Beyond fixed workflows: the next opportunity for AI in claims

Where should a claims handler still be involved, and where are they involved simply because the process was designed that way?

For P&C insurers, that question is becoming harder to ignore.

Featured in this article:

• Claims are hard to automate because of variation, not complexity. Fixed workflows have to map every path in advance. 

• AI-supported workflows can handle some of that variation, checking coverage, requesting missing information and moving straightforward claims on within governed rules. 

• People stay where judgement matters. The opportunity is more straight-through processing without automating every decision. 

Why variation has made claims harder to automate

Claims are not difficult to automate because every step is unique. The challenge lies in what happens when a claim does not follow the expected path. Information may be missing, coverage needs to be checked, or a decision may depend on circumstances that were difficult to anticipate when the workflow was designed.

Fixed workflows need those paths to be mapped in advance. That works well for predictable processes but becomes harder as variation increases. The whitepaper describes a different model in which a workflow can reason through the specifics of a case against governed data, handle some of that variation and escalate when the situation warrants human involvement.

This changes the automation question. Instead of asking how to force more claims through the same predefined path, insurers can start looking at which parts of the process can respond to the claim itself.

From individual AI tasks to adaptive workflows

AI does not need to take over an entire claims process at once. In our whitepaper, Innovation Built on Trust, we outline the three levels at which AI can play a role, starting with embedded capabilities such as summarisation or extraction before moving into task-level agents and orchestrated workflows.

At task level, AI can perform a specific activity within a familiar process. Checking coverage is one example. Gathering information that a handler would otherwise collect manually is another. The process itself remains recognisable, but the aim is to reduce handling time and the cost per claim.

The next step is an orchestrated, AI-native workflow in which agents work with other agents and people across an end-to-end process. In claims, that means the workflow no longer has to follow a path fully modelled in advance. It can adapt to the case in front of it, while still operating within a governed process.

Let straightforward claims move, and keep judgement where it matters

The most immediate opportunity within P&C sits in high-volume, repeatable work. Today a handler touches almost every claim, including those that stalled because a document was missing or coverage needed confirming.

In a governed AI workflow, those claims keep moving. The workflow checks coverage, asks the customer for what’s missing, and decides the next step. When a case warrants human judgement or accountability, it is transferred to a person with the context already gathered.

Governed has a specific meaning here. The workflow runs on data and business rules the insurer already controls. Every automated step is recorded, so a decision can be explained to a customer, an auditor or a regulator. And the insurer sets the boundaries: which decisions the workflow can make, and which always go to a person.

This doesn’t mean every claim should become fully automated.

Nelson Petracek, CTO at Keylane, poses a deliberately challenging question to test where people are really needed:

The point is to look at the current process with fewer assumptions about where a person needs to be involved, then identify the moments where human judgement really matters. 

A different route to straight-through processing 

Straight-through processing has traditionally depended on a process predictable enough to run without manual intervention. Variation is what pushed claims off that path. When the workflow can handle some of that variation itself, more claims stay on it. 

The whitepaper links this directly to higher STP rates and lower leakage, with claims teams focused on the cases where their expertise is needed.

Rethink the workflow, not just the task 

The bigger shift is from adding AI tools around an existing claims process to reconsidering how the process itself could work. Chatbots, summarisation and individual agents can all be useful, but the larger opportunity comes from redesigning business processes rather than leaving them largely unchanged. 

Keylane approaches this from the core outward. Axon already runs claims administration for P&C insurers across Europe, and was named a Technology Standout in Celent’s 2026 P&C Claims Administration Systems: EMEA Edition. Keylane Automate builds on that foundation: AI that works with the data, rules and processes insurers already trust, governed by design and owned by the insurer. 

That is where AI can add something different to claims automation: not by removing variation from the process, but by helping the process deal with it. 

Read the broader perspective 

This article draws on Innovation Built on Trust, Keylane’s whitepaper on how AI can create value in insurance and pensions. The whitepaper explores how governed data, AI-supported workflows and human oversight can support the move from individual AI use cases towards redesigned business processes.