered by a web of hidden-from-sight gas lines. Although gas utility organizations commonly offer free gas line demarcation services, whenever anyone engages in an activity that encompasses ground excavation, ranging from large industrial projects to simple homeowner yard work, a possibility of rapturing a gas line arises.


With that in mind, a major gas utility organization was interested in developing data-driven capabilities to anticipate and aggressively mitigate the threat posed by accidental gas line raptures. The project entailed leveraging structured (numeric) and unstructured (text) data to delineate the leading causes of gas line raptures, fit, validate and productize a multivariate likelihood of gas line rapture model, and to develop an ongoing risk mitigation strategy.


To accomplished the stated goals, the Company's own accident database, as well as partners' datasets were utilized; following extensive data re-coding, re-engineering, (text) mining, agglomeration and amalgamation efforts, statistically robust exploratory (inclusive of text mining), explanatory and predictive analyses were undertaken yielding numerous decision-guiding insights, inclusive of a highly reliable, 30 effect-coded-factors-driven predictive model. To further enhance the efficacy of the resultant data-driven risk mitigation strategy, model predictions were geocoded to enable GIS mapping display of 'high risk excavation' activities, as well as summative period risk heat maps.

One-Call Sample1.png

Identifying 'At-Risk' Drivers

A transportation and logistics organization wanted to develop proactive means of mitigating the risk of driver accidents. An array of Company-captured demographic, employment and driving behavior related factors were utilized to examine different types of avoidable driving incidents, which ultimately led to identification of two distinct types of avoidable accidents.


Following re-coding and re-engineering of the available and applicable data, a comprehensive driver risk profile was derived for each individual driver, which in turn led to the development of two separate, multi-attribute predictive models, each focused on estimating the probability of an a priori defined type of driving incident. Productized as a recurring, period risk scoring facility, the set of the two predictive models forms the basis for early intervention focused risk mitigation strategy.


Text Mining Insurance Adjuster Notes

A large insurance carrier desires to extract periodic insights out of claim handlers' electronic notes systems, with the goal of identifying specific behavioural patterns and changes in those patterns. Given that some of the more complex claims can have several hundred individual notations associated with each claim, an organization that manages thousands of claims may be sitting atop of a huge mountain for text data hiding worthwhile, though difficult to extract insights.


Still, the nuanced nature of human communications coupled with the heavy use of technical language and abbreviations make the already difficult task of text mining especially challenging. However, with the help of a custom-developed thesaurus and a proven method for combining text mining-derived insights with the numerically-coded ones, the otherwise inaccessible data can become a fruitful source of decision-aiding insights.