From AI pilots to production: protecting margins in a softening insurance market

A softening insurance market is good news for buyers, but it puts more pressure on insurer margins. Aon’s September 2026 UK market outlook describes an overall soft market, with plentiful capacity, strong competition and premium reductions across most commercial lines. For insurers, MGAs and brokers, profitable growth will depend on disciplined underwriting, tighter expense control and strong broker and customer relationships.  

AI can support each of these priorities, but only when it improves real work. 

Where AI can make a difference 

Cost matters, but the bigger opportunity is to help people make better decisions, handle more work and spend less time on routine administration. 

Experienced insurance teams still spend too much time finding information, reviewing documents, rekeying data and moving work between systems that don’t connect. AI can triage submissions, extract information from policy and claims documents and give teams the right knowledge when they need it. 

That could help underwriters assess more of the risks they want to write, claims teams respond faster and brokers spend more time with clients. The value comes from removing avoidable effort while keeping informed people in control. 

Marsh’s new Broker WorkBench offers one example. The AI-assisted platform combines intelligent data processing with streamlined placement workflows. Marsh says it aims to reduce placement times from the industry average of two to four weeks to days or hours, while its brokers will continue to approve every recommended placement.  

That balance matters. It is not technology instead of people; it is technology removing routine administration so people can deliver more value. 

Insurance remains a judgement-led and relationship-driven industry, so AI shouldn’t become a headcount conversation. Used well, it creates capacity for better risk selection, faster claims decisions, more responsive broker service and the development of skilled people. 

More pilots are not the answer 

The real challenge is no longer proving that AI can do something useful. It is turning a promising use case into a reliable part of day-to-day operations. 

TXP’s 2026 research, The AI Value Gap, shows how difficult that move remains. Among 200 UK mid-market decision-makers responsible for IT and AI strategy, including leaders in banking and insurance, 47% had scaled AI pilots into production. 

At the same time, 49% said AI initiatives had failed to deliver value or had been underwhelming, while 61% had paused or wound down projects that did not perform as expected.  

The barriers were practical. Leaders pointed to the cost of moving from pilot to production, governance and risk concerns, poor data quality or accessibility, and the difficulty of integrating AI with existing systems.  

Insurance teams will recognise these problems. Policy, claims, underwriting and broker data often sits across legacy platforms, inboxes, PDFs and spreadsheets. A standalone AI tool won’t resolve that fragmentation. Without the right data, workflow and integration around it, another tool can add complexity rather than remove it. 

One finding from the research is particularly important. Seventy-seven per cent of leaders wished they had invested more time and budget in discovery before starting. A further 69% said critical AI knowledge remained with specific individuals rather than being documented and accessible.  

The problem isn’t a shortage of AI tools. It is that the technology often lacks the business context needed to work reliably. 

What it takes to move AI into production 

In a softening market, another impressive pilot isn’t enough. Insurers need a clear line between AI investment and a measurable operational outcome. 

That could mean reducing handling time, reaching quote or claims decisions faster, improving conversion and retention, increasing underwriting capacity or giving brokers and customers a better service. 

The starting point should be the business problem, not the technology. 

That means working with the people who understand the process and its exceptions. Their knowledge needs to be captured before the workflow can be redesigned around it. The data must be accessible and trusted, and the solution must connect with the wider technology estate rather than sit alongside it. 

Governance and human oversight also need to be part of the design from the start. Teams should be clear about what the technology can automate, where it should support a decision and where human judgement must remain in control. 

This work can sound less exciting than launching a pilot, but it is what turns an isolated demonstration into a dependable workflow. 

Turning AI ambition into measurable value 

TXP helps insurers turn a defined operational problem into a governed AI workflow that can operate in the real world. 

We start by understanding the problem and capturing the knowledge behind it. We then bring together the data, workflow, integration and governance needed to move the right use cases into production. 

This discovery-led approach helps teams test their assumptions before committing to a larger investment. It also creates a clearer link between the technology and the outcome it needs to deliver. 

In a market where every point of margin, every broker interaction and every skilled hour matters, the goal isn’t another pilot. It is measurable value, delivered by technology and people working better together.