How IBM, AXA and Others are Capturing Real Value from AI

After years of headlines about AI, UK businesses are moving beyond experimentation to capture measurable returns on investment.

According to recent IBM research, 66% of UK businesses are already achieving productivity gains from AI, while 74% of consumers are comfortable with AI playing a role in decision-making.

Yet many organisations still struggle to scale beyond pilot projects. At a recent IBM roundtable in London, industry leaders from insurance, education, technology and consulting gathered to discuss how they’re overcoming barriers, capturing ROI and preparing the UK workforce for an AI-powered future.

They were:

• Leon Butler, CEO UKI at IBM
  

• Natasha Davydova, Chief Information Officer at AXA UK

• Vishaal Gupta, President, Enterprise Learning & Skills at Pearson

• Sue Daley, Director, Tech & Innovation at techUK

Where is AI delivering measurable ROI in your organisation?

Natasha Davydova, AXA: In insurance, we’re seeing measurable ROI across claims processing, fraud detection, underwriting, risk assessment and personalised insurance products. Claims processing automation is cutting administrative costs by 30-40%, while fraud detection accuracy has improved dramatically. In underwriting and risk assessment, we’re seeing accuracy improvements of up to 50%. When we monitor policies with AI and automation, we’ve reduced processing times by 70%, which enhances both compliance and risk management.

At AXA specifically, we’ve introduced AI to help our contact centre agents. Previously, when a customer asked whether their policy covered a specific aspect, it would take on average five minutes for the agent to search through policy documents. Now, the AI picks up the question, searches the policy, and brings up the reference within five seconds. The agent can provide an answer immediately, which dramatically improves customer satisfaction.

We’re also using AI for automation in our contact centres, so agents can focus on value – paying attention to customers rather than administrative tasks. In underwriting, where you process vast amounts of data, we can move from months and weeks of analysing historic data to hours and minutes. That's transformational.

Vishaal Gupta, Pearson: We operate across the entire learning spectrum from schools through to the workplace, which gives us a massive database to work with. One area where we've seen real scale is our AI study tools and study tutors. Students are using these extensively. Interestingly, our highest usage happens between 11pm and 2am, when human tutors aren't available. Students are getting ready for exams and using these tools for last-minute revision.

From a market standpoint, AI is providing us with context about our customers that we simply didn't have before. That allows us to launch highly personalised learning experiences, and those productivity gains can be transformative for individuals. We’re also doing important work supporting teachers, because they’re another community facing the impact of AI and many are struggling with how to integrate it effectively.

But here’s what’s critical: assessments. If you look at LinkedIn profiles today, you’d think everyone can walk on water. So how do you actually validate those skills? Assessments become absolutely vital. Within the assessment space, we’re using AI to target questions more effectively and to support marking processes.

Leon Butler, IBM: At IBM, we’ve taken a ‘Client Zero’ approach, using our own technologies internally before deploying them to customers. Over the last couple of years, we’ve generated approximately US$4.5bn in productivity gains. That’s not a target we set lightly, and we’ve exceeded it.

Sixty-six per cent of businesses we surveyed across Europe, the Middle East and Africa are achieving productivity gains. Sixty-three per cent of leaders report operational efficiency improvements and 62% believe they need to do even better to reach the next level of productivity gains.

For us internally, it’s about systematically applying AI across domains: HR, supply chain, procurement and sales. We have something called the Skills Hub, which uses natural language to query everything from talent management to employee movements. Tasks like promoting an employee, which used to be complex multi-system processes, can now be orchestrated by AI that navigates across multiple applications automatically.

We’ve also implemented the IBM-er watsonx Challenge, where every employee can submit ideas for how AI could improve their daily work. About 178,000 employees registered for the 2024 challenge and 70% of teams who submitted projects created improvements they’re now applying to their work.

What are the biggest barriers preventing organisations from scaling AI?

Sue Daley OBE, techUK: Skills comes up as a challenge all the time, but there are two levels to it. First, if an organisation is investing in AI solutions, do the employees actually have the skills to use it? Even back to basics: how do I write a prompt?

The other side of the skills gap is: does the UK have the people to deliver that AI to companies? AI experts, quantum engineers: what kind of skills do we need at that level?

Then there’s data. I might have loads of data for my AI to consume, but is it good data? Rubbish data in, rubbish data out. There are still a lot of organisations struggling with that.

And the last one is trust. Can I trust this tool? We need good AI assurance – risk management approaches, auditing tools and technologies that can assure people that the AI products and services they’re putting into their organisations are actually doing what they're supposed to do.

Natasha: I’d add regulatory compliance to that. If you take the EU AI Act, for example, it requires you to explain how the decision has been made. You can’t just push in massive datasets and ask AI to choose without being able to explain it.

At AXA, we’ve launched AI and data academies which have several tracks of complexity depending on whether you're a data scientist or somebody who’s less experienced with AI. We also work with our technology partners to create proof-of-concept projects, and that actually helps quite a lot.

Vishaal: From a customer perspective, it’s really hard. We work with Microsoft and Google and all these companies, and there are so many options. It’s difficult to work out what’s actually valuable versus what’s hype. You might not even need AI. You might need something else.

What infrastructure challenges are organisations facing as they scale AI?

Leon: Complexity is a major one. If you look at the average enterprise, you’ve got over a thousand applications. When you’re building AI applications across on-premise, private cloud and multiple public clouds, that becomes a real challenge to deal with.

Energy is another one. AI is very energy-intensive, and that’s a real conversation we need to have.

From an agentic perspective, there’s a risk at the moment: you’ve got agents all over the place. Agents are uncoordinated, not talking to each other. The ability to evaluate what agents are doing, especially when they can access multiple applications, is going to be really important.

Sue: This connects back to skills as well. Someone said to me recently that this will be the last generation that only manages humans. The next generation will be managing both humans and AI agents. So what does that mean for the management skills we’re teaching people in workplaces right now? What does that look like for people entering the workforce? I think that’s a fascinating area, but we haven’t yet figured out the answers.

What role should government and industry play in developing AI skills?

Sue: The government made a great announcement in June about working with industry to train 7.5 million people in AI in 2025. That's a really powerful goal, but the question is how we make that happen. Government’s role is being an enabler.

There’s obviously curriculum changes: AI is going to be introduced into the curriculum. There’s something around apprenticeships, and how those funds could be used towards upskilling and retraining.

From an industry point of view, there need to be internal pathways within organisations to develop AI skills. I think there’s also something about industry’s role as a recruiter: how do we position these new roles so that we can get people from diverse regions and backgrounds across the UK into our sector?

Vishaal: Our research shows that in the UK alone, because of lack of skills, the UK economy is losing about £96 billion a year. That happens because as you come out of college, you're not ready for work. Between jobs, you take very long to get employed because you don't have the right skills. Or even when you are at work, you don't have the correct skills.

In schools, we’re working closely on curriculum. But we’re also very focused on vocational skills: nursing, hospitality, construction. That’s where jobs are getting created.

For companies, we start by helping them evaluate their current skills. If you’ve been following the last six months, learning how to learn is critical. If you don’t have learning-to-learn capabilities, you're going to fail, because things are evolving so rapidly.

What industry-specific challenges have you encountered in scaling AI?

Natasha: Insurance companies have been around for hundreds of years. Legacy systems are a real issue, and there’s vulnerability when you’re integrating new technologies. You have to be strategic about where you apply AI.

What’s crucial is selecting use cases that deliver genuine value. When customers get married, they might take out life insurance, but then that policy might just sit there for 30 years. So you need to think carefully about where AI adds real value versus where you're just automating for the sake of it.

Personalised products can increase customer retention by 5-10%, and for large insurance companies, that's substantial value. When you align AI with strategy and measure against clear business outcomes, you can demonstrate real ROI.

Vishaal: In education, when we introduce AI tutoring tools that students are accessing at 11pm, we’re not just providing a service but changing how learning happens.

The assessment question is critical. In an age where AI can help write essays or complete assignments, how do we validate that students have actually learned? It's about fundamentally rethinking how we measure learning and competency in an AI-enabled world.

Looking ahead, what excites you most about AI’s potential?

Leon: What excites me is moving beyond individual use cases to seeing AI embedded across entire enterprises. The Client Zero approach means we can validate solutions at scale before we offer them to customers, which builds trust.

The productivity gains we’re seeing, US$4.5bn and counting, that creates a flywheel effect. Those savings fund further innovation, which drives growth. That’s sustainable progress.

Natasha: When we can answer a customer’s policy question in five seconds instead of five minutes, that’s fundamentally better service. When we can analyse underwriting data in hours instead of months, we can be more responsive, more competitive and provide better value to customers.

Vishaal: I’m excited about democratising access to high-quality education. When a student can access an AI tutor at 2am while they’re preparing for an exam, we're breaking down barriers of time and geography. When we can personalise learning pathways based on individual needs and learning styles, we’re moving toward truly individualised education at scale.

Sue: What excites me is seeing UK companies not just adopting AI but leading globally in responsible AI implementation. The UK has an opportunity to set standards for ethical, transparent, trustworthy AI. If we can combine technical excellence with strong governance and inclusive skills development, we can build AI systems that genuinely serve society while driving economic growth.