The end of 'good enough' risk management
For decades, insurance risk management operated on a reassuringly stable foundation: actuarial tables, historical loss data, and expert judgment. The system was reactive, a total rearview mirror industry, addressing damage only after it occurred.
Today, that approach no longer works.
The less predictable nature of climate events is difficult to explain using historical frequency curves. Cyber threats change faster than underwriting rules can keep up. New risks related to autonomous vehicles, AI liability, and pandemics have little loss history to guide pricing. Throughout the policy lifecycle, from risk modelling to underwriting, pricing, and claims, the manual, isolated processes that once worked well now create competitive and financial risks.
Enterprise technology and business leaders in and around the insurance sector now face a new challenge: not if they should modernise risk management, but whether their organisations are truly architecturally prepared to keep up with the market’s accelerating demands.
The real cost of fragmented risk functions
Modern insurance risk management is a chain of interdependent functions, each of which must perform well individually and in coordination. Risk modelling informs risk assessment. Assessment drives underwriting decisions. Underwriting assumptions feed pricing. Claims outcomes reshape the models that started the cycle.
When these functions operate in isolation, the compound effect is significant: decisions at every stage are made on information that is incomplete, dated, or inconsistent with what other parts of the organisation already know. This is the structural failure mode affecting most established insurers. Underwriting guidelines are updated quarterly or annually, far too slowly for markets where conditions shift in weeks.
The architecture mismatch
Capgemini's report calls this structural problem the "architecture mismatch." Their research shows that transformation often stalls because business units are fragmented and organisations lack the expertise needed to drive large-scale innovation. This mismatch shows up in three main areas:
- On the strategy side, only 35% of the top 20 global P&C insurers clearly connect their AI strategy to business goals beyond just efficiency. Most AI projects are still seen mainly as ways to cut costs, not as tools for growth or standing out from competitors.
- From a technical perspective, 81% of insurers say that old systems and IT limitations are the main barriers to scaling AI. Another 74% mention problems with data quality and sharing data across teams, while 61% highlight regulatory and compliance issues.
- On the organisational side, 67% say a lack of AI skills is a major barrier. Another 55% point to unclear ownership of AI projects and the lack of clear ways to measure ROI. As a result, these programs often depend on a few motivated individuals instead of strong support from the whole organisation.
This creates a cycle where limited ambition leads to modest results, and those results reinforce the lack of ambition. Capgemini calls this "AI fatigue," a growing credibility problem that cannot be fixed just by investing in more technology.
Move from static risk models to dynamic intelligence
Traditional actuarial models rely on past loss data and statistics, assuming the future will look much like the past. This approach works for slow-changing, familiar risks. But with rapid climate change, new cyber threats, and risks like accidents involving autonomous vehicles, these old models are no longer enough.
Still, no modelling approach, whether traditional or AI-based, can fully avoid relying on the past. AI systems are trained on historical data and learn from patterns that have already happened. The future's ontology and dependencies remain fundamentally unknown.
So, the change that leading insurers are making is not a jump from looking at the past to fully predicting the future. Instead, they are getting better at quickly processing signals from both the past and present, using more sources and greater detail. Still, they face the same basic limit as always: the future can only be estimated based on what is already known.
AI technologies are central to this transformation. Modern map-reduce/HPC tools mixed with further deep learning models can process massive datasets to identify patterns invisible to traditional methods. Another study has shown that AI models can predict credit defaults at rates exceeding conventional methods by up to 30% when integrating employment data, industry risk factors, and even climatic variables into their analysis.
This idea applies across different insurance types. For example, building sensors that monitor vibrations, the environment, and fire suppression give a continuous view of property condition instead of a one-time survey. Supply chain finance data can show if a commercial policyholder’s risk has changed significantly since the policy started. In all cases, insurers need data systems that can handle large amounts of both structured and unstructured data, linked to assessment processes that work quickly enough for real-world needs.
This is not just adding a feature to existing systems. It usually means redesigning the core assessment infrastructure. Organisations that haven’t started this process are building up technical debt that will cost more over time.
Modernising underwriting through AI and automation
Underwriting is the point where risk assessment turns into a business decision: whether to accept or decline, at what price, and on what terms. It is also where the costs of poor data infrastructure show up most clearly. Manual underwriting processes like document review, reference checks, guideline checks, and approval routing are slow, inconsistent, and hard to audit.
The case for modernisation rests on three improvements that current technology can deliver:
- Automation brings speed and scale. Today’s automated underwriting systems can review thousands of risk factors at once, instead of just 50 to 75 as before, and they work over 70% faster. For simple cases, straight-through processing reduces turnaround from days to minutes, which helps boost conversion rates and cut costs.
- Automation improves consistency and makes audits easier. Model-assisted and rules-based decisions produce reliable, well-documented results. This is important for internal checks and for meeting regulations. Insurers under Solvency II, IFRS 17, or similar rules now need clear audit trails. Explaining why a decision was made, and showing that similar cases are treated the same, is now a basic compliance need.
- Modern systems expand the market insurers can reach. By using more types of data, including non-traditional sources, they can accurately price risks that were once declined due to missing information. This lets insurers serve more customers and offer pricing that better matches each person’s risk, rather than relying on broad averages.
AI underwriting workbenches are already showing results. According to Capgemini's research, underwriters who use workbenches with AI recommendations are about 1.4 times more likely to grow their books. However, the decision flow itself has not been redesigned at most firms. Outcomes still depend heavily on individual judgment rather than a consistent, AI-guided framework. Until that changes, AI might help underwriters work faster, but it doesn't fundamentally change the quality or consistency of decisions.
Reinventing claims processing
Claims are where insurers deliver on their promises. For insurance companies, it is also the biggest part of operational costs. Both efficiency and accuracy here have a direct financial impact. Paying out on fraudulent or inflated claims to overloss. On the other hand, delaying or underpaying valid claims can result in lawsuits, regulatory problems, and losing customers, which affects renewal rates.
Traditional claims processing has built-in inefficiencies. The first notice of loss is often logged by hand. Reserve estimates are made with only partial information at the start. Adjusters can get overloaded, causing backlogs and slower resolutions. Complex commercial claims might take months of reviewing documents, consulting experts, and negotiating before they are settled. All these delays cost money and can hurt the company’s reputation.
Intelligent triage and routing
Modern claims platforms can review new claims as soon as they arrive and sort them by complexity, value, signs of fraud, and the expertise needed. Simple, low-value claims with clear paperwork can be quickly approved and settled through automation. More complex or suspicious claims are sent to specialists, along with a prepared analysis to help them handle the case.
For example, a global insurer working with Capgemini deployed more than 12 AI agents across 12 countries, reporting a 95% straight-through processing rate and 100% accuracy in claims validation. While these numbers look impressive, it is important to be cautious. Real-world accuracy depends on the model/approach used and the data used for validation/testing. Any outlier/contradictory claim request will decrease the accuracy; it’s important to remember that no system based on probability is perfect all the time.
Reserve accuracy and capital efficiency
Reserving, or estimating how much money is needed to cover outstanding claims, has a direct effect on financial statements and regulatory capital requirements. If reserves are set too low, it can create solvency risks. If they are set too high, it ties up money that could be used elsewhere. Predictive models that use past settlement data, litigation rates, medical cost trends, and details of each case can make reserve estimates much more accurate than traditional methods, which improves capital efficiency.
AI-powered fraud detection across the value chain
Insurance fraud drives up costs, and these costs are passed on to policyholders through higher premiums. To catch fraud effectively, insurers need to review the entire claims process, including application data, policy history, third-party records, and the claim itself. It is much easier to spot inconsistencies when all this information is considered together instead of separately.
Traditional rule-based systems with set thresholds are not very effective because they often miss new types of fraud and create too many false alarms. AI and machine learning algorithms can find complex patterns and adapt as fraud schemes change by learning from transaction records. Studies have found that neural networks and Bayesian networks are the most effective AI models for detecting financial fraud.
Natural language processing, especially LLM-based evaluators, are now used to spot identity theft and social engineering attacks. It helps identify phishing attempts and manipulative language in fraudulent claims and emails.
Still, there are challenges in using AI for fraud detection. The biggest issue is a lack of labelled training data, mostly due to a skewed distribution; for example, there may be only 10 to 100 fraud cases out of 10,000 to 1 million claims, so the positive (fraud) class makes up less than 1%. Also, as cybercriminals change their tactics, AI models need regular retraining and updates.
Barriers to scaling AI in risk management
While the benefits of AI-driven risk management are becoming clearer, there are still major barriers to overcome, and they are not purely technical.
- Data quality is often the most overlooked challenge. Insurance companies have collected decades of data stored in incompatible systems, with inconsistent categories, incomplete records, and isolated storage. AI models depend on good data, and poor or fragmented data can lead to misleading results that cause worse decisions than simpler traditional methods.
- With the U.S. adopting AI governance guidelines and the EU AI Act classifying life and health insurance AI as high-risk, the insurance industry must balance AI use with regulations and be realistic about what models can and cannot do. Explainability is becoming a must. Regulators in both the EU and the US want insurers to explain AI decisions in ways policyholders can understand. This is driving demand for Explainable AI (XAI) systems and pushing the industry away from black-box deep learning models that focus on accuracy but lack transparency.
- Algorithmic bias raises both ethical and regulatory concerns. If training data reflects past underwriting practices that unfairly affected certain demographic groups, machine learning models will continue and possibly worsen those biases. This issue has already caught the attention of regulators in the auto and life insurance markets.
- Legacy infrastructure is still a major limitation for many established insurers. Core policy systems built on mainframes from the 1980s and 1990s cannot easily connect with cloud-based AI platforms. Modernising means either costly and risky phased replacements or using advanced middleware that lets old and new systems work together during the transition.
Four strategic questions every insurance leader must answer
The technological shifts described above are not purely operational; they are enabling a fundamentally different business model for insurance and risk management. For technology and business leaders, whether inside insurance organisations or in adjacent industries considering insurance-embedded products, the strategic questions are becoming urgent:
- Is your data infrastructure capable of supporting real-time risk modelling, or are decisions being made on data that is days or weeks old by the time it reaches decision-makers?
- Have you assessed your algorithmic governance framework — not just for performance, but for bias, explainability, and regulatory compliance under frameworks like the EU AI Act?
- What is your legacy modernisation roadmap? Organisations that have not yet begun the transition from monolithic core systems to cloud-native or hybrid architectures are building technical debt that will become progressively more expensive to retire.
- Are you building the talent architecture to support this transition? AI-driven risk management requires a different skill mix: data engineers, ML practitioners, and explainability specialists sitting alongside actuaries and underwriters.
Conclusions
The insurance industry stands at a crossroads: the debate is no longer about adopting AI, but about how boldly organisations are willing to reinvent their operations. The results speak for themselves. Insurers who weave AI into the fabric of their business are surging ahead in revenue, pricing precision, and customer satisfaction. Meanwhile, those clinging to isolated pilots and outdated workflows are weighed down by technical debt and losing ground.
Yet technology is only part of the equation. True frontrunners are reimagining their entire value chain, forging seamless links from risk modelling to underwriting to claims in a dynamic feedback loop. They are reshaping decision-making around AI’s new capabilities and investing equally in people and processes as in technology. The leap from reactive to predictive, from siloed to unified, from slow to real-time is now the standard to meet, not just a distant goal.
The opportunity to lay these foundations is shrinking fast. As agentic AI evolves and the protection gap grows, the divide between innovators and followers will only widen. The moment to act is now, beginning with a candid look at your organisation’s current position.
Related Insights
The breadth of knowledge and understanding that ELEKS has within its walls allows us to leverage that expertise to make superior deliverables for our customers. When you work with ELEKS, you are working with the top 1% of the aptitude and engineering excellence of the whole country.
Right from the start, we really liked ELEKS’ commitment and engagement. They came to us with their best people to try to understand our context, our business idea, and developed the first prototype with us. They were very professional and very customer oriented. I think, without ELEKS it probably would not have been possible to have such a successful product in such a short period of time.
ELEKS has been involved in the development of a number of our consumer-facing websites and mobile applications that allow our customers to easily track their shipments, get the information they need as well as stay in touch with us. We’ve appreciated the level of ELEKS’ expertise, responsiveness and attention to details.
