Why insurance AI fails without legacy system modernisation Only one-third of insurers have seen a positive return on their artificial intelligence investments and just 7% have scaled AI across their organizations, according to Datos Insights research cited by software consultancy Nearform. Nearform reported that a three-week AI-led proof of concept modernized a policy administration platform running 1.5 million lines of aging code for a major insurer, translating buried logic into plain specifications checked by senior engineers and the insurer's domain experts before any new code was written, with the new build writing to the existing database and no downtime to critical processes. Nearform's engineers are now modernizing the entire codebase, with the first module completed in weeks rather than months. Insurers have made significant investment into artificial intelligence over the past few years, hoping it would transform underwriting, claims handling and compliance, but, unfortunately, the results haven't matched the ambition. While most carriers now use AI in some form day-to-day, only a third https://datos-insights.com/reports/ai-implementations-in-2026-ins-2026-102370/ have seen a positive return from that investment, and a mere seven percent have managed to scale it across their organisation. This gap between adoption and impact isn't a failure of the technology itself. Large language models are capable, accessible and improving quickly. The real issue is where insurers are trying to apply them. Many attach AI tools directly onto underwriting, claims and compliance processes that already exist, then wonder why progress stalls. The tools work, but the systems beneath them hold everything back. What happens when AI is bolted onto outdated insurance platforms? Plenty of the processes insurers rely on still run on platforms built decades ago, and some of this software is no longer actively supported, which creates a meaningful security risk on top of the operational drag. Teams spend their time maintaining fragile systems rather than improving them, and new AI tools end up constrained by the same limitations as everything else running on that infrastructure. An AI-native operating model can't function on these foundations. Modernising legacy systems, so both the models and the people using them work in sync, has to come first. But this is also the step most boards and investors treat as too risky to approve, given the cost and disruption traditionally associated with replacing core systems. How can insurers modernise legacy systems without major disruption? At Nearform, we recently worked with a major insurer to modernise a policy administration platform that had sat at the heart of its organisation for more than 30 years. The existing platform ran on 1.5 million lines of aging code, understood by only a handful of people left in the company. The system still worked, which is exactly why nobody had touched it. Replacing it through conventional means was estimated to cost tens of millions of dollars and would have caused significant disruption to daily operations, so the project kept getting pushed down the priority list. Our proof of concept, using AI-led modernisation, took just three weeks. Our team oversaw the AI which read the existing code and translated its buried logic into plain specifications, cataloguing hundreds of forms and over a thousand points of complexity in a matter of hours. Our senior engineers and the insurer’s own domain experts then checked each specification against live data before a single line of new code was written. The new build wrote to the existing database, so nothing needed to be migrated, and the old system stayed live throughout. Critical processes experienced no downtime, so when access to one part of the system was blocked, the team simply pivoted to another area, rather than losing momentum. Nearform’s engineers are now modernising the entire codebase. The first module was completed in weeks, rather than months, with minimal disruption to how the insurer's teams already worked. New capabilities are shared with the insurer's own staff as the project progresses, so the skills and understanding stay in-house rather than locked away with an outside vendor. This approach works because of how responsibilities are divided between machine and human. The AI handles the heavy reading and reproduction, tasks that would otherwise consume months of skilled engineering time. Judgement and sign-off stay firmly with the people in the organisation. That division is the core of what makes an AI-native operating model succeed where bolt-on approaches fail. What value does an AI-native operating model create for underwriting and claims? With the foundational platform modernised, the next step was applying it to generate real value across the business. AI could now work across underwriting and claims processes at consistent quality and speed, freeing underwriters and adjusters to focus on weighing risk rather than chasing data between disconnected systems. Compliance checks, anti-money laundering reviews and regulatory reporting now take far less manual time and effort. Crucially, decisions made by the system can be audited and explained, because the underlying logic has been written down and is finally understood by the people responsible for it. These improvements in reliability, efficiency and expert effectiveness were achieved without the downtime or the enormous bill that a conventional system replacement would have required. What's the real competitive edge for insurers using AI? McKinsey & Company https://www.mckinsey.com/industries/financial-services/our-insights/ai-in-insurance-understanding-the-implications-for-investors estimates that generative AI could add between $50 billion and $70 billion in revenue to the insurance industry. The carriers who actually capture that value won't be the ones who simply spend the most, since every organisation has access to the same large language models. The edge belongs to insurers willing to spend in a measured, prioritised way, starting with the legacy foundations their organisation runs on. This approach reduces risk while improving the underlying codebase faster and more reliably than layering new tools onto old infrastructure ever could. It also respects something boards often overlook: the people inside these organisations who understand the systems, the risk and the regulatory environment better than any model trained from outside it. What should insurers do next to turn AI investment into results? Insurers chasing AI returns need to resist the temptation to buy another tool and attach it to what already exists. The organisations seeing real gains are the ones treating legacy modernisation as the starting point, not an afterthought. That means auditing which core systems are holding teams back, identifying where thoughtful use of AI can genuinely remove manual effort, rather than just add a layer of automation, and keeping human judgement at the centre of every decision the system informs. Carriers that get this sequence right will spend less, move faster and keep their best people engaged in the work that actually matters. Those that keep bolting AI onto ageing platforms will likely keep seeing the same disappointing returns, regardless of how much they invest. But wait - there's more. Nearform publishes real-world learnings on data & AI, engineering, and digital strategy - with more merged in weekly. Insights Perspectives on AI in engineering, product development, and strategy, for enterprise executives. Community Deep dives and tutorials by engineers, for engineers. You may also like The first step businesses can take to move legacy systems to agentic AI Bulbul Pandya https://nearform.com/authors/bulbul-pandya , Colin Houlihan · 28 Oct 2026 · 3 min read