A low-speed collision, a severe impact and a claim containing inconsistent information cannot be handled in the same way. The severity, the information required, the people involved and, above all, the decisions to be made are all different.
Yet when claims automation is discussed, it is often imagined as a uniform process: the system receives information, applies a rule and initiates a predefined sequence of actions. This model can work very well for straightforward cases, but its limitations become apparent when a document is missing, an anomaly emerges or the event requires a more detailed assessment.
It is precisely in this less predictable territory that Agentic AI could mark an important step forward.
Unlike systems that simply analyse data or suggest a response, an AI agent can interpret the context, identify missing information and determine the next appropriate step among the actions it is authorised to perform. It does not necessarily follow the same path in every situation but can adapt it as new information emerges during the handling of the claim.
The most interesting question, therefore, is not whether artificial intelligence will one day manage every claim autonomously. That is an extreme prospect and probably not even a desirable one. The real question is how far AI should be allowed to go and under which circumstances it should stop.
Consider a minor accident. The circumstances are clear, all the necessary information is available, no anomalies emerge and the damage is consistent with what has been reported. In this situation, an AI agent could collect the data, verify coverage, open the claim, request any necessary images and update the customer on the next steps.
Human intervention could be minimal.
Faced with a more serious collision, the same agent should behave differently. Assisting the people involved could become the immediate priority, followed by collecting the necessary information and promptly assigning the case to a specialist. Some claims may also appear straightforward at first but become more complex as they develop. An unclear reconstruction, a discrepancy between the reported circumstances and the detected event, or the emergence of new information can completely change the appropriate course of action.
In these cases, Agentic AI could:
- Recognise that the available information is insufficient;
- Interrupt the automated process and request further verification;
- Assign the claim to the person with the most appropriate expertise;
- Provide the operator with all the context already collected.
The advantage would not therefore lie in automating as many activities as possible. It would lie in preventing fundamentally different claims from being treated as though they were the same.
Well-designed autonomy should not be measured solely by what a system can do on its own, but also by its ability to recognise its own limitations.
To make this distinction, artificial intelligence must be able to rely on trustworthy information and the most complete possible view of the event.
In claims management, integrating telematics data, images, vehicle information, policy conditions and the insurer’s rules can help to:
- Assess the potential severity of the event more quickly;
- Verify consistency between the impact, the claim report and the damage;
- Identify cases requiring further investigation;
- Reduce duplicate requests, manual handovers and waiting times.
This is where Agentic AI can find one of its most practical applications in the insurance industry. Not as a separate layer of technology, but as a tool capable of transforming information from different sources into a coherent operational process.
Opening a claim, checking whether documents are available or sending an update are activities that can be automated. Decisions with significant financial, legal or personal consequences, however, require expertise, accountability and human oversight.
Freeing people from repetitive checks and the management of straightforward cases allows them to focus their expertise and experience where they genuinely make a difference. For customers, this can mean receiving faster responses without being left alone in an impersonal process. For insurers, it can mean using resources more effectively and assigning complex cases to the most appropriate professionals.
This distinction cannot be improvised while a claim is being handled. It must be established in advance by clearly defining which activities the agent can perform autonomously, which require confirmation and which cannot be delegated.
Traceability also becomes essential. If an agent changes information, initiates an action or directs a claim towards a particular path, the insurer must be able to reconstruct what happened and understand the information on which the action was based. Autonomy does not remove the insurer’s responsibility or reduce the importance of oversight. On the contrary, it requires even stronger rules.
The future of claims management will not necessarily be free from human involvement. It will be more selective: some cases will move quickly through automated processes, while others will be assigned to people as soon as their complexity demands it.
The evolution will not consist in handing everything over to artificial intelligence, but in understanding what AI can do on its own, when it requires confirmation and when it needs to stop.