Chatbots and AI Assistants for Insurance Agencies: What They Can Really Do (and Where Humans Are Still Needed)
Over 55% of Italian insurers have launched AI projects, but only one in five has moved beyond the pilot stage. For an agency, the real question is not whether to use a chatbot, but where to draw the line between automation and advice.
The insurance chatbot: between promise and proof
Over the past two years, the chatbot has become the most repeated promise made by insurance technology vendors: round-the-clock support, instant answers, customer-service costs reduced to zero. On the marketing materials, everything works flawlessly. In the reality of an agency with a few thousand clients and a team of three or four people, things are far more nuanced.
The question is not whether conversational artificial intelligence "works": it does, and in some tasks it works very well. The question is understanding where it works and where, instead, introducing an automated assistant risks doing more harm than good. A chatbot that mishandles a distressed client after a claim does not save time: in a matter of minutes it erodes the trust built over years. This article sets out to separate what an AI assistant can concretely do in an insurance agency today from what remains — and will long remain — the exclusive competence of the intermediary.
Where AI in Italian insurance really stands
The Italian insurance market is in good health. According to ANIA, in the first nine months of 2025 total premium income exceeded 88 billion euros, growing by close to 10% year on year, driven by both the life and the non-life segments. It is against this backdrop that artificial intelligence has stopped being a laboratory experiment.
Cetif's Research Insight reports indicate that the most widespread applications among insurers concern fraud prevention — adopted by roughly 65% of operators — the automation of claims handling (over 55%) and, precisely, customer-service support through virtual assistants based on natural language processing. On paper, adoption looks mature.
Yet one figure tempers the enthusiasm. According to the same industry surveys, over 55% of Italian insurers launched AI projects in 2025, but only one in five moved beyond the pilot stage. The gap between "we have a project underway" and "we actually use it every day" is enormous, and it is precisely the space where the difference between hype and concrete results is decided.
On the customer-relationship front, something is genuinely shifting. IVASS data on complaints show that in the first half of 2025 the grievances received by insurance companies fell by more than 13% compared with the same period the previous year, with average response times steadily below 25 days, well within the regulatory limits. This is not down to AI alone, but the ability to clear simpler requests more quickly frees up time for the complex ones: and this is where conversational automation makes its most tangible contribution.
It is worth recalling a distinction that changes everything for those working in an agency: most of these figures concern insurers, not intermediaries. Insurers develop models on large volumes and substantial budgets; an agency or a brokerage firm operates on a different scale, with different needs and — this is the good news — with automation opportunities that are often far simpler and faster to deliver.
What an AI assistant can handle well in an agency
An AI assistant truly delivers when it focuses on high-volume tasks with low advisory value and low emotional charge. These are the tasks that today absorb hours of front-office staff time without generating any additional revenue. Let us look at them concretely.
Repetitive questions
Every agency receives the same questions over and over: "When does my car policy expire?", "What documents do I need to file a claim?", "Are you open on Saturdays?", "Does my cover include roadside assistance?". A conversational assistant connected to the client records and policy data can answer this first level instantly, at any time and on any channel — website, client portal or messaging app. For the client it means an answer in ten seconds instead of a phone call during office hours; for the agency it means removing dozens of micro-interruptions from the table each day.
The condition for it to work is that the assistant draws on real, up-to-date information, not generic answers. A chatbot that "invents" an expiry date or a coverage detail is worse than no chatbot at all: it generates errors that someone will later have to answer for.
The first filter on claims
Opening a claim is notoriously the most time-consuming phase: the client reports the event in an unstructured way, often by phone or WhatsApp, and the intermediary has to gather the initial information, work out what type of claim it is and route it appropriately. As we explored in our article on claims handling for intermediaries, this phase costs an average of 30 to 45 minutes per event.
An AI assistant can take care of the initial triage: gathering the essential data (date, place, type, parties involved), automatically classifying the claim, indicating which documents the client needs to prepare, and opening an already structured case in the agency's system. It does not settle the claim — that is not its job — but it hands the operator an orderly record instead of a phone call to transcribe by hand. The time recovered shifts from paperwork to the relationship.
Appointment booking and preliminary data collection
The third area where automation pays off is appointment management. An assistant can offer the available slots in the calendar, set the meeting and — a step that is often overlooked — collect the useful information in advance: the vehicle details for a motor policy, turnover and sector for a business policy, the composition of the household for a personal cover. The intermediary arrives at the meeting already informed and can devote the appointment to genuine advice, rather than filling in a form in front of the client.
Where the machine stops (and the human does not)
Here lies the boundary that no serious vendor should promise to cross. An AI assistant does not provide insurance advice. Choosing the right cover for a business, assessing a complex risk, balancing premium and deductible against the client's actual situation are tasks that require judgment, experience and professional responsibility. They are also the heart of an intermediary's value: what justifies their commission and what sets them apart from a direct purchase on an online comparison site.
Then there is the emotional dimension. A client who calls after an accident, a theft or a bereavement is not looking for efficiency: they are looking for someone who understands them. Entrusting that moment to a bot, however well trained, is the fastest way to turn a loyal client into a lost one. IVASS itself, in the speech by its Secretary General on AI and insurance intermediation, stressed how artificial intelligence redefines the agent's role not in competition with the human being, but as a tool to enhance their advisory and relational skills. This is the right framing: the automated assistant frees time from repetitive tasks so that the intermediary can devote more of it to those in which they are irreplaceable.
The obligation many forget
There is one aspect of conversational technology that marketing materials tend to skip over: regulatory obligations. The European Regulation on artificial intelligence (AI Act, Reg. EU 2024/1689) sets out precise transparency obligations for those using systems that interact directly with people. In practice: when a client converses with an AI assistant, they must be informed that they are talking to a machine and not to a flesh-and-blood operator. This is not a formality — it is a consumer protection, and ignoring it exposes the agency to reputational as well as regulatory risks.
In Italy the framework was further defined by Law 132/2025, in force since 10 October 2025, which confirmed IVASS's role as market surveillance authority for the insurance sector, including as regards artificial intelligence, alongside AgID and ACN as national authorities. For an agency this means one simple thing: the AI assistant must be designed from the outset to be transparent, traceable and compliant. A bespoke solution makes it possible to build these rules into the very workings of the system; a generic solution, designed for different markets and regulations, often does not.
How to build an assistant that truly serves
The difference between a chatbot the client happily uses and one abandoned after two attempts does not lie in the underlying technology — language models are now accessible to everyone — but in three design choices.
The first is a clear definition of scope. An effective assistant knows exactly what it can do and, above all, knows how to recognise when it must hand the conversation over to a person. The handover to the operator is not a failure of the system: it is its best-designed function. A bot that insists on answering a question it cannot handle produces frustration; a bot that says "let me put you in touch with one of our advisors on this" and transfers the conversation with all the context already gathered is doing exactly its job.
The second is integration with the agency's systems. An assistant disconnected from the client records, the policy portfolio and the management software can only give generic answers, as if from a brochure website. What generates value is the assistant that knows the individual client's position: renewal dates, active coverages, ongoing cases. This requires the system to talk to the existing infrastructure, not to replace it from scratch.
The third is tone of voice. An agency has an identity and a way of speaking to its clients built up over time. An assistant that answers in the anonymous language of any old service weakens that relationship. Customising the language, the workflows and the information is what turns a technical tool into a coherent extension of the agency.
This is why, at A126, we develop conversational assistants as vertical micro-applications built on the intermediary's real processes, rather than as standard modules to be adapted as best one can. The tool is shaped around the organisation and integrates with the systems already in use — the same logic we apply to all our solutions for intermediaries, as described in our article on software for insurance intermediaries.
The right AI in the right place
The chatbot is neither the revolution promised by marketing nor the useless gadget feared by those wary of technology. It is a tool that, applied to the right tasks — repetitive questions, first filter on claims, bookings and preliminary data collection — gives back precious time and improves the agency's responsiveness. Applied to the wrong tasks — advice, risk assessment, handling difficult moments — it becomes a boomerang. Maturity does not lie in automating everything, but in knowing where to stop.
For an agency or a brokerage firm, the choice is not "chatbot yes or no", but "which part of my work makes sense to automate, and how to do it without distorting the relationship with clients". It is a question that requires starting from real processes, not from the tool.
At A126 we support intermediaries in exactly this: we analyse operational workflows, identify the tasks where conversational automation genuinely creates value and build bespoke solutions, integrated with existing systems and compliant with the regulations. Get in touch for a free consultation and let us work out together where an AI assistant can make a difference in your organisation — and where, instead, it is better to leave room for people.
A126 Corporate Advisors — Bespoke digital solutions for insurance intermediaries.
