The winning paper, by Tony Boobier, of the business management category of the 2001 Lumina awards sets out how a major composite insurer has increased operational efficiency, and used its accumulating knowledge and data to enhance its operational effectiveness and increase its risk management knowledge in subsidence claims, one of the most technically difficult of all domestic claims.

What is subsidence?
Whilst generally viewed as a UK insurer phenomena, the problem of subsidence occurs world-wide and relates to the movement and subsequent damage of buildings as a result of factors unconnected with the weight of the property itself.

This is a complex problem but with some underlying themes; 70% of cases involve the interaction of trees on buildings, but there is considerable variability between trees, sites, property shapes, ages and policyholder behaviour.

In the UK, The Association of British Insurers (ABI) indicates insurers will spend in excess of £350m per annum in a non-drought year and in excess of £500m in a drought, or `event' year. The Chartered Insurance Institute (CII) suggests that a `super drought' may cost the industry in excess of £2bn. This particular insurer has an exposure of £100m pa in respect of subsidence claims alone. Reinsurance has traditionally not been readily available.

There is a high degree of uncertainty whether any particular year will be an `event year', when new claims numbers increase by a factor of four within a very short time-spell, creating a major supply/demand imbalance (figure 1).

This imbalance creates serious problems in the provision of the claims service to policyholders, the very great majority being domestic homeowners.

Although an `event year' tends to occur on average every seven years, climatic changes indicate the potential for an event year to occur with increasing frequency, perhaps as much as every three years.

This problem also needs to be seen against a background of subsidence cases creating a high degree of anxiety to the customer, due to the historically fragmented and mainly reactive industry. This anxiety was analysed as part of the project and is described as the `emotional rollercoaster' (figure 2).

The challenge is therefore one of developing a methodology for the `routing' of new case notifications to allow insurers to place greatest resource into those areas where claims cost leakage is at its highest, and in doing so manage the loss, and to place less emphasis on those cases which were most likely not to be due to subsidence or any other insured peril, yet at the same time not adversely prejudicing the policyholder/end user experience.

Greater understanding of the subsidence claims characteristic will also allow preventative action being taken, by targeting those areas where claims are most likely to occur and offer advice/take action accordingly.

Framework for progress
This section of the report deals with the sources of, and manner of collection of data.

This insurer has already heavily invested in its people and systems. Current claims administration systems are working efficiently and constantly improving, despite the legacy of merger and its associated complications regarding policies, customers and disparate systems.

This therefore provided a solid foundation to take the claims process beyond that of administration and into `smart' handling and cost management, risk management, and ultimately as a cornerstone of a fully mature customer, relationship management.

The first step was to introduce the capability for decision-making and routing into what is has necessarily been, until now, a purely administrative issue.

For subsidence in particular, where the cost of handling each claim - declined or otherwise - starts at several hundred pounds minimum, any measures which provided for alternative resolution based on lower cost, are particularly effective.

Therefore the principal, which was applied to the original system, was to effectively gather, accumulate and harness the knowledge, information and expertise already existing, into a more cohesive, systematic and focussed approach, and thus:

  • building improved systems now for immediate and substantial operational gains;

  • designing a coherent framework for a practical integrated system; and

  • providing a core strand to the wider strategic aim of total customer relationship management.

    Capturing knowledge and data
    Understanding what data and information is available, where and how it arrives, and how it is stored, distributed and utilised, is a prerequisite to progress. This was achieved by the construction of a distinct claims database.

    This first section looks at where the key areas of data and knowledge are, and how they were collected to maximum effect. It deals with the data issues of data and knowledge only. Separate discussion of the mechanics of the various systems, and how all information might be utilised, follows later. The three key areas were:

  • contact with the customer;

  • contact with the expert; and

  • external and internal databases.

    Contact with the customer
    This insurer has built over time a tremendous asset in its claims handling staff. Experience of the issues, the process, how to service and guide customers, and how to liaise with external engineers are all bound into successful operations. With appropriate support systems, guidance and training, it was considered possible to leverage this skill base to the full.

    From key observations of the notification process, it was discovered that:

  • casual information obtained during the conversation with the customer at the point of notification was not `needed' and therefore not recorded;

  • information recorded was not `needed' for administration purposes and therefore not be captured;

  • information that was gathered was not held in a form suitable for systematic retrieval and analysis; and

  • improvements in operational effectiveness relied upon non-system improvements through handler, engineer and manager experience, rather than through systematic knowledge building shared across the operation and built into practice and analysis

    Systematising the current approach to create a uniform minimum data-gathering standard at point of initial contact was considered a cornerstone of any operational improvements. Whatever approach was adopted, it needed, however, to allow for the flexibility necessary to deal with different handler preferences and expertise, customer demands and channel differentiation.

    Therefore it was recommended, as a starting point, that a standard data collection form was introduced with an increased number of data fields, to be used by all handlers at the point of claim initiation by the customer.

    Rationale of new data fields
    A number of fields needed were already standard pieces of gathered information (although not necessarily as captured database fields on current systems). The data identified as additional provided the means to move the process onto a `smarter' footing, and requires additional explanation.

    Building data
    Already partly available from policy details, but gathered additionally if there is any doubt. Checks against engineer site reports are made to investigate the veracity of data supplied by the customer at either stage, providing evidence of reliability of current data and the degree of confidence that can be placed on customer information.

    Rigorous accumulation of building type and age data (and over time additional characteristics) provides a cornerstone of statistical analysis, used in risk management.

    Buildings vulnerability
    Available hazard models provide a geographical distribution of expected risk. How each building responds - its vulnerability - depends upon its engineered characteristics, and these vary according to age and form.

    We `expect' more claims from areas of highly shrinkable clay (greater hazard). Within those areas of clay we would `expect' more claims from inter-war property than from `modern' property due to shallower foundation depth. Alongside other site and customer specific data, this provides the basis for both scoring, and ultimately resource planning.

    Claim details
    This was the main point of departure from current practise. Much of this information passes in an unstructured way during the initial conversation. It finds its way into paper notes, but for original requirements it was not necessary to systematically record.

    What was being sought to achieve here was the `best' understanding of the condition of the building prior to an expert visit, and also to identify key characteristics of the customer, which may provide insight, and over time statistically valid data.

    This demanded the detailed and systematic recording of all data at all points of the claim process.

    Interaction with the expert
    Current practise is to automatically appoint an expert as first response to a notified claim.

    Given that more systematic recording is necessary for other reasons at notification, it made sense to produce a report template for the engineer to submit relevant data based on the new notification form.

    The expert's initial appraisal report
    The intention of this initiative is to transform current administrative-based systems into dynamic knowledge-based business management systems. Therefore switching information from paper-based filing into retrievable and useful data, is fundamental.

    Appraisal reports contain a wealth of information regarding the property and details of the claim.

    Reports existing contained specific fields which could be readily transferred to a database structure, or used to correct/amend/augment similar fields (such as house type/age) already gathered.

    Systematic analysis and use of data for routing decisions, streamlined administration, as marketing and point of sale information, and as mining material for underwriting/pricing is currently stored, but in many ways also `lost', in a number of locations, through lack of collection.

    A careful, longer-term approach was needed in this area, not only to ensure that data collected is not subject to continuous change and the learning curve is climbed, but also to ensure systems dovetailing with any strategic operational changes in this area.

    Storing site-led information on the claims database
    By using an appropriate typology for building types, damage causes and works, with appropriate freeform text, it would be possible to build an insurer-wide UK buildings database as a subset of the claims database, and from historic data held elsewhere (whether the current MIS has retained enough buildings data to utilise seriously, is subject to further investigation).

    On notification, the claims database enables identification of:

  • previous experience of the same property;

  • previous experience from the same street/locality; and

  • previous experience of similar types (by building, cause or other circumstance).

    This informs the process, improves judgment for procedure, and ultimately adds to any automatic scoring capability.

    Such a database will also:

  • improve quotations;

  • identify shortfalls in cover/sums insured and opportunities to develop the customer relationship;

  • provides an `alert' capability;

  • provides deep, consistent and valuable data for risk/price modelling and underwriting; and

  • provides stronger data for modelling of rebuilding costs - reserving and pricing.

    Additional information not currently captured anywhere in the process is quick and easy to obtain and provides useful background - especially as reference to further claims. This included:

  • more detailed and systematic recording of building characteristics;

  • observations regarding the uniqueness of this property's problems in the context of the neighbourhood;

  • is it a specific problem or are more houses in the street/vicinity likely to be susceptible;

  • is the house typical of the street/neighbourhood in terms of construction; and

  • is there a widespread tree issue in the neighbourhood.

    This data not only aids scoring and provides a wider footprint of data than for a specific property, but aids issues such as prevention where RSA has other policies nearby that are likely to fall under the same influence.

    Satellite telegraphy
    Given that photographs are scanned routinely as part of document storage, implementing a database retrieval system to link images with other database information is also an achievable goal. The use of digital imagery is likely to become routine, and should be geared up for now.

    Risk models, estates lists, other data and analyses
    Knowledge, information and useful data are the foundations of a `smart' approach to subsidence claims management. This data is not only available from within the totality of the claims process itself, but can be external; analysis and understanding which can be applied to a particular occurrence at a particular location.

    Location analysis - the geography of subsidence claims
    Subsidence claims do not occur on an even-distributed or random basis across the UK. Claims vary by exposure, but also in accordance with the underlying risk.

    Primary causes of subsidence vary by location, and can be categorised by the following types:

  • differential ground movement associated with shrinkage of clay-type ground conditions (often associated with tree impact);

  • ground movement associated with previous mining activity (by various extraction types, which, also very geographically, including coal mining);

  • natural consolidation of geologies such as peats and alluviums over time;

  • landslip;

  • in-fill of previous land-use causing weaknesses in underlying load-bearing capacity;

  • solution cavities caused by water impact on underlying geologies;

  • other site-specific causes such as loss of fines due to drain failure etc.

    Underwriting subsidence database
    This insurer currently licenses a full postcode risk model for underwriting purposes, which provides core data on the `risk' associated with the first six causes. It is based on UK-wide geo-technical research commissioned from consulting engineers. It also provides vulnerability assessments for a broad band of building types, and is a fully developed `claims cost' model, including an expected claims rate for each postcode.

    This model was used as a `reference' table to help with any expert judgment applied to each claim in the routing process. Which of the data columns (hazard through to claims frequency) proves most useful over time will ultimately be determined by monitoring.

    Limitations of underwriting model
    It must be stressed that pricing models, despite major technical and modelling investment over time, remain models. Subsidence processes are highly complex and on a site case-by-case basis extremely difficult to predict (apparently near identical situations can produce different outcomes). However, they are based on probabilities of outcome, and those probabilities provide a more efficient management of risk over the book of exposure, and more accurate individual pricing, than would otherwise be possible.

    The underwriting model is extremely useful in scoring:

  • claims will be more `likely' where the claims frequency is highest;

  • certain causes will be more `likely' in different locations (shrinkable clay in London, mining and other causes in non-clay areas); and

  • the geo-technical hazard assessment will provide the likely geo-technical regime for each notified claim (not geology, but geo-technical ground conditions, based on plasticity etc.).

    Accessing the underwriting data table at point of notification provides both a scoring variable, and useful data to be passed to the engineer.

    Insurer risk pricing models
    The risk pricing subsidence model was brought into the insurer as an aid to improve existing risk pricing and rating structures - improving actuarial analysis of claims, location, and costs. Detailed modelling in this area indicates which factors are statistically valid in claims cost pricing (but of course not in determining validity of claims).

    This data is useful on a number of levels. At a headline level, the final rating tables provide a look-up for expected `risk' by postcode (normally in 30 rating district bands), with additional multipliers by house age (four bands) and house type/size category. By matching these parameters to those on claims notification we can identify:

  • whether a claim from that particular locality is more or less likely than from other localities; and

  • whether by building characteristics (from the notification form and/or policy record) it is a property type susceptible to claims.

    Current and further reporting and analysis
    Allied to core systems is the issue of current reporting and available analysis. In order to obtain a complete picture of some of the fundamental issues, some key overview statistics are also helpful.

    The following analyses have been included in the model to date. These would be followed by more detailed investigation as any project is pursued:

  • customer geo-demographics;

  • estates list - a record of well-known problem areas;

  • credit rating;

  • scoring analysis (including expected lifetime values); and

  • County Court Judgments.

    Whilst these data are likely to be of secondary value, any insights provided can be incorporated over time, and add to CRM processes.

    Developing external data relationship and co-operation
    Third parties such as engineers, loss adjusters, underpinners and their associations contain a wealth of information. Developing relationships that bring these into use (such as a database of previously underpinned properties) would be prudent. Success will depend on utilising and developing current relationships and being prepared to provide value in return.

    Co-ordinating and integrating data sources into a claims and buildings database
    A picture thus emerges of various data streams gathered for differing activities, but underpinned by a common thread - the need to understand risk and operational issues associated with particular locations, properties within those locations, and policies taken out by customers living in those properties.

    A central proposal here is that as a result of the `smart claims' process, a clear, well-defined data structure is designed which feeds and is fed by the different areas of the business, and becomes a growing knowledge base with an intrinsic and increased business value in its own right.

    The single framework
    At the heart of the claims management system therefore is the accumulation of knowledge and data within a newly structured `claims' database, containing details regarding the claim, which is additional to those currently required for administration and operations.

    Within that claims database are details keyed in directly regarding the property - the property database.

    The claims and property databases are fed by interaction with the customer, the policy databases, claims administration systems, engineers' reports and observations, risk data and pricing models, underwriting knowledge and the estates list, and other external data.

    This does not necessarily mean re-engineering current organisational structures, merely the building of a concise, adaptable, additional database structure to act as a container for currently lost information, and to act as a link between currently disparate data, harnessed into a coherent structure (figure 3).

    Harnessing the data - active claims routing
    This section deals with how the accumulated knowledge and data is incorporated into an operational model.

    As in the accumulation of data, the application of claims routing need not involve major systems re-engineering at the outset - at its simplest it is an informed and systematic interpretation of the available data: Whether the routing decision uses paper files, database tables, incomplete initial data or other limitations, the key decision is to make a routing decision, and then act upon it. The accuracy of the predicted outcome will improve over time, hand in hand with the systems supporting it.

    At the time of writing, the current level of successful prediction through the automatic model was in excess of 70% accuracy, exceeding the accuracy of the most experienced claims handler.

    This is the fulcrum of bringing the concept of `smart' claims handling into current practice, and the key point from which benefits will flow. It is providing a powerful mechanism to decide, at the point of notification, the most cost-effective way of dealing with that particular claim.

    Not all claims are equally valid or will produce the same outcome - in time or cost - and therefore need to be treated differently.

    The system ultimately provides for a `triage approach' to the handling of subsidence cases.

    Integrating claims routing systems into current and evolving structures
    Figure 5 shows a schema of the current data flow through the claims operation, including storage and sources of data referenced in the process. Current priorities are fast handling and earliest closure, quality tracking, engineer integration and reporting.

    At each step, from initial contact with the client through to recording data into paper files and onto systems, data which is useful as both a central resource and to be used in decision-making is being lost or not stored as retrievable, discrete data.

    The elements necessary to transform the situation by enhanced data collection, storage and use have been outlined. How the mechanisms devised slot into current systems and operations are laid out as in figure 6.

    The transformation of data collection to allow for decision routing entails, at the first level, the implementation of:

  • enhanced data collection as previously outlined (facilitated by new forms at notification and completion and enhanced data collection and capture from the engineer); and

  • the claims/property databases as an addition to current systems.

    By addressing all the key areas of data collection and utilisation, and embedding claims specific data gathered from notification and completion forms and interaction with the engineer, the operation has the capability to function at a `higher level'.

    The wider strategic context and risk management
    This section deals with the strategic and tactical application of the routing and data collection models, including risk assessment of individual properties.

    There are a number of wider strategic and tactical issues to which this development lends itself, dealing with risk and claims management, and also reinsurance. These include:

  • support and enhance current operational changes, including supplier management issues;

  • preventative campaigns around `hot-spots', where there is a greater potential for anticipating damage and recommending action be taken, particularly tree management;

  • historical data analysis;

  • observation of trends;

  • re-investigation into predictive analyses, using all data to continuously refine the model;

  • flattening the claims curve (figure 1), increasing efficiency during the supply/demand imbalance;

  • block policy analysis, where cover is placed for corporate and affinity partners, including surveying, predictability, improving efficiency, bottom line profitability, and reinsurance opportunity; and

  • increased excesses and their sensitivity on the claims process (£500 - £1000, variable, optional excess, etc.).

    Current development plans focus on achievable, step-by-step use of current systems - gathering core data vital for any longer term decision-making. The new model also allows fundamental input into strategic re-engineering of the whole process, from `claims' management through to business/customer relationship management.

    Risk management
    The objective of this element of the project, as defined by this insurer, is to produce a model from the data that is "able to produce a credible, useable and quantified measure of the risk of damage to any given property caused by shrinkage-related subsidence associated with local trees or other sizeable vegetation".

    Complex variables
    In recognising that 70% of all cases involve the interaction of trees on buildings, estimating the potential for tree related subsidence is a complex problem.

    The behaviour of the soil in the vicinity of trees depends on the species, size and growth rate of the tree, the pattern of moisture and moisture movement in the underlying soil, and the soil properties. It varies with time, and is affected over a period of seasons by climatic conditions. The behaviour of a building located on shrinking and/or heaving soil near trees depends on its distance from the tree, the magnitude and distribution of the ground movement, the depth and type of its foundations, and the construction characteristics of the building. Because of these variables, any attempt to give a definite prediction as to whether a given building will incur subsidence damage is unlikely to be successful.

    Nevertheless, there is now adequate theoretical knowledge and performance data on tree-growth, on moisture-related soil movements and on cracking of buildings resulting from subfoundation soil movements, to create a viable probability model.

    Currently under development, the model has the following features:

  • it will be based on an engineering understanding of the risk, distinguishing hazard and vulnerability and offering a probabilistic estimate of damage;

  • it will make use of the substantial databases of damage assessments that are being currently collected by this insurer, as described earlier in this paper; and

  • it will estimate the probability of damage over different time periods, allowing for probable climatic patterns and tree growth.

    A key feature of the proposed model is that it is based on an engineering understanding of the basis of risk, distinguishing hazard and vulnerability, each a quantity defined on a probabilistic basis.

    The provision of such a model, applicable through palm top IT hardware, will have a significant impact on future claims practice and underwriting but also will have application within the housing market where many problems are discovered at point of sale/purchase.

    Subsidence is a complex, well-established experience and knowledge base, but historically with each case considered as `unique', and with data collected through a legacy of different systems. Many more issues and information are being gathered as context and as understanding, but this report has focussed on the key issues and looked for the simplest and most effective approach..

    The key change has been that of structured data collection, and this will need to develop in line with actual experience, following agreed principals set out. This ultimately will save claims cost, improve efficiency, assist the management of suppliers and provide a framework for risk management and loss prevention.

    Already the process demonstrates 70% success of predicting the outcome of any case, allowing effective routing decisions to be made. This will be invaluable in coping with the next major drought when there will be major supply/demand imbalances in the supply chain. During the course of this year, it is anticipated that direct advice will be capable of being given to homeowners to act in tree management issues, and in doing so, reduce or avoid damage occurring.

    In a product which after 25 years of cover has remained mainly fragmented, this programme represents the first real attempt in the marketplace to harness knowledge on a large scale, and in doing so will unlock the problems of the past in this traditionally difficult area.

    As a result, it is anticipated that the market will for the first time show a semblance of being in control of its destiny despite the potential for major problems due to changing environmental conditions. With greater understanding of the problem and higher predictability of damage, there will be increased opportunity for reinsurance of the subsidence risk.

    By Tony Boobier

    Tony Boobier, subsidence programme director of Royal & SunAlliance, based in Horsham, UK is a Chartered Civil Engineer, Chartered Loss Adjuster, Associate of the Chartered Institute of Marketing and European Loss Adjusting Expert. Tony is a member of the Subsidence Steering Committee of the Association of British Insurers.

    The changes highlighted in this report together with other related initiatives will provide financial savings independently measured at £120m over a five-year programme. Currently in year three, the programme is on track for those savings.

    Recognition is made of the major contribution to this work of Garry Stone, subsidence programme manager at Royal & SunAlliance.

  • Topics