How CDL is using Dynatrace to prepare for agentic AI in insurance

The insurance industry has weathered major technological shifts before – from telephone quotes and paper applications to price comparison websites and digital self-service. CDL, a UK-based insurtech provider serving brands including the AA, Co-op Insurance, and Swinton, believes the next transformation will be even more fundamental – and it's using AI observability to make sure it's ready.

Robert Trueman, Chief Technology Officer at CDL, doesn't mince words:

I think a radical shift isn't even enough to explain what's coming down the road in terms of how AI will impact consumer behaviors. We like to view Alexa Plus, Gemini, things like that. So what we always try and do is get ahead of the game in these instances.

It's why CDL is now extending its work with Dynatrace to build the observability infrastructure needed to develop, test, and deploy AI agents in a heavily regulated environment. The company has consolidated from 10 separate monitoring solutions to a single platform, while establishing the frameworks necessary for AI deployment in financial services.

From reactive to proactive

CDL's relationship with Dynatrace predates its current AI ambitions. The company was an early adopter of Dynatrace's Software-as-a-Service (SaaS) platform, and the two have been working together on new capabilities for several years. Trueman explains:

We've always tried to say to them, look, we'll be the first to try this or we'll work with this, because we're typically trying to be in the same space. So I think the AI space was no different, really.

The reason for consolidating was simple: maintaining multiple monitoring systems was slowing teams down. Before the shift to Dynatrace, CDL's teams operated with disparate tooling, each adequate for its specific purpose but collectively making problem diagnosis harder than it needed to be. Trueman notes: 

To actually solve a problem or be proactive about solving the problem, you might have had to have pulled data from five of those 10 different systems together. And I think having a single source of truth in Dynatrace, where it typically just tells you what the problem is before you've even got to that without even having to delve into it – that's mega.

The consolidation has had measurable effects. CDL reports a 10% improvement in customer satisfaction scores, which Trueman attributes to a broader strategic transformation that included migration to Amazon Web Services (AWS) and moving away from legacy Oracle systems – with observability as a key part of making that work.

Building for model volatility

As CDL develops its agentic AI capabilities, the observability challenge has become more complex. AI is changing so quickly – with models being released "probably daily, if not hourly" – and that creates particular challenges for organizations that need to validate performance, cost, and regulatory requirements before deployment.

CDL has already encountered the kind of issues that make AI observability essential. Through Dynatrace's monitoring, the company identified that an outdated version of Anthropic's model was causing token exhaustion – a problem resolved by updating to a newer release. But for Trueman, these specific examples matter less than the underlying capability they demonstrate. He explains: 

Those examples and other things will happen all the time as we move through models and model adoption rates increase. So building a stack where we can effectively test models and see those differences and figure out how a new model variant or shift into a different model can solve some of the challenges that we see at that point is key to the platform that we're building out.

This extends to evaluating new options as they emerge. Trueman references recent developments in Chinese AI models:

We've seen some of the Chinese models come out that are a lot more efficient. So how do we get them into our system? How do we prove they're more efficient through the data, not just marketing blurb? How do we prove that, get it in, prove that it has all of the governance that we need, all the performance that we need?

Governance as foundation, not afterthought

Operating in financial services means CDL cannot bolt on AI governance later. The company's approach has been to establish policies, procedures, and processes around AI before pushing capabilities into production – addressing ethics, hallucination risks, and bias alongside technical performance. Trueman acknowledges:

I think AI is bringing a rethink. Part of the POC research development we do is around the tech stack side of things. We also went very early in terms of looking at our corporate policies, procedures, processes around AI across the board.

The company uses OpenLLMetry support through Dynatrace to give its teams visibility into Large Language Model (LLM) behavior, including cost optimization for token consumption. This transparency serves both regulatory requirements and internal trust-building as AI capabilities expand.

CDL's roadmap reflects this cautious approach. The company began with what Trueman calls "safer interactions" – use cases where the scope of AI action is more bounded – before progressing to policy changes, customer data modifications, and transactional interactions.

Trueman elaborates: 

We started with the safe change and things that you can do, but that's really the proof point to start to get into the higher value items where you will start to make actual policy changes, changes to customer data, changes to information that the customer holds with the brand.

The partnership model

As a privately-owned company, CDL doesn't face the same quarterly earnings pressure as publicly-traded competitors, giving it more freedom to invest in longer-term research. That means it can explore technologies like AI and quantum computing, bringing partners like Dynatrace and AWS into the process as ideas move toward real products.

Trueman believes that collaboration is key:

We quite often don't just ask our partners about the specific thing we're asking. We'll ask them about their experience of all those things, because when you're innovating on things, hopefully, you know, if you're picking the right partners, then you think you've got to consider all these different aspects.

This extends to CDL's own customer relationships, connecting what the insurers it serves are asking for with what its technology partners can support.

My take

CDL's approach is worth noting for what it prioritizes. Plenty of organizations are rushing to deploy AI capabilities – sometimes without enough attention to observability or long-term supportability. CDL has chosen to build infrastructure before acceleration. You cannot optimize what you cannot measure, and you cannot govern what you cannot see.

Trueman's advice to peers – invest in the frameworks before diving into "the cool stuff" – reflects real experience. Models change constantly, regulatory scrutiny is intensifying, and the ability to validate, compare, and audit AI behavior is likely to become a real advantage rather than just overhead.