Know thy Risk: the emergence of digital twins and the re-imagination of insurance

With exponentially increasing compute power, the proliferation of data producing sensors, and data production and consumption through devices like mobile phones, the ability to measure aspects of individual objects (cars, people, buildings) or processes (e.g., factory production lines, delivery logistics) is rapidly accelerating.  Models can now be highly complex, and with a combination of high-performance computing and advances in real-time data streaming, physical objects can have a model that is close to itself both in time, and bespoke to an individual instance of a physical object.  Welcome to the new kid on the block in the universe of mathematical models - the Digital Twin. 

So how do Digital Twins fit into the world of modelling physical reality?  A good way to think about this is to consider complexity versus uncertainty.  Some systems have a simplicity about them whilst still being very uncertain.  For example, in computational fluid dynamics, the rules governing the dynamics of an object moving through a medium, with fluidity like air, are simple, but there are a lot of unknowns when that object is propelled into space upwards through a thinning atmosphere.   

Conversely, manufacturing a car is very complex but extremely well understood - which is helpful when something goes wrong in the manufacturing process.  The situations where highly complex systems become very uncertain are the ones where digital twins play a unique role. In the subtle dance between the digital twin model and its physical counterpart, the model can tap into its predictive capability to test alternative more optimal possibilities, adjust reality, observe the subsequent outcomes and reduce uncertainty on the fly. As Dan Klein, Director of Data and AI at Zühlke, notes, "that fast feedback loop changes the game".  When the physical objects are things like cars, ships, people, buildings, that speaks to radically modifying an approach to risk - and therefore changing the insurance paradigm.   

Fast Forward to a Population Scale Digital Twin – The NHS England Test & Trace App

The concept of the Digital Twin has steadily emerged over the last couple of decades, and visually evokes ghostly images of aeroplane engines, buildings, wind turbines and oil rigs.  That upward trend would have smoothly continued as-is, were it not for a pandemic that gripped the world.   

Suddenly, urgent questions were being asked, and needed to be answered, on what the mechanisms and parameters of virus transmission are, how to break the chains of virus transmission, and understanding how the population of a country is moving around, all the way down to understanding how individuals come close enough to one another to be at risk.  As a situation, this was both highly complex and uncertain.  

One of the tools to help answer these questions is in your pocket - your mobile phone.  Packed with various sensors, a decent user interface, and a tendency for the item to be on your person most of the time, the possibility exists to estimate in detail when one human being is close to another, using Bluetooth signal strength between devices to estimate the distance between two devices that are close enough.   

That spatial relationship through time, combined with individuals updating their Covid test status, enabled risk management of those that came into close contact with a person who subsequently reported a positive virus test.  This was all done whilst respecting the anonymity of users and contacts, so privacy was not compromised in the process. 

Zühlke supported the NHS Test and Trace program by building the COVID-19 Contact Tracing application, capable of scaling to tens of millions of users within weeks of going live, and downloaded onto 22 million unique devices.  More than just a medical device, it was classified as "Software As a Medical Device" (SaMD).  With so many downloads, and really digging into becoming a real-time reflection of the distance dynamics of a large fraction of a country's population, that's well on its way to being a Digital Twin at population scale and helping to inform and manage the risk of the population.   

The feedback loop to users of their potential exposure enabled users to manage their own risks and behaviours. The data from the app, combined with other data such as rates of positive tests, could not only inform policy, but also be used to estimate the impact of the so-called non-pharmaceutical interventions by analysing data after interventions, such as lockdown or tiering decisions.    

The contact tracing app worked well and as intended, but there are lessons for the future, particularly around the public understanding of the technology and that is informative for the public understanding of digital twins.  For example, the inbuilt privacy in the contact tracing app meant that nobody would know who else had been pinged, but then users needed to have clarity on whether the alerts they receive are advisory, otherwise there is a misguided compulsion to delete the very app trying to help them.  The communications strategy around a service to users, and the value exchange involved is key. 

Commercial Property, IoT and the Start of Digital Twin Usage for Insurance

From an insurance perspective, an obvious target for the development of Digital Twins is with commercial property and the associated increasing adoption of IoT devices.  As Hélène Stanway, President of SENSE Consortium, says, “More data is an opportunity to drive more certainty, predict, and understand the uncertain more accurately.   

It is possible to determine a lot about a building far sooner than would otherwise be possible.”  With the top three causes of loss being escape of water, fire, and theft, with an estimated $120bn paid out in claims in 2020, one can easily imagine that sensor technology is the key to stopping the fire rather than having to pay for it.  This would be a fundamental shift in the way risk of a physical asset is managed and mitigated 

Such a shift has its own inherent set of challenges – for example, cyber-security around sensor devices is critical.  There is then the analytical challenge of pooling together all the data from the sensors to be able to construct a holistic picture of the system and what is happening to it in real time, and what might happen to it.  Subtly embedded in all of that is the role of uncertainty quantification and propagation.  Data from each sensor is noisy and imperfect, and sensors can fail, so aggregating imperfect data streams and correctly calculating and displaying the uncertainty around that holistic picture is critical to interpretation and decision making.  Practically, this means cross-disciplinary work between data specialists and people who are experts in the building itself is one of the success factors.   

The example of insuring commercial buildings illustrates one of the biggest challenges of all - adoption.  Greater knowledge of the building and the environment it is in should in principle contribute to reducing risk and driving down premiums.   There is of course the moral hazard that greater transparency exposes more risk and so premiums may increase, at least in the short term. 

Andries Smit, former MD of Aviva's Digital Garage and Partner at Stryber Venture Builders is sceptical. "Insurers are not in the business of reducing premiums. The common interest is in driving down claims (preferably whilst keeping premiums the same). Using IoT devices in insurance needs a complete rethink of the business model. There needs to be upside for both the insurer and the customer. Insurers will only care about this if it can be proven that claims are reduced disproportionately to any premium reduction, or if there are new services and revenue streams to them." 

The customer is sharing data and offering a more accurate view of their asset, which enhances the opportunity to personalise premiums.  The insurer could offer other services such as advice to mitigate risk, and real-time analytics to predict and prevent incidents.  Services might also support other outcomes, such as improved experience of people moving around the building, managing the building's energy consumption, and offering support on ESG compliance. 

Human Scale Digital Twin: Possible or Needed?

When the risk object is a person, it becomes obvious to ask, "Are Digital Twins applicable to health?"  We have already seen from the perspective of the pandemic that the answer is clearly “Yes", but the spirit of the question is to dig more deeply into the idea that a single human being could have an informative virtual counterpart where human health can be modelled.   

As Benedikt von Thüngen, Founder and CEO of Sanome, says, "building a human digital twin is the only way that we’re going to change healthcare – as it stands today, healthcare is broken. People around the world have demonstrated their ability and interest in engaging with digital healthcare monitoring during covid. We have the power, technology, and responsibility to change it for the better.”  

It does not help that the useful information of an individual's health journey through time is scattered across legacy estates of complex health systems, making it challenging to construct useful healthcare pathways.   

However, the advance of technology is giving us the means to measure the individual nearer real-time, with wearable devices, bespoke portable medical devices, and smartphone apps.  This only underlines the gap between what digital fingerprints can be established in near real-time, with being able to link the data to other data from more traditional clinical tests such as scans, and treatment pathways scattered across those legacy systems.   

Bringing these two worlds closer promises to transform the early detection and treatment of disease and, as time goes on, the fidelity of the digital picture of an individual will surely increase and get closer to real time.   

Humans by and large are used to the idea that devices are aiding them.  Glasses help us see, and hearing aids connected to mobile phones now provide a digital way of improving hearing.  Various kinds of implants and transplants provide a longevity to functionality otherwise lost.  Now with phones and watches, we can nudge the behaviour of the humans to improve their existence.  "Time to stand!". "Loud! - Repeated, long-term exposure to sounds at this level can damage your hearing." 

However, this desire to lead stronger healthier lives, supported by knowledge and predictive capability, comes with the greatest challenges of them all - chaining the individual's information together in a way where there is trust, respects privacy, and strikes the right balance between utility and anonymity.  

It is therefore no surprise that some of the most important data trends alongside the emergence of digital twins are around privacy enhancing technologies, federated data ecosystems, trusted research environments, and analytical solutions that navigate around a data ecosystem in a round-robin fashion to capture just what a measurement needs to get the right job done at the right time.   

Such approaches that might be so hyper-personalised and which will have to meet ethical requirements, are at risk of bias of various kinds, and could have unintended consequences even when individual data are aggregated up into larger cohorts.  With insurance, charging lower premiums for a low-risk cohort would have to be counterbalanced by higher premiums for higher risk cohorts. 

Quo Vadis, Insurance and Digital Twins

We are walking into a future full of risk and uncertainty - but measuring ourselves and our environment more accurately, to the point of the models being a twin with reality, opens that value exchange between being measured appropriately, and understanding and managing those risks at a granular level.   

Any digital twin of the physical or human world that can be used to monitor, model and predict outcomes that could be catastrophic, expensive to remedy, or both has the potential to be used for insurance 

That will change the game in the insurance industry, for health, property, and many other industries and sectors and in this article we have discussed just three digital twin applications that impact insurance: population level health, commercial buildings and the individual humans.  In many other papers, the areas that have already been identified where digital twins could be applied in insurance include: improve the accuracy of risk assessment, develop more effective pricing and underwriting models, parametric claims management, and improve overall outcomes for insurers and policyholders alike through the mitigation of risk and the offering of new services.  However, the potential use cases are much more varied and numerous. 

Digital twins are already being built for many other non-insurance related applications for process optimisation, product design, and predictive maintenance.  Many of these are in large capital intensive industries and applications such as shipping, aerospace, manufacturing, power generation, infrastructure, logistics, infrastructure.  What many of these have in common is that high value and the large costs when things go wrong.  So, combining the innovative applications in this article and the burgeoning digital twin implementations in other sectors, what, other than our imagination, is stopping the growing use of digital twins for insurance? 

If we build the future well, it will reduce some of the costly and disastrous outcomes we endure.  The challenge is the journey there, but we shall not be alone - our digital counterparts will grow closer to us, criss-crossing and nudging our real existence, guiding the way.