In a nutshell
- Using AI successfully in insurance claims goes beyond simple automation – it requires a well-structured ecosystem. High-quality inputs, seamless workflows and ongoing optimisation are key to making it work.
- Customers value speed and accuracy above all. Given the choice, they prefer AI-driven processes that provide fast, reliable answers over slower traditional methods.
- AI requires continuous investment. To deliver real impact, businesses must document AI models, refine processes and commit to long-term improvements.
As a growth- and customer-focused expert, Ed Ackerman has been helping businesses scale through strategic partnerships for over 15 years – from start-ups like Onfido to established giants like Google.
He started his career as a consultant at Deloitte before joining Google in London, where he became Head of Channel Partnerships for UK & Ireland. Prior to joining Qover, Ed was the Director of Alliances and Partnerships at Onfido. At Qover, he's gone from leading the customer success function to Chief Customer Officer and now Chief Operating Officer – overseeing Qover’s delivery organisation and customer-facing teams.
In 2024, Qover’s CEO explored how the insurance sector could harness AI to maximise customer outcomes and efficiency. He highlighted that the natural entry point for AI adoption in insurance – like in most industries – would be in frontline customer interactions.
Fast forward 2026, and we've now had over a year of hands-on experience putting these principles into practice – with results that have both surprised and humbled us. This post – the first in our AI blog series – focuses on the lessons we've learned about the business impact of AI. Since I (Chief Operating Officer at Qover Ed Ackerman, reporting for duty 🫡), wrote it in March of 2025 as Chief Customer Officer, we've updated it to reflect where we are today.
We don’t have all the answers – far from it. But we do want to contribute to the ongoing conversation about the real-world utility of AI and encourage others to do the same.
Navigating the AI hype cycle
Despite ChatGPT-4 being widely released only a couple of years ago, it feels as though we’ve already reached the trough of disillusionment. The initial excitement around generative AI tools has settled, and businesses are now grappling with the real challenge: turning AI potential into meaningful outcomes.
Big Tech and startup investors have poured billions into AI, but whether these investments will drive the mass adoption to fulfil these big bets remains an open question. Some analysts predicted 2025 as the make-or-break year for AI – would we look back on it as the year AI revolutionised industries at scale? Or would we see a burst of the AI bubble, followed by a slower, more measured adoption curve?
At Qover, we see AI’s potential and are investing heavily in its capabilities. However, I also acknowledge the reality: as pioneers in our sector, we will experience both the advantages (first-mover benefits) and the challenges (navigating the unknown and course-correcting when necessary). We’re committed to being at the forefront of AI adoption in insurance, but we want to do so responsibly, ensuring the needs of our partners and their customers remain the priority.
Writing this in 2026, I think it's fair to say 2025 was neither the year AI revolutionised everything overnight, nor the year the bubble burst. It was, quietly, the year the gap between companies who committed early and those who waited started to show. At Qover, we're glad we moved when we did. Not because we got everything right, but because we built enough of a foundation to learn quickly.
Qover’s customer sentiment analysis showed us that customers value speed almost above all else – speed of accessing answers to queries and speed of claim settlement, as Accenture has also found.
This insight led us to prioritise AI claims processing applications that would significantly improve speed to outcome and resolution efficiency. However, we recognised that achieving these gains couldn’t come at the expense of customer satisfaction or quality of service. Our goal was to enhance both speed and accuracy simultaneously.
To do this, we focused on two key AI applications:
- Supporting written communication with customers: using generative AI to refine, speed up and enhance responses.
- Enhancing claim evidence capture: automatically extracting and analysing documentary evidence to expedite claim decisions.
Watch the webinar: benefits & risk of AI in claims →
The mantra: hurry up and slow down
In our experience, hurry up and slow down is the mantra. AI is undeniably transformative. However, expecting AI alone to generate substantial returns overnight without broader operational changes is perhaps naive.
AI-driven improvements require investment in multiple areas: for example, better documentation to feed AI models, enhanced process flows to manage human-in-the-loop exceptions and revised quality assurance mechanisms to measure the effectiveness of AI-human collaboration.
At Qover, implementing AI in customer service and claims processing was never going to be dependency-free. Instead, it demanded significant improvements to our supporting capabilities. Here’s what we’ve learned along the way.
.png)
Lesson 1: AI’s power lies in high-quality inputs
A key finding in our AI journey is that people are, in general, quite happy to interact with AI. The AI Act required that we clearly indicate when AI is being used and provide an alternative, non-AI option for those who prefer it.
In our claims project, we found that less than 1% of claimants opted for the non-AI version. This suggests that what truly matters is not whether the agent is human or AI, but its competence. A poorly trained human agent can be a major turn-off, whereas a well-trained AI agent that delivers fast and accurate responses is well accepted.
Over time, as AI systems become even more capable, I believe users will care less about who or what is on the other end, as long as the service is competent and efficient.
AI-powered agents are impressively articulate, but their accuracy is only as good as their training data. Generative AI tools, like ChatGPT and Gemini, are excellent at producing professional, well-structured responses. However, they need well-organised, up-to-date and highly detailed information to deliver truly useful answers.
In the insurance sector, policy documents are critical, but they don’t always cover practical, real-world applications. Human agents instinctively know the nuances of FAQs and edge cases – AI, on the other hand, needs those insights documented explicitly.
Our approach at Qover:
- When deploying an AI agent for a new product, we spend the first few weeks reviewing every response manually. Our human agents provide feedback, refining AI outputs until it's as good as human output. That way, our AI agent achieves customer satisfaction scores comparable to human agents.
- Post-launch, we analyse queries that AI couldn’t answer and expand its knowledge base accordingly. This leads to a 50% increase in the number of queries the AI can resolve over time.
- We've also extended AI to voice. Following a successful pilot, voice AI is now live for one of our largest clients – handling inbound queries in real time, in multiple languages, around the clock.
- We also invest in the AI literacy of our staff to foster awareness and proper oversight.
Bottom line: Our experience shows human expertise must be systematically documented to unlock AI’s full potential – which takes significant time and effort.
.png)
Lesson 2: AI’s success depends on robust upstream and downstream processes
When returning from an operation to repair a tendon, we need time to rebuild the supporting muscle through physiotherapy. Your achilles (heel) doesn't work in isolation.
Much the same, you can't simply insert AI into a process without levelling-up the business muscle around it; everything will fall over somewhere else in the chain.
For AI claims processing, for instance, success depends on:
- High-quality document capture: ensuring that claim documents are clear, well-structured and digitised correctly.
- User-friendly experiences: making the claims submission process intuitive and ensuring customers provide accurate inputs.
- Sensible prompt engineering: refining AI queries to extract meaningful and relevant data.
- Seamless AI-human integration: transforming AI-generated insights into actionable claims decisions.
What we’ve learned at Qover:
- Well-structured claim forms can enable automation of up to 50% of claims without AI.
- Around two-thirds of our automation success comes from rule-based verifications. We achieve this by improving the questions in our claims forms and converting key policy conditions into system-run checks.
- To tackle the remaining one-third of claims, AI is essential. But for AI to work effectively, prior investments in data accuracy, flexible validation frameworks and seamless user interfaces are crucial.
- Using AI for document verification, we’ve been able to automate up to 25% of complex claims, significantly reducing manual workload and improving efficiency.
- As of 2026, around a quarter of all customer contact is now automated, with 80% of those interactions requiring no human in the loop at any stage. Our target for claims automation this year is 60% – of which half will be fully straight-through, with no human review needed.
Bottom line: It’s important that AI claims processing adoption be accompanied by robust process and insurtech user experience improvements across the board.
.png)
Lesson 3: Optimisation is an ongoing process
As with people, AI needs time and continuous learning to achieve mastery. Organisations expecting overnight success will be disappointed. Getting to launch is step one; realising the full benefits of AI requires patience, experimentation and determination.
Before launching AI initiatives, it’s essential to clearly define:
- The core objectives the AI solution is meant to achieve.
- Metrics to measure success before and after implementation.
- A structured optimisation plan post-launch to ensure continuous improvement.
Our AI optimisation approach at Qover:
- Every AI automation project enters a hypercare phase post-launch, where project teams dedicate over half their time to monitoring performance, gathering frontline feedback and fine-tuning the AI.
- This allows us to swiftly address false positives, enhance AI understanding and improve its integration into workflows.
Bottom line: Companies should plan for a period of AI fine-tuning to maximise effectiveness and long-term gains.
One year on: is it compounding?
The honest question to ask after a year of this work is: are we building something that gets smarter over time or are we just running faster on the same treadmill?
I've been thinking about this a lot since a panel discussion at Insurtech Insights in April 2026, where strategist Simon Torrance introduced the idea of intelligence capital. The idea is that AI, when done properly, creates a new kind of organisational asset – one that appreciates rather than depreciates. Unlike human expertise, which walks out of the door when someone leaves, the reasoning and decision-making logic you encode into your systems stays, compounds and improves.
I think that describes what we've been inadvertently building. Every hypercare cycle that refines an agent's outputs, every edge case we document, every claims handler who moves from processing decisions to overseeing the AI that processes them – that's all intelligence capital accumulating.
The thing that's surprised me most is the people story. When we started, the natural worry was what happens to our frontline teams as automation takes over routine tasks. What we've found in practice is something more interesting: the people who were closest to the work – who knew exactly where the process broke down, where customers got confused, where a claim needed a second look – are becoming our best AI overseers. They're not being replaced, butredeployed into roles that didn't exist 18 months ago, and they're better at those roles than anyone we could have hired from outside.
I don't want to overstate it. We're still learning, still course-correcting and 60% claims automation is a target, not a given. But the trajectory feels different now to how it felt when I wrote the original version of this post. Less like we're chasing the technology, more like we're starting to shape it.
Final thoughts
AI has already demonstrated its potential in improving customer experience and operational efficiency. However, unlocking its full power requires careful planning, investment in the team’s AI literacy, strengthening supporting processes and a commitment to continuous improvement.
For insurance companies and other industries navigating AI adoption, our advice is simple:
- Be bold in experimenting with AI, but realistic about its limitations.
- Ensure high-quality inputs underpin AI initiatives.
- Think holistically about how AI integrates into broader business workflows.
- Plan for ongoing optimisation rather than a one-and-done implementation.
- Think about what you're building, not just what you're saving. Efficiency gains are the floor, but the ceiling is an organisation that gets meaningfully smarter every quarter.
This is just the beginning of our AI journey. This series continues with Thibault Gillis, our former Head of AI & Transformation, on how we scaled AI across the whole business.
For a recap of what the wider industry is grappling with, and how we're deploying agentic AI in insurance at Qover, you can also check out my panel at InsTech London. Happy reading.
.jpg)
