There is a paradox every insurer knows well. Of all customers, the insurance customer is the one who speaks to their insurer least. Three or four occasions a year, between expiries, renewals, and claims. Yet each of those occasions decides, in full, the perceived quality of the relationship. A badly handled claim wipes out ten years of premiums paid on time.
That is why AI Agents in insurance are judged not by the quality of their answers, which is by now a baseline requirement, but by the depth of what they execute in the few moments that matter. Taking a complete claim notification, not explaining how to file one. Preparing a renewal, not reminding the customer that one is due. And the stakes have risen on both fronts, with more policies on the books and new rules governing the use of AI.
Larger portfolios, harsher peaks
The obligation to insure against natural catastrophes, introduced for businesses by Italy's Budget Law for 2024, has brought an entire population of newly insured companies into insurers' property portfolios, with deadlines phased by company size. Businesses that had never bought cover against floods and earthquakes, and that now contact their insurer to understand it, manage it, and, when the need arises, use it. Non-compliance, moreover, carries a defined cost. The law requires it to be taken into account when public grants, subsidies, and incentives are awarded, and a ministerial decree has already aligned access to incentives with the obligation. The industry's volume of interactions has grown structurally, and it will not fall back.
Larger portfolios bear peaks that no amount of planning can anticipate. A single weather event concentrates into a few hours volumes of claim notifications that would ordinarily be spread across months, beyond the capacity of any contact centre sized on headcount. AI Agents are the first structural answer to this asymmetry, because their capacity expands during the emergency and contracts once the peak subsides. And because a claim notification taken with precision during the event is what underpins settlement in the weeks that follow.
Claim notification is the moment of truth
A poorly taken claim notification is paid for three times over. In cycle time, an incomplete file bounces between the claims handler and the policyholder. In repeat contacts, because a customer left without news calls back. And in complaints, which in general insurance arise more often from handling times than from the amounts settled.
Entrusted to AI Agents integrated with the insurer's systems, the first notification of loss, the FNOL, changes in nature. The policyholder is recognised, and the policy record is retrieved instantly. Information gathering is adaptive because hail damage to a car and a flooded warehouse call for different questions, and the AI Agent asks only those relevant to the line of business and the coverage in force. Documents and photographs of the damage flow through the same channel as the conversation, starting with WhatsApp, without diverting the policyholder to portals or email inboxes. Cover is verified in real time against the policy systems, and a claim that falls outside cover is clarified at once, not after weeks of waiting. The file reaches the appropriate claims handler already classified, complete, and ranked by priority. In the days that follow, it is the AI Agents that keep the policyholder informed of progress, the single request that by itself generates a substantial share of calls to claims lines.
Assessment and settlement remain with people. But a claims handler who receives fully prepared files makes better decisions faster, and cycle time shrinks precisely where the policyholder feels it.
Renewals, where outbound is written into law
In Italian motor insurance, tacit renewal has been prohibited for more than a decade. Every motor liability policy, at every expiry, is to all intents and purposes a new sale, by operation of law. Few sectors have such a reason to run outbound, and almost none manage to do so across the entire portfolio, because no call centre covers every expiry, every time.
AI Agents change the economics of renewal by turning it from a campaign into a process. Every customer is contacted at the right moment, on the channel they prefer. The quote takes shape within the conversation, aided by industry databases that make it possible to price from very few declared details. Objections are answered, data is brought up to date, and the cases that warrant negotiation pass to a consultant with the full context already in place. Retention stops depending on the capacity to work through call lists and becomes systematic coverage of the portfolio.
The same rigour applies to complaints, on the opposite side of the relationship. The AI Agents gather the elements required by regulation, route the case to the appropriate function, and give the policyholder a reference and a response deadline, with the end-to-end traceability that supervisors expect. Assessment of the merits remains with people.
Proof from the field. Voice that executes
Voice is the most demanding channel for AI Agents, and it is also the one where the difference between answering and executing shows most clearly.
Net Insurance has entrusted AI Agents with the medical interviews required for policy applications, a regulated process in which completeness and precision leave no margin for error. The operator identifies the customer and hands over to the Agent, which conducts the interview autonomously and records the answers in the company's systems, already in structured form. Around 1,200 interviews a month, 83% completed without operator intervention.
The regulatory map rewards good design
On AI governance, the industry labours under a misconception worth clearing up. The AI Act's high-risk classification covers, for insurance, risk assessment and pricing in life and health insurance for natural persons, with obligations that, following the postponement introduced by the EU's Digital Omnibus package, will apply from 2 December 2027. Conversational Agents handling claims, renewals, and complaints generally fall outside that category. Those who delay adoption by invoking high risk are delaying in the name of a constraint that, in most cases, does not concern them.
Outside that category, however, lies not a regulatory vacuum but a framework already written. Since 2 August 2026, the AI Act's transparency obligations have been in force, beginning with the most visible. Customers must know they are conversing with an artificial intelligence system. And the opinion with which EIOPA framed the governance of AI systems in August 2025 places everything else within the rules insurers already apply - Solvency II, the IDD, DORA and the GDPR, under a principle of proportionality. The operational message is clear. AI Agents are governed with the instruments insurers have always used, applied with rigour. Full logging of every action, granular permissions for every Agent, health data handled under the GDPR's minimisation principle and legal bases, and suppliers within DORA's perimeter.
This arrangement has a name, human in the lead. The person decides on settlements, on complaints, on exceptions. The Agents execute, before and after the decision, and every action they take can be reconstructed. It is the arrangement that satisfies the current framework by design, and that will not need rethinking when the postponed deadlines arrive.
Where to start
The request for an update on the status of a claim is almost always the right starting point. High volumes, a defined procedure, immediate value for the policyholder, and for the contact centre. From there, the perimeter extends based on the strength of the data. First notification of loss in the highest-frequency lines, then expiries and renewals, with autonomy levels defined together with compliance before every release and metrics fixed from the outset. Claim cycle time, first-contact resolution, repeat contacts, retention.
The insurance contact centre has long been the place where volume was absorbed. With AI Agents that execute, it becomes the place where processes get shorter, and the few moments that matter are played well. The insurers that have made this shift have not made their answers faster. They have made their processes shorter.
FAQ
Which insurance processes should be entrusted to AI Agents first?
Those that combine high volumes with a defined procedure. Claim status updates, which alone generate a substantial share of calls, first notification of loss in the highest-frequency lines, renewal reminders, expiry management, and the structured intake of complaints. These are processes that repay the integration quickly and build the track record of reliability needed to extend autonomy.
Is a conversational AI Agent a high-risk system under the AI Act?
In most cases, no. For insurance, the high-risk classification concerns risk assessment and pricing in life and health insurance for natural persons, with obligations applying from 2 December 2027 following the postponement introduced by the Digital Omnibus. An Agent handling claims, renewals, and complaints generally falls outside that category, but it is subject to the AI Act's transparency obligations, in force since 2 August 2026, and to the framework EIOPA has set out within Solvency II, the IDD, DORA, and the GDPR.
Can a claim be settled automatically by an AI Agent?
Settlement is a decision, and it remains with people. The AI Agents handle everything that precedes and follows it, the complete intake of the claim notification, the verification of cover, the opening and routing of the file, updates to the policyholder, and the communications that come after. The claims handler receives a fully prepared file and decides with more context in less time.
