Last week, we had the pleasure of collaborating with InsTech on a panel and networking event in the heart of the London Market: the Lloyd's Building.
Between drinks and canapés, we brought together a panel of insurance experts to discuss one of the industry's biggest challenges: moving AI beyond experimentation and into production at scale.
Stefan Schrijnen, Partner, Commercial & Specialty Insurance Lead at Capco, hosted the discussion, joined by three fantastic panellists:
- Sasha Haco, CEO & Co-founder at Unitary
- Hasani Jess, Group CTO at Aventum
- Jade Mounir, SVP of Finance at Flock
The conversation covered everything from cultural resistance and legacy systems to the business case for AI and the rising expectations around what it should deliver.
Here are five of the biggest takeaways from the discussion.
1. Scaling AI is much harder than adopting it
Insurance is leading the way in AI adoption. According to the Bank of England and FCA's 2026 survey, 93% of UK insurers are already using AI. Yet BCG research found that just 7% of insurers surveyed had successfully scaled AI across their organisations. Our findings at a recent roundtable with Artex Risk reflected a similar picture.
So why the disconnect? One distinction highlighted by the panel was between personal productivity AI and systemic AI.
Personal productivity tools, such as copilots that help employees write emails, analyse information or prepare presentations, are relatively easy to introduce. But systemic AI, embedded into the core processes and systems that run a business, is a much bigger undertaking.
Scaling systemic AI often means navigating legacy technology, fragmented data, complex workflows and longstanding ways of working. Unlike individual productivity tools, it can't simply be rolled out to employees and left to run. It requires changes to how work is organised, how different systems interact, and how decisions are made.
That's why widespread AI adoption hasn't yet translated into widespread business transformation. The technology itself is only one part of the challenge. The harder work is making it function reliably within the complexity of a real organisation.
2. AI adoption works best when employees drive the change
One of the strongest themes of the discussion was that successful AI adoption depends on getting employees bought into the change. As Hasani explained, people don't necessarily resist change because they're incapable of changing, but because they don't believe it will benefit them. Helping employees understand what's changing, why it matters, and how it will improve their working lives is essential to bringing them on board.
Jade shared how Flock is empowering its employees to actively drive AI adoption. Rather than leaving innovation to the technology team, Flock encourages the people closest to a process to identify opportunities for improvement. Leadership makes dedicated time for experimentation, including a bi-weekly AI breakfast where employees showcase practical use cases.
In one example, Flock's finance team built an application to process complex, bespoke deals that underwriters had previously been told couldn't be handled. Engineering is now using it as the blueprint to build the capability into Flock's core platform. As Jade explained:
"The people closest to a process are often the ones best placed to see where AI could make a difference. Our approach is to empower those teams to experiment, then bring in engineering to help turn the best ideas into scalable solutions."
3. Perfect data is no longer a prerequisite for getting started
Legacy systems, fragmented data and inconsistent processes are familiar challenges across insurance. But as Sasha argued, waiting until everything is cleaned up and modernised can mean spending months preparing for automation without seeing any return.
"The reality is that most insurers have legacy systems and messy data. But that doesn't mean they can't automate. The technology exists to work with those systems today, so businesses can start seeing value without waiting for perfect data or major system overhauls." — Sasha Haco
Hasani also highlighted how AI has fundamentally changed what's possible with unstructured data. Historically, technology transformation has often required businesses to force complex information into standardised formats. AI is increasingly capable of interpreting unstructured documents directly, which is particularly valuable in a market built around bespoke risks and policy terms.
Of course, accuracy requirements vary depending on the task, and appropriate controls and human oversight remain essential. But imperfect data and legacy infrastructure don't necessarily have to prevent insurers from getting started.
4. AI ROI starts with cost, but it doesn't end there
When evaluating AI investments, the conversation often focuses on cost savings, hours saved, or headcount reduction. However, the panel agreed that the value of automation can extend well beyond these measures.
For many insurance businesses, particularly those operating at lower volumes, automation might not lead to direct headcount savings. But faster broker responses, fewer errors, better customer experiences, and more time for relationship-building can all contribute to a company's competitive advantage. In a softening market, being easier to do business with can make a real difference to customer acquisition and retention.
Hasani introduced the concept of a quality-of-life investment: if an initiative doesn't directly increase revenue, reduce costs or mitigate risk, it may still be worth pursuing because it improves employees' working lives. The important thing is to recognise this explicitly and establish how the wider benefits, such as improved retention and reduced recruitment costs, will be measured.
5. Not every successful AI pilot should make it into production
An interesting question raised during the panel was how businesses should decide which AI pilots deserve further investment. Hasani challenged the assumption that a high proportion of pilots should progress into production:
"We shouldn't measure the success of AI experimentation by how many pilots make it into production. Learning that something isn't worth scaling is a useful outcome too. What matters is making the right investment decisions." — Hasani Jess
The first consideration should be strategic alignment: does the initiative support what the business is trying to achieve? Only then should the business case determine whether it's worth scaling. A pilot might demonstrate impressive technical accuracy, but that alone doesn't guarantee meaningful business value.
The discussion also touched on the economics of scaling, particularly the unpredictability of AI token costs as usage grows. Using AI selectively, choosing models appropriate to each task and relying on conventional software where possible can help make automation more cost-effective and predictable. Ultimately, success isn't about deploying as much AI as possible, but investing in the applications that deliver the greatest value.
Moving beyond experimentation
The overarching message from the discussion was that scaling AI is as much an organisational challenge as a technical one. Insurers have shown they're willing to experiment, but turning that experimentation into lasting business impact requires the right people, processes, and measures of success.
A successful pilot proves that something can work. The next challenge is proving that it's worth scaling.
A big thank you to InsTech, Stefan, Hasani and Jade for such an insightful discussion, and to everyone who joined us at Lloyd's for the evening.
