In the race to adopt AI, many insurers might just assume that wherever the technology can be integrated, it will automatically provide the most value, but simply adding a model to their ecosystem is not the way to maximise the potential of AI.
AI models are only as good as the foundation they are built upon, and insurers need to remember that when they look to adopt the technology across their ecosystem. Put simply, Simon Fagg, SVP and Go To Market Leader at IntellectAI, noted that “Pilots fail the move to production on the record, not the model.”
This sentiment was shared by Allison Lagosh, V.P. and head of compliance at Saifr. She said, “Insurers need more than models.” To get the most out of the technology, an insurer needs data, governance frameworks, regulatory requirements, human oversight and workflow integration. This can help AI models reach their full potential, while also having guardrails in place to prevent mistakes.
When it comes to implementing the technology, Dan Simmons, managing director and founder at Quensus, sees three core aspects needed for insurers to get real commercial value from AI. The first is to consolidate IoT telemetry, claims history, and policy records from siloed departments into a unified repository. Second, they need to build a robust deployment pipeline so models can be continually monitored for drift and updated without the need to start from scratch. Finally, they should bring actuaries, underwriters and data scientists together to ensure AI targets real challenges.
One way to maximise AI’s ROI is to ensure it is connected to relevant systems and data. A solution that is independent can only do so much. By contrast, a tool that connects datasets and departments can provide greater efficiency and better insights that remain consistent across the business.
This is something Matt Twist, vice president sales EMEA at Earnix, also believes. He explained that AI should connect to business processes and measurable outcomes rather than treating it as a standalone experiment. “This means connected, accessible data across pricing, underwriting, claims and customer functions, rather than AI operating on fragmented information. As well as decisioning infrastructure that allows AI outputs to be embedded directly into business workflows and acted on in real time.”
When transitioning from pilots to effective use cases, IntellectAI’s Fagg believes four measures are needed. The first is implementing an effective governance framework to manage the widespread adoption of independent solutions across organisations. They said, “The point-solution approach has dominated, and the returns have been woeful. Token cost has escalated as everyone has jumped in feet-first to avoid the fear of missing out.
“Meanwhile, this ground-up approach to problem-solving has not been through the rigor of the interconnected governance process.”
Secondly, competitive advantage will come down to the insurer’s ability to attain a differing perception of risk vs profit on presented risk. The third measure is to ensure real-time data capture, classification, normalisation, clearance, triage and enrichment to ensure real-time decision-making is possible. “All this must be done within a predetermined cost per transaction offset against profit and risk. You cannot verify and update post-transaction data, as it’s not efficient or cost-effective,” they said. “Most pilots survive contact with production only to discover the constraint was never product or model accuracy; it was that half the fields the AI needed to reason are incomplete.”
Finally, Fagg emphasises that insurers need to start with an end in sight, but flexibility is needed as it will likely throw questions that were never conceived of at the beginning. “Yes, you can conceive and build at a 10-100x rate and at a quality and resolution never envisaged until recently. But you need to keep the basics collated: what is your required end-state reporting to drive decisions, compliance, insights? What resolution is required on the data? What frequency are updates to positions required? How dynamic should your reporting be? What type of analytics and insights will enable STP and manage by exception? Is data in line with your organisation’s key drivers, market alignment? Can you be nimble when appetite changes?”
Balancing speed and governance
AI adoption can often feel like a balancing act. On one side there is a demand to move fast to ensure competitors do not get too far ahead and steal market share. On the other, there is a need to ensure strong governance frameworks are in place that mean AI tools are accurate, explainable and meet regulatory standards.
There are no standardised regulations for AI, with some countries opting for specific rules, like the EU’s AI Act; others, such as the US, have issued various guidelines, and places like the UK have existing regulations where AI tools will need to adhere to. This creates a regulatory minefield for insurers to navigate and opaque AI models are not enough to satisfy regulators.
Quensus’ Simmons believes the answer rests in keeping humans involved for high-stakes decisions. He said, “Insurers should deploy explainable AI frameworks for anything affecting pricing or claims denials, ensuring the underlying logic can be audited easily. Continuous testing for model bias, combined with human sign-off on edge cases, lets carriers move quickly on routine automation while keeping regulators, underwriters, and policyholders on side.”
One thing that became clear from the industry experts is that insurers should not see speed and governance as opposing forces, but rather as partners.
Saifr’s Lagosh noted, “The goal is not to choose between innovation and governance. Insurers should adopt AI within a framework that embeds explainability, human review, data controls, and regulatory compliance requirements and oversight from the outset. They should also be able to track and implement legal and product changes and limitations.
“Insurers can be required to report to a regulator that their use of AI fits within historical limitations, especially relating to underwriting. Specifically, just because AI can help provide a more accurate answer from an actuarial standpoint does not mean that it can be utilized legally.”
Melanie Hayes, co-founder at KYND, also believed that speed and responsible AI are not competing priorities. “The key is to be clear about where AI can accelerate decision-making and where human judgement and oversight remain essential.”
As such, Hayes explained that insurers must apply the greatest scrutiny where the consequences of an error are highest. They will need to understand what data is used, how AI supports a decision, whether it can be explained and evidenced, and who is accountable for the outcome. “If those questions can’t be answered, the issue isn’t that the technology is moving too quickly, it’s that the right governance isn’t in place.”
Adding to the complexity of the situation is the fact that it is not just insurers that are adopting AI, Hayes said. Their insureds will also be adopting the technology, and insurers will not have a clear view of where and how it is deployed. This just creates new exposures across cyber, professional indemnity, employment practices and other areas that need accounting for.
“So, for insurers, responsible AI adoption ultimately comes down to visibility and accountability. They need both within their own use of AI and across the risks they underwrite. The better the evidence available to them, the more confidently they can embrace AI without compromising the standards of accuracy, privacy, fairness and regulatory compliance they are expected to uphold.”
What needs to change
The era of pilots is coming to an end. Rather than just experimenting with various solutions, insurance companies are moving ahead with adoption. For this to happen effectively, there will need to be a change within their operating model and culture.
At the centre of this, Simmons emphasised, is the need to move away from the notion that humans will be replaced. “Scaling AI requires shifting the internal mindset from employee replacement to human augmentation.” This means framing tools as assistants that can manage heavy data crunching to save time. Operational teams should also be rewarded for proactive loss prevention and verified risk improvement. Finally, Simmons noted that companies should not expect perfect, massive rollouts. Instead, AI models should be seen as evolving tools that need constant feedback from all parties to improve.
Part of that culture change involves ensuring everyone is on the same page. Twist stated that as AI becomes incorporated into daily operational processes, business, technology, data and governance teams need to work together rather than having their own siloed AI tools.
Insurers will need to have clear ownership and accountability in place and employees will need to have tools that can support their expertise, rather than replace them, to ensure internal trust and adoption. This will also require a culture of continuous improvement to ensure models, rules and processes evolve with the market and demands.
He said, “The goal is not to use AI for the sake of it, but to connect it to real business processes so insurers can make faster, smarter and more consistent decisions.”
Finally, Fagg believes that smooth AI adoption requires the creation of a new role within the business. They said, “There is a role some carriers have not created yet. The role involves reviewing and reconciling agent-generated outputs as part of a human-in-the-loop process, a broad spectrum of checks, and the ability to resolve guideline issues and take action. The role sits as oversight between the underwriter and the model risk function and is independent of both. Compliance, peer review, and the auditor role are a key part of what ensures the systems remain controlled and accountable.”
Big opportunities for the technology
AI has the potential to transform most facets within insurance. Its growing capabilities and flexibility mean it could find its way into most areas of an insurance company. Once firms can pave the way for widespread adoption, there are many areas it can support.
For instance, Earnix’s Twist sees some of the best opportunities being in pricing and underwriting, because they can offer faster analysis, precise risk assessment and quicker response times. Elsewhere, it can bolster customer engagement through more personalized and timely interactions, while within operational decision-making the technology can automate routine analysis and recommendations. Finally, Twist sees significant benefits in greater cross-functional coordination, enabling pricing, underwriting, claims and customer decisions to work together.
Improving decision-making capabilities was the most common area people pointed to for where AI can improve insurance. Its ability to collate information helps underwriters gain a full picture of clients and identify signals earlier, while also allowing them to focus on making judgements rather than manual admin tasks.
One area that is not discussed as often is how clients are using AI and how that impacts risk. Hayes explained that, while many insurers are focused on their own use of the technology, many are forgetting the businesses they insure are also exploring the technology. Without that visibility, the technology creates blind spots in risk assessment. Whether it is an AI making an error, a vendor outage that halts business or a security vulnerability that leaves businesses exposed, how a client uses AI is vital knowledge for an insurer. Tools are needed that can provide visibility into the technologies an organisation uses.
“For me, that’s where some of the real potential lies. AI shouldn’t simply make existing insurance processes faster; it should help insurers understand risk better. Giving underwriters better visibility, at the point they’re making a decision, means they can move faster and with greater confidence.”
IntellectAI’s Fagg noted that the industry has focused heavily on point solutions—especially the automated underwriting “front door,” including intake, clearance, triage, insights, and workflow optimization through to bind.
But there is a major gap after bind. Carriers also need a holistic view of the portfolio: scenario modeling, full-spectrum claims insights, industry loss trends, comparison of in-force risks against what was originally underwritten, pre-renewal assessment, and dynamic capital and capacity management. This needs to consider both inward and outward perspectives.
Historically, the approach has been too narrow and asymmetrical. Post-bind analysis has not been treated as an important input into pre-bind decisions. The focus has been more on individual policies than the portfolio, individual risk ratings rather than marginal contribution, and limited rather than multi-dimensional analysis.
AI now makes it possible to move beyond simple trend analysis, Fagg stated. An AI agent can understand underwriting guidelines, operate like an experienced subject-matter expert, evaluate a presented risk against internal and external data, and apply a different reasoning process than a traditional underwriter.
The next step is for AI to continuously identify meaningful changes in the internal and external environment and recommend actions, such as adjustments, benchmarking, anomaly detection, risk alerts, and profitability improvements. Importantly, this should go beyond simply comparing an individual policy to the existing book.
They said, “Reading the whole bound book against the carrier’s own guideline turns a lagging indicator into a leading one. Guideline drift and loss-ratio drift can be caught early to mitigate underwriting patterns, educate the underwriting team, advise or take specific actions within the realms of the corporate-defined boundaries of the AI.”
Read the latest InsurTech news
Copyright © 2026 InsurTech Analyst
