Digitalisation & Technology, 4 August 2026
Quantum, synthetic data & beyond
The emerging technologies insurers are already testing
What do a weather forecast for the year 2035, real-time fraud detection and a new cancer therapy have in common? At first glance, nothing. On closer inspection: all three could one day benefit from technologies that are still at the research stage today – but may soon be part of everyday life in the insurance industry.
Quantum computers, artificially generated data and AI systems that can understand the physical world instead of merely spotting patterns in vast datasets: all three technologies are still in their infancy. Yet they are already highly relevant, because AI is massively accelerating the pace of technological progress. In other words: the course for the future is being set now. Through pilots, research partnerships and cautious early test runs, these technologies are already being explored and refined.
This article looks at three technologies that are not production-ready yet, but already hint at where things may be heading. Our aim is to show why it pays to pay close attention today rather than only reacting once competitors are already ahead.
Quantum computing: Not just faster, but fundamentally different
We’ve seen quite a few “quantum leaps” in computing over the past decades. While this label was handy from a marketing perspective – and each new generation did deliver major advances – the real, literal quantum leap is still to come. The next generation of quantum computers won’t just be faster; they’ll be based on a completely different computational paradigm.
A classical computer works with bits that are either 0 or 1. There are only these two states. A quantum computer, by contrast, works with qubits that, according to the rules of quantum mechanics, can also exist in a state of superposition between 0 and 1 – in other words, 0 and 1 at the same time. Qubits can also be entangled, so that the state of one instantly determines the state of another.
The theory may sound strange, but the impact becomes clearer when you look at what qubits enable in practice. The combination of superposition and entanglement allows quantum computers to explore many solution paths in parallel, instead of stepping through them sequentially as classic 0/1 machines do. The result: computational capacity increases exponentially.
Why is quantum computing relevant for insurance?
At its core, much of the insurance business is about calculating probabilities and quantifying the cost when a given event actually occurs. This involves running large-scale simulations across many possible futures. The accuracy of these estimates depends on how many simulations can be carried out. The practical challenge: if you want to double the accuracy, you typically need around four times the computing power.
For less complex scenarios, that trade-off may be acceptable. But not for extreme events such as natural catastrophes, pandemics or major cyber incidents. These events are severely underrepresented in traditional models precisely because they are rare. Yet they matter enormously to insurers, because they drive the largest losses.
New quantum-based numerical methods could push these boundaries significantly. Probability calculations become cost-effective even for extreme events.
Another important use case is risk optimisation: how do you assemble a portfolio from countless possible policies that meets customer needs without breaching risk tolerance? Here too, the number of variables is a decisive factor – and something quantum computing is well-suited to handle.
For insurers, quantum computing is less about speed and more about uncovering risk patterns that are hidden in today’s highly complex insurance scenarios.
Synthetic data: Data for things that haven’t (yet) happened
Mainstream AI models have a built-in limitation in an insurance context: they are trained on historical data and produce answers based on statistical probabilities. That’s not exactly ideal when you’re trying to forecast rare events that lie in the future.
Synthetic data tackles this issue by augmenting the training set with artificial – but statistically plausible – data points. In essence, you expand the dataset used for training specialised AI models to such an extent that even rare cases are represented in sufficient numbers.
You can compare this to pilot training in a flight simulator. Extreme situations such as engine failure are simulated in numerous variations until the pilot can respond optimally at any time. Only very few pilots will ever face this situation on an actual flight, but they’ll still be fully prepared.
Why is synthetic data relevant for insurance?
The rarity of certain events – large-scale natural disasters, novel cyber attacks or new fraud schemes – poses a major challenge for insurers. Even with AI support, these issues are hard to tackle because historical data doesn’t contain enough examples to derive robust patterns.
Artificially generated data can fill out the “tail of the distribution” with realistic scenarios, so insurers don’t have to wait for real-world incidents before they can learn from them.
Two further aspects are mainly practical: data protection and scalability. Synthetic data, unlike real customer data, is not sensitive from a privacy perspective and can safely be used for AI training. It also doesn’t need to be painstakingly collected; it can be generated relatively easily.
Important: synthetic data is always a supplement, not a replacement for real data.
The biggest change ahead of us is prevention at scale. With improved simulated forecasts, insurers can take proactive measures to reduce claim frequency and improve combined ratios.
World models: AI learns the physics of the real world
As limitless as the knowledge of large language models can sometimes appear, there is one area where they still lag far behind even a toddler: they don’t understand how objects behave in the real world. While toddlers quickly grasp what gravity means, today’s AI tools only “know” the theory. Consequently, even the latest AI-controlled humanoid robots struggle to mimic the fluid movements of humans.
World models are designed to close this gap. Instead of training a model purely on text, these systems ingest vast amounts of video and motion data. The aim is for world models to learn the physical world: gravity, inertia, cause and effect, and the spatial relationships between objects.
Put simply, a conventional AI model is like a highly intelligent artificial brain without a body. It may contain a wealth of information on balance and be able to explain the theory perfectly. But it has never experienced balance itself and could not keep a physical body steady.
Why are world models relevant for insurance?
The practical value of world models for insurers lies in their predictive nature. These systems don’t just describe what has already happened; they can estimate what is likely to happen next. That’s exactly what many emerging technologies depend on:
- Autonomous vehicles need to predict whether a pedestrian is about to step into the road.
- Robots must apply the right amount of force when grasping objects.
- Climate models become more precise when AI can place physical data in causal relationships.
Another advantage emerges when you combine world models with synthetic data. A model with a “physical understanding” can generate artificial data that reflects realistic dynamics. The pedestrian mentioned above could be extrapolated into many different scenarios: how do things change in wet or icy conditions, what difference does it make if the pedestrian is elderly or young, how do time of day and season affect the situation?
Conclusion: Risks become more predictable
All three technologies are still early-stage, but they already show considerable potential for the insurance industry. They enable more accurate climate and weather models, extend fraud detection beyond known patterns and could help AI systems detect signals for rare diseases that lead to new treatment options.
Last but not least, they may fundamentally transform the insurance sector and turn it into a new magnet for tech talent.
Text: Falk Hedemann
Your opinion
If you would like to share your opinion on this topic with us, please send us a message to: radar@ergo.de
Further articles