An Approach for improved Data and Analytics within the Wealth Management and Life Insurance Industries
In an earlier article, I discussed the challenges faced by Australian Life Insurers. The challenges include increased Government oversight and intervention, greater regulatory obligations and a more volatile customer base, each of which can equally be applied to the Wealth Management Industry. Add on the recent focus from Government Commissions (Royal and Productivity) and you have an environment that is forcing the better management of data within the industry. The earlier article identified opportunities to improve transparency of data within the organisation and to take advantage of an increase in external data sources. Here I discuss how an insurer or wealth manager can review the current data environment, identify areas for improvement or advancement, and provide a conceptual model for end to end analytics.
A good place to start when looking to improve analytical capability is an assessment of the current focus of Data & Analytics within an organisation. There are many different maturity models available. Popular ones include TDWI  and Gartner . The one below is the one we prefer to use for high-level assessments at Shoreline:
The phrases ‘Descriptive’, ‘Diagnostic’, ‘Predictive’ and ‘Prescriptive’ have been used for a while within the industry to describe different types of analysis functions but the distinction is not always clear. The following sections expand of these and provide some examples of metrics typically employed.
An assessment of the focus of most businesses often reveal that the greatest effort is still spent on ‘Traditional Business Intelligence’. Focus is on analysis of the current portfolio of policies and reporting historic movements in premium income or liabilities for an Insurer, or changes in account balances and flows for a Wealth Manager. This information tends to be well structured and has been centralised in a data warehouse for broader use throughout the business. It forms the foundation for deeper analysis and more advanced analytics. If historic lapse rates are not understood, for example, it will be difficult to predict propensity to lapse.
Some examples of descriptive analysis include:
- Premium inflow, outflow analysis
- Superannuation contribution and rollover reports
- Training and education management within advice networks
Although this type of analysis is highly valuable in providing information about the operations of a business (this point can easily be forgotten until the reporting goes wrong, however), it does not provide insight as to why the numbers are the way they are.
The next stage in the maturity model relies on better information, about customers or other offerings in the market for example, to be able to perform diagnostics about why something occurred. Knowing that a customer lapsed is interesting, but knowing they took up a competitor’s product or that there was a lifestyle or life stage trigger that caused the lapse is the result of powerful diagnosis.
When it comes to moving up the maturity scale, Life Insurers and Wealth Managers have traditionally taken a conservative approach, driven partly by the poor availability of data. The increased focus on data privacy considerations, such as GDPR in Europe, may also be resulting in a cautionary approach. The availability of a rich data sets has historically been limited by minimal direct interaction with customers, a situation exacerbated by the traditional adviser-led model of selling insurance and superannuation, or by the anonymity of corporate life insurance plans.
However, opportunities do exist to deepen customer relationships. The key to obtaining better information about customers lies in the ability to develop trust through clearly understood mutual benefit. One such example is the introduction of ‘vitality’ programs, whereby customers are rewarded for participation in healthy activity, in return for greater access to personal lifestyle or behaviour information.
Also, as customer interaction via online channels is increasing, better information is available about the navigation paths that a customer takes before making a purchasing decision, for example. Analysis of this information supports improved understanding of buyer decisions and conversion rates, and offer the opportunity to fine tune website navigation paths.
Another use for historical diagnostics is to facilitate expansion into new markets by deeper analysis of risk, therefore supporting improved underwriting.
Some examples of diagnostic analysis include:
- Marketing Campaign effectiveness
- Fraud detection
- Competitor analysis
- Customer retention diagnostics.
The ‘holy grail’ of analytics has long been the prediction of future behaviour based upon statistical analysis, and mining of rich data sets to identify hidden trends.
Questions that can be answered by predictive models include ‘how likely is a customer to take up on offer’ or ‘what is the likelihood that a customer lapses’, and providing the ability to adjust responses accordingly.
Of course, it is not easy to predict the future and few current models are capable of accurately forecasting behaviour at an individual customer level, but the approach of analysing past performance to prioritise focus and investment for customer retention activity, for example, is a valid one. This is the domain of the Data Scientist, equipped with the statistical modelling skills and familiar with the specialist tools of the trade in order to query large data sets and extract significant data points and trends.
We encourage businesses to take a broad view when determining what data sets to include in this kind of analysis. The best predictive models will blend internal customer transactional data with external macro market information to improve the accuracy of the result. For example, 2014 research by the consultancy Oliver Wyman  suggests that there is a correlation between the unemployment rate and customer lapse rates, suggesting predictive models should include at least some macro-economic data to provide better insight. Furthermore, there are an increasing number of data providers supplying (for a fee) access to broader information about individual customers, such as a recent house sale data or estimated mortgage balance. Leveraging these broader, external datasets can lead to better predictive results.
Types of predictive analysis include:
- Customer Value Management
- Predictive models for cross-selling and up-selling
- Product association analysis
- Churn propensity
- More advanced fraud prediction models for Claims Management
Typically the underlying data that supports this type of analysis is jointly owned and managed by IT and other business functions, such as Sales. Flexible infrastructure should be made available to support rapid testing of hypotheses in sandbox or test-and-learn type environments. The information used for analysis is typically less structured than that used for Diagnostics and may include once-off data loads from spreadsheets (downloaded from the ABS or APRA, for example) as opposed to structured batch processes.
It also requires a collaborative culture within IT and other business functions. This can be challenging where traditional IT data functions are used to operating in a structured fashion, producing repeatable, reconcilable data sets, versus analytics functions embedded in business units which may come across as ‘cavalier’ to a traditionalist. Cross-skilling and secondment between teams is one way to break down these barriers.
A further challenge with this type of analytics is managing stakeholder expectations. Building a once-off model relying on manually entered data or massaged spreadsheets may be a relatively simple and cost effective ‘quick-win’. Building models that can be reliably and sustainable executed over a broad set of data may require a much larger level of investment. Analytics managers must carefully balance the desire to demonstrate value in the short term with the need to build robust platforms over the longer term.
If predictive analytics is aimed at forecasting future outcomes, answering the question “what will happen?”, prescriptive analytics goes one step further and is aimed at providing a suggested response, or answered “what should I do (about it)?”. Examples of prescriptive analytics include dynamic pricing models, aimed at improving customer retention rates, or next best activity models that can be used as part of a cross/up sell campaign.
Typically this type of analytics will require test-and-learn techniques on a smaller cohort of customers, or experimentation within a contained environment. The expectation here should be that experiments will probably fail more times that they succeed, and that even when a successful model has been developed, it must be closely monitored to ensure that external forces have not caused the model to become invalid, usually via some kind of feedback mechanism.
As the term ‘maturity’ implies, this type of analysis typically can only be performed once there is confidence in the data sets that support Descriptive analytics, analysts are experienced and skilled in performing Diagnostics and the appropriate Predictive models are in place and have been proven to work. The implications of getting dynamic pricing wrong are large, so only the most mature organisations attempt this kind of analytics.
Types of predictive models include:
- Dynamic pricing
- Next Best Offer
- Predictive Life-stage Triggers
As we move into an era of ever more advanced analytical techniques and algorithms that are operating in a semi-autonomous fashion, we begin to approach the field of Artificial Intelligence, which is a topic for another day. What can be said, however, is that the best companies employing AI techniques have learned by progressing through the earlier types of analytics functions to develop the necessary maturity required to realise the potential benefits of AI.
The diagram below shows how different levels of analytics might be implemented across an organisation’s architecture (using a Life Insurance business in the example below). The flow of data throughout the organisation can be through of as a value chain, with each stage relying on, and enriching, the data from the previous stage. No one step is more important than another but strategic insight is improved as the diagram is traversed from left to right.
In conclusion, a life insurer or wealth manager is likely to have examples of the four kinds of analysis functions: Descriptive, Diagnostic, Predicative and Prescriptive. As the focus of organisations shifts to more advanced analytics, however, care should be taken to ensure the foundational data sets are robust and well governed, a culture of collaboration between IT and business-focussed analytics functions exists, and the expectations of senior executives are well managed so that it is understood that reliable and repeatable models take time and effort to develop, and may initially fail several times before they succeed.
At Shoreline, we have real, hands-on experience of implementing across the 4 kinds of analytics. We provide strategic analytics advice through to analytics project implementation services. Contact us for a discussion about how best to approach your analytics initiatives.
———————————– TDWI Analytics Maturity Model: https://tdwi.org/pages/maturity-model/analytics-maturity-model-assessment-tool.aspx  ITScore Overview for BI and Analytics, Gartner: https://www.gartner.com/doc/3136418/itscore-overview-bi-analytics
 “Don’t Leave Me This Way”, Oliver Wyman 2014: http://www.oliverwyman.com/content/dam/oliver-wyman/global/en/2014/jul/Life-Insurance-Customer-Retention.pdf