Technological advances are providing the tools and resources needed to improve reserving capabilities, says Lewis Maddock
Inefficient use of human capital is the principal reserving challenge of today. It leads to opportunity costs, because companies cannot deploy these human resources on continuous development, and reserving teams also lose talent when people feel their full capabilities are not being used.
The amount of time spent on reporting and monitoring is far from where it needs to be if we are to enable the innovation and incremental developments required for real change. In terms of control, the reporting process and supporting activities are too onerous, and consume too much human capital resource, to manage operational risks and control the output to meet requirements.
Given this strain on resources, and the compromises made due to deficiencies in the data and information needed for effective decision-making, why has there been so little innovation?
The reporting process has different objectives, constraints and timescales to typical modelling processes. For one thing, the governance controls and audit requirements are more onerous. This adds to the difficulty of maintaining the process – let alone considering innovative techniques and new approaches.
For example, in some markets there is a requirement to print out detailed information on the model parameterisation and the models used with enough detail for someone to independently recreate the results. This becomes even more challenging when you add complexity to the models involved.
The reserving process is an optimisation exercise. We recognise that there are a range of reasonable best estimates at any given valuation date, and optimise the selection of our best estimates to meet reserving objectives over time. This is a non-trivial process control problem, and real progress can only be made if we change the way we think about and perform reserving.
There is more to reserving than the reporting process, but until we address the fundamental issues, there is little opportunity for growth. Because of this, the common perception is that reserving is just a statutory requirement – however, it is a critical part of business intelligence.
What has changed?
Technological advancements can now provide the tools and resources needed to enact change in capabilities. For example, workflow management solutions allow more decision support material to be produced in a shorter timeframe, while maintaining and improving governance and controls. The leading solutions in this space reduce the need for ad hoc analysis, provide more data-driven analysis, and free up time for embedding further capabilities.
Such tools allow us to integrate software and systems in an end-to-end process, supporting a best-in-class approach to the architecture. This means we can use the best tool for a particular job in an end-to-end process.
Cheap computing power is also enabling us to better leverage data assets. For example, robotic process automation is being used to produce more for less with increasing granularity; interpretation techniques are helping decision-makers to identify pertinent information and prevent it from being lost among large volumes of analytical output; and machine learning is being used to unlock value from unstructured data.
The ability to integrate a range of systems and applications into a coherent process has expanded the possibilities not just within reserving, but also across an organisation. For this reason, real change in reserving appears feasible.
The impact of machine learning
Machine learning’s role should be limited primarily to enabling targeted elements of the reserving solution, rather than being a one-stop remedy. It needs to be considered as part of a wider roadmap towards a future target operating model. Investing in the right development at the wrong time is probably the biggest pitfall when it comes to machine learning and reserving.
Besides operational efficiency and process control, machine learning’s main benefit in reserving is in providing improved insights, and at its core it is designed to tackle the problem of optimisation. In the context of reserving, we are effectively running an optimisation in order to hit a moving target, with the objective of considering the cost of being wrong in order to provide meaningful outputs. This is different from the machine learning techniques used elsewhere by insurers, in which supervised learning methods are typically used with a well-defined response to minimise the error in fitting to historic data.
In pricing and risk modelling, for example, an insurer can assume that the risk differentials in its recent historic policy and claims data are representative of near-future experience. This is a reasonable assumption in practice – if the insurer fits a model that is predictive of risk relativities based on recent history, it can create a predictive model of experience in the near future.
The difficulty in reserving is threefold. First, the information needed to inform an answer in the more distant future may not be in the historic data. Second, estimates will change over time as the insurer generates more information. Finally, the reserving process and the way the insurer communicates results will be different in the future. It needs to be supported by many more visualisations than the insurer currently uses. Back-testing diagnostics, for example, will be essential. This is why the roadmap is so important – it helps insurers to put developments in the right order so they can get to where they need to be.
No single machine learning method will solve all the problems in reserving. Indeed, some developments will exacerbate existing problems if certain parts of the wider solution are not in place beforehand.
We are seeing a gradual move towards the use of machine learning in projecting ultimate losses. In considering the end goal for machine learning in reserving, a useful analogy may be found in process control applications in robotics. Imagine a quadcopter is programmed to fly through a hoop that is thrown through the air at random. The quadcopter must monitor the data feed from sensors tracking the position of the quadcopter and the hoop, and then determine adjustments to its speed and direction to optimise its trajectory.
How the hoop is thrown, how the wind blows and numerous other variables mean the quadcopter cannot know where the hoop will be when it flies through it. The algorithms optimise a decision, given all the information available at the time, to minimise the probability of missing the target, and repeat this at regular intervals until the target is reached.
In reserving, we cannot know what the ultimate loss for any given cohort will be, just as the quadcopter cannot know where the hoop will be. However, we can optimise the output from our reserving processes to acknowledge the fact that we're on a journey and minimise the cost of being wrong at any given time.
The roadmap to unlocking reserving capabilities
The insurers that have most improved their reserving capabilities are those that have made the most progress towards their defined future operating model. They are differentiated from the rest by their strategic development planning and clear upfront objectives for these exercises. A lot can be done to improve operational efficiency and process control before using machine learning. It has its place, but it's not the whole solution.
Embedding a workflow management solution in the reserving process is critical. Better automation capabilities will enable insurers to produce more decision support material while mitigating the problems of maintaining governance, controls and audit, and freeing up resources for development. This tooling must be in place in order to enable real growth in data-driven analytics, as well as to support reserving without expenses spiralling out of control. Insurers will be able to readily integrate new tooling into the environment with improvements in governance, control and audit. Robotic process automation will further complement these capabilities by producing increasingly sophisticated data-driven analysis and output in a timely manner for the same headcount.
Lewis Maddock leads the Nordic Insurance Consulting and Technology business and global future of reserving intellectual property development at WTW
Image credit | Shutterstock