As the insurance industry manages a whirlwind of challenges: evolving risks, technological disruptions, and shifting customer expectations, those organisations leaning into AI are finding success.  

They are already unlocking significant productivity gains and cost savings while laying the groundwork for disruptive innovation. Right now, these businesses are building strong data foundations and developing an understanding of the different types of AI, their benefits, and how to apply these for impact across the entire insurance value chain. 

‍

We’re working with one global insurance firm projecting a $100m increase in Gross Written Premiums as a result of transforming their underwriting process - and they’re already getting started on this journey. Many are now at a phase where they begin scaling AI-powered innovations across the Enterprise. 

‍

In this article we set our vision for what the AI opportunity is within insurance and our key recommendations to accelerate and scale AI adoption across your organisation, plus the questions you need to answer to choose the right AI initiatives for your business. But let’s set the scene first. 

‍

Insurance Industry's AI Maturity

‍

The insurance companies we work with fall within three broad categories. 

‍

• "Conservative adopters” generally find themselves in the “trough of disillusionment”. Having enabled ChatGPT at the enterprise level, encouraged experiments and new use cases through AI committees, the results have been underwhelming. This broad approach was useful to understand Generative AI’s benefits for content creation and language generation, unstructured data summarisation  (think documents, voice, pictures) and entity / attribute / relationships recognition and extraction. Ultimately this drove bottom-up buy in and drove AI literacy through the organisation.

• “Forward thinkers” have identified a small number of high impact applications of AI, focussing on automating key processes while keeping humans in the loop and significantly increasing their productivity. These organisations have been more deliberate about how to apply AI and are entering a scale up phase. Through being methodical, they have unlocked even bigger opportunities which can now address key industry challenges in a more fundamental way. Despite the new set of challenges that this scale may bring, they are better placed to overcome these and continue innovating.
• Finally, a few “innovators” are now building intelligent systems which go beyond automation to take humans out of the loop and drive more powerful outcomes, using early advances related to reasoning models (e.g. OpenAI o1) and Agentic AI.  

‍

It is worth noting that while there is a disproportionate amount of focus on Generative AI, general machine learning and deep learning techniques should also be part of any comprehensive AI strategy. Predictive analytics, classification, clustering, and segmentation, for example, can be useful for churn prediction, customer categorisation or better underwriting risk; on the other hand, simulation techniques coming from scenario modelling can help with actuarial modelling, for instance to model climate risk and impact. The trick here is to allow your actuaries, statisticians and engineers the freedom to experiment, while giving them the platforms to take these ideas into production securely, scalably and repeatably. 

‍

Where Does AI Impact the Insurance Value Chain?

‍

If we look at the value chain of insurance, from product creation all the way to managing the lifecycle of products, there are many opportunities for AI for immediate cost reduction opportunities and more long-term growth opportunities - both in general and speciality insurance.

‍

‍

The Four Key Ingredients to Scaling AI Successfully

‍

So how should insurers go about unlocking these opportunities, whether they reside in the “conservative organic adopters” or the “forward thinking” camp? What are the “innovators” doing today?We believe there are four key ingredients to successfully scaling AI at the enterprise level:

‍

• Build comprehensive business cases, preparing for AI Agents,
• Create various funding mechanisms to scale AI adoption at pace,
• Continuously iterate over the Data Strategy,
• Design an end to end framework to deliver AI to production and govern it.

‍

1. Build Comprehensive Business Cases, Preparing for AI Agents

‍

Insurance organisations venturing beyond tactical AI experimentation need to consider how business cases can support both immediate value and future capabilities. Cost reduction is a zero sum game after a while, so it’s important to look at the other benefits that come with any AI implementation, such as new revenue streams, improved customer satisfaction and risk reduction.

‍

When developing these business cases, consider exploring these questions:

‍

1. How might AI transform core operational processes? Consider areas where manual processing creates bottlenecks or inaccuracies in claims, underwriting, or policy administration.
2. Where do customers experience friction in their journey? Ensure you can measure/quantify and then reflect on moments that create dissatisfaction or drive negative or high contact volumes.
3. Which risk assessment approaches might benefit from incorporating alternative data sources? Think about where current models struggle with emerging risks.
4. Which unique datasets does the organisation possess and how can they be exploited through automated intelligence? Could those unique datasets be matured enough to be monetised into the market? Or do they give a competitive edge?
5. How does the business case take into account the accumulative benefits of an Enterprise AI platform? The more AI “engines” and data products are developed across the organisation, the more the marginal cost of the next AI goes down.
6. What are Insurtech disruptors doing which could be applied?

‍

Looking forward, consider how today's individual AI capabilities might evolve into more sophisticated agent systems. This requires thinking about:

‍

1. How might modular AI components developed today combine into more comprehensive solutions tomorrow?
2. What skills and knowledge transfer will help domain experts collaborate effectively with technical teams?
3. How will you balance automation with human expertise, particularly for complex or sensitive decisions? Where could automation make better decisions through the processing of a (much) broader set of data?  How will you incorporate human-in-the-loop feedback into your models to improve them? 
4. Which processes currently automated within the organisation can become part of a broader workflow, powered by reasoning Agents?

‍

AI Agents bring an additional dimension to what has been done to date with Generative AI: the ability to autonomously make decisions on an optimal course of action, executing various tasks leveraging Enterprise systems. This autonomy in planning needs to be organised with the right controls in place to ensure consistent, ethical and business aligned outcomes - but it opens the door to more ambitious use cases and outcomes which go beyond the pure “automation + human in the loop”.

‍

2. Create Various Funding Mechanisms to Scale AI Adoption at Pace

‍

Traditional inward investment approaches often struggle to accommodate AI's experimental nature and cross-functional impact. The pace of AI innovation is also unprecedented and what might have been challenging to build 6 months ago is now built into the latest versions of a product or a platform.

‍

This often collides with the traditional yearly funding cycle most organisations have. The requires an alternative approach:

‍

1. How might you create funding structures that allow for exploration without demanding immediate returns?
2. What would a portfolio approach to AI investment look like, balancing quick wins with transformative opportunities?
3. How could business units collaborate on shared AI capabilities that benefit multiple functions? Would this be funded from a central pot or through co-investment by business areas?
4. What internal service models might help democratise access to AI capabilities across the organisation and promote collaboration and reuse?
5. What investments in your people would allow them to do more with less when it comes to AI?

‍

These questions need to be answered from the very top of the organisation and require business and technology vision - or the price to pay will be lagging behind. 

‍

3. Design an End-to-End Framework to Deliver AI to Production and Govern it

‍

An end-to-end framework to transform ideas into products is crucial if your organisation is serious about productionising AI use cases while minimising risk and enabling reuse of individual components. It needs to consider the following:

‍

1. How might you standardise the development and deployment of AI models while maintaining flexibility for innovation?
2. What diverse perspectives should be represented when evaluating potential ethical implications of AI applications?
3. How will you monitor AI systems in production to ensure they continue performing as expected? And how do you link technical performance with business metrics?
4. What approach will you take to designing AI that augments rather than replaces human judgment?
5. How might cross-functional teams share knowledge and perspective throughout the AI development lifecycle?

‍

Governing AI and managing the risks related to its usage are nascent fields, with new industry standards, frameworks and regulations still being developed and iterated across the globe. Finding a proportionate response to this challenge is key, balancing sufficient assurance of AI systems’ robustness and reliability, appropriately managing risks related to data used in these systems and allowing for innovation to follow its cycle without unnecessary red tape. 

‍

A few questions to be asked about an insurer’s AI maturity may help guide the right responses:

‍

1. Is data already part of the official risk taxonomy and does it get reviewed regularly to include novel risks posed by AI?
2. Is the usage of AI constrained to proof of concepts and experiments or are there use cases in production?
3. Is sensitive data, such as personal information and material non-public information, being used to train or use AI systems?
4. To what degree are AI models developed and hosted internally, used from open source libraries or purchased from third parties? What about the usage of AI “bolt-ons” such as SaaS and productivity tools offering transcription, summarisation and other content generation tasks? 
5. Do you have the right governance processes in place to guard against immature or even dangerous tooling being used, while also fostering experimentation and innovation and bringing it quickly into production?

‍

Reflect on whether your current technology governance frameworks adequately address AI's unique characteristics. Consider how traditional IT governance and testing might need to evolve to accommodate the probabilistic nature of AI outcomes. And invest in an LLMOps strategy to support the deployment of the AI strategy at scale so teams can continuously put models in production, safely and in line with the strategy. 

‍

4. Continuously Iterate Over The Data Strategy

‍

A sound data strategy is paramount for successfully harnessing the power of AI, laying the foundations for more advanced use cases, reducing time-to-insight and bringing measurable business value. Some of the key considerations for an AI-ready data strategy:

‍

1. Alignment to the business strategy is key to help prioritise AI use cases that align with overall organisational objectives and can be effectively measured. Data capabilities can be uplifted gradually as part of delivering use cases, linking foundational work with real value for the business.
2. Implementing solid data quality and governance frameworks will help get the right data to the right AI applications at the right times – with full lineage and auditability of data sources and accountability for the data itself. This is especially important for unstructured or dark data that tends to be drawn in from disparate sources/repositories and emails unnoticed, with little control.
3. Data required for risk modelling often needs to cover multiple decades: the data strategy needs to offer a solution to maintain the quality of the data over time, even if it is hosted on legacy platforms.
4. Data integration and harmonisation, whether physically or virtually, will accelerate the delivery of AI use cases, such as creation of more comprehensive risk profiles that lead to more accurate underwriting decisions or the ability to price more dynamically.

‍

The points above could still be true for a data strategy written 15 years ago. But the recent pace of technological advancement brings a need to be agile and iterate, not only with use cases, but with the strategy itself. New AI systems, new architecture paradigms and new datasets are becoming available at an unprecedented rate. This calls for a data strategy that is living and breathing, ready to be adapted or even course-corrected as new regulation or technological advancements arise.

‍

A data strategy that makes high quality data more accessible to those who need it allows insurers to capitalise on the latest AI techniques with greater agility, building the muscle memory to continue innovating and solving complex problems with technology. 

‍

The Time for Tentative Experimentation Has Passed

‍

The insurance industry stands at an inflection point where AI adoption will increasingly differentiate market leaders from laggards. If you aren’t spearheading this, we guarantee your competitors are. As we've explored, the most successful insurance organisations are moving beyond experimental AI implementations to strategic, enterprise-wide adoption that addresses fundamental industry challenges.

‍

Forward-thinking insurers are building comprehensive business cases that balance immediate returns with strategic positioning for an AI-enabled future. They're creating innovative funding mechanisms that accelerate development and adoption while managing investment risk. Their data strategies continuously evolve to incorporate new sources and capabilities while maintaining governance standards. And they're implementing end-to-end frameworks that ensure AI solutions move reliably from concept to production with appropriate governance.

‍

The path forward requires both technical excellence and organisational transformation. Insurance leaders must foster cultures that embrace informed, data-driven decision-making while maintaining the human judgement and relationship focus that has always differentiated great insurers.

‍

Those who successfully navigate this transformation will create significant competitive advantages: more accurate risk assessment, personalised customer experiences, streamlined operations, and the agility to respond to an increasingly volatile environment. The AI revolution in insurance isn't merely about technology adoption –  it's about reimagining the fundamental ways insurers create value for customers and shareholders alike.

‍

As we look ahead, the most successful insurers will be those who view AI not as a standalone capability but as a transformative force integrated throughout their business model, culture, and operations. The time for tentative experimentation has passed. The era of comprehensive AI-powered reinvention has begun.

‍