Analytic insightsForensic Focus - February 2009 |
Analytic Insights is a new service offering for Deloitte in New Zealand, so for this newsletter we thought we’d share a little bit of the typical process that is involved, and how that brings value to an organisation.
Essentially what we do is analyse data, in order to obtain actionable insights. Organisations these days are amassing enormous amounts of data in their various systems, and within this data is a comprehensive history of the organisation’s interactions with customers, suppliers, staff and so on. Getting insights from this data means that the organisation can fix problems, make more money, or support their customers better.
Our typical Analytics Insights process consists of three steps:
- Source and optimise
- Analyse and interpret
- Model and predict
Source and optimise
Sourcing data is identifying and obtaining the data for the Analytics Insight task. Typically we would work with key personnel to identify the potential data sources and the level of detail required. We then work with the required people to access and extract the data to our secure analytics environment.
Optimising entails organising the data into an analytical data set. There may be anomalies in the source data from systems problems, upgrades or changes, or data entry errors; this is all about cleaning it up sufficiently to be able to analyse.
Analyse and interpret
Once the data is structured, we interrogate and analyse it. This step uses subject-matter expertise from the client, and combines it with our collective experience built up from 100’s of client engagements. There’s also a place to let the data tell its own story, using assumption-free advanced analytical techniques which often brings brand-new insight to the situation.
For example, a credit lending institution knew that the employment type of an applicant was a significant factor in predicting defaults; full-time employed being a better risk than casually employed, for instance. However, analysis of the data set using Analytic Insights techniques identified a pattern in the data for full-time applicants: when the time of day and day of week of the application was considered, hot spots of potentially fraudulent loans became apparent for times such as Monday 9:00am when typical full-time workers might be expected to be at work.
Model and predict
Following the previous phases, we develop models to describe the underlying process. These are developed in such a way that they accurately reflect what is being seen in the data, but are simple enough to clearly communicate those findings.
Predictive models fulfil a particular need to be able to assess new, previously unseen cases. For example, in the insurance industry there is a need to assess whether claims need to be investigated for possible fraud. It is not feasible to investigate every claim, so a predictive model is used. The model examines the data for every claim and ranks them according to how likely they are to be suspect. Investigators can then work more efficiently by directing their attention to the cases to be more likely fraudulent.
For more information about how to earn a dividend from your data asset, please contact our team.