How Allianz Partners is transforming the claims experience with Agentic Process Automation

Stefan: Can you tell us about your role, goals, and this collaboration?

Pieter:  Sure! We’re a leading brand in the assistance business, operating across North America, South America, Europe, and Asia. One of our biggest challenges is that over the years, we’ve acquired multiple companies, each with its own legacy systems. We’ve never had the chance to build a unified operational backbone.

With the advancements in language models and automation, we finally have the tools to create that backbone. We’re starting with decision-making, using DeepOpinion to bring best-in-class skills, tools, and workflows into our operations.

It’s not just about automating repetitive tasks, it’s about improving efficiency, reducing errors, and making more accurate decisions - as we make thousands of decisions every hour across the globe. We’re helping to bring consistency and quality to decision making and lifting the work of operations to a 90% automation rate.

‍

Stefan: What challenges did you face in claims and why did you start this project?

Pieter: Insurance is in many ways the obvious choice for applying large language models. The main trigger was cost because claims processing is expensive. It requires a lot of people handling different types of claims, each with lots of variability and inconsistency across our lines of business. We spend a significant amount of money managing these manual processes on a daily basis.

That’s why claims became the obvious starting point. It stood out due to the complexity and volume of work involved. It’s good to have state of the art tools that are expert in LLMs and most importantly hands on knowledge in the insurance domain - specifically claims processing.  

The key challenge isn’t just improving the accuracy of information extraction or classification but addressing more difficult claims handling problems. That’s where domain knowledge makes all the difference.

‍

Stefan: Do you see increased customer experience and competitiveness, what value is being delivered?

Pieter: The biggest value drivers are consistent and accurate decision-making, which helps prevent poor service quality for customers, avoids regulatory issues and reduces stress on our workforce.

‍

Stefan: Why did you move beyond technologies like RPA and document processing point solutions - to an end-to-end Agentic Process Automation layer?

Pieter: Technology often comes with a lot of hype and high expectations, and we didn’t want to put LLMs in the same bucket. Automation itself isn’t new; it’s been around for years. But what really matters is understanding what’s “good enough” versus what you really want to achieve.

You want to make sure that things flow with a platform that can handle requests and dispatch providers - you want all that automated and well organised. Or do you want marginal improvements where you can apply a bit of RPA or some bots? We wanted the first option as we have big operations globally, so it made sense scale wise.

‍

Stefan: What have you been able to achieve so far and what’s next?

Pieter: This year we went live with supporting travel claims automation and we're about to go live with some invoice automation. 

The outlook for us is scaling this up and being able to really replicate what we've built across all our geographical regions and also doing a lot more on the automation side.

‍

Stefan: How does DeepOpinion integrate into existing infrastructure, what are important factors?  

Pieter: Obviously, APIs are important as always. If you have a relatively old set of core banking systems that’s not necessarily an issue. You’re not replacing them, you’re interfacing with them, the system of record remains in place and DeepOpinion works alongside it.

The second part is the human-in-the-loop. We’ve worked with DeepOpinion to take this a notch further, so operations teams can monitor, control, and fine-tune the system. We are also looking at the flexibility of LLMs to ask it any question and make a good job of transforming and making outputs.

‍

Stefan: How does DeepOpnion ensure secure, scalable and compliant Agentic Process Automation in an enterprise environment?

Pieter: From a security and privacy perspective, we were comfortable with you guys from the start because you truly understood our constraints. The first checkpoint was ensuring we had a secure environment for data exchange. 

Secondly, from a privacy perspective, is to ensure we don't make any mistakes in data handling and to operate within established processes that customers are aware of, whether under legitimate interest - or with customer consent. 

DeepOpinion has helped us be aware of compliance and privacy laws, and have a really easy setup for us to work with. We work quite well with interfacing back and forth between DeepOpinion backends and our backends.  

‍

Stefan: Are there any best practices you can share that you've learned so far in the collaboration? 

Pieter: Product and service ownership. Claims handling is a service, so it becomes a permanent thing. You need to staff correctly; people that understand both handling claims and insurance, as well as the technology.

Product and service owners are the one skill that’s hardest to source. You need strong programme management, especially if you’re at our scale. You absolutely need AI ML engineers that understand how to maintain, run, and monitor plug into these decision flows effectively. Most importantly, you need willingness in each of your teams to want this.

Our senior leadership are quite mature in understanding the value of this. We are progressing so fast because we've got a full leadership team completely understanding, committed to driving this forward in a lot of detail. It's not just me and my team, it's not just the operations team; it's all the company coming together.