Will AI replace insurance brokers? 6 honest questions and answers
· FRANK
Every insurance conference this year has had an AI panel, and brokers keep asking us the same handful of questions. Some of them have clear answers. Others don’t, and we would rather say so than pretend otherwise.
What follows is what we have learned from building FRANK, an AI Operator now used by more than 350 agencies in Portugal and several other countries, and from watching closely where it works and where it doesn’t. By AI Operator we mean software that supports insurance brokers by actually doing several types of back-office tasks itself. For example, it reads a client’s request, logs into the insurer portals, fills in the forms and brings back a quote or a claim status for the broker to check.
Will AI replace insurance brokers?
Nobody can answer that with certainty, us included. But at FRANK we believe AI will not replace brokers, and history gives us some support. We believe AI is empowering insurance brokers, and will keep doing so.
In the past, several technological shifts, such as the dot-com boom and the rise of self-service insurance platforms, were expected to make the broker’s job disappear. The job survived and thrived through both, and through other industry transformations as well.
We believe that is mostly because people prefer to trust a human’s expertise in assessing risk and explaining the complexity and language of insurance, and above all because they value the ease, trust and support of a human relationship.
What we can describe is what AI does well today and what it doesn’t.
It is good at the repetitive work that surrounds a policy: typing the same client data into several insurer portals, collecting quotes, opening a claim, chasing a status update. “Most brokers spend 80% of their time on back-office tasks instead of being with clients,” says António Castro, our CEO. McKinsey estimates that around a third of an agent’s sales tasks could be automated with current technology. The two figures measure different things, but they point the same way: a large part of the job is administration.
AI is much weaker at the parts clients actually pay a broker for. Judging which cover fits a family or a small business. Explaining an exclusion before it becomes a problem. Noticing that a client’s life has changed. Being the person who picks up the phone after a fire or a flood. We have not seen software that does these well, and it is not what we are trying to build.
The bigger unknown sits on the client’s side. Self-service insurance platforms already exist, and AI agents are now starting to shop for insurance too. In September 2026, for example, Meta launched Muse in the United States, an AI assistant that, like many others, can shop for insurance on the user’s behalf. Insurify blocked it from its site, while Kin and Neptune opened theirs to it. It is not available in Portugal. As more channels for selling insurance become available, they will probably take their share of the market, as technology has done before. But we believe the broker will remain the main player in this industry.
In Portugal, the channel is getting smaller in headcount but not in business. The insurance regulator ASF counted 10,024 brokers and agents in 2025, fewer than in 2024, while their commissions grew 9.7% to €1,475M. Fewer people handling more business is likely to put more pressure on back-office time, whatever AI ends up doing.
At the same time, with all this technology and information at hand, clients themselves are changing. They are becoming more knowledgeable about insurance, more price-aware and more demanding about service. Brokers must move with the times to meet, and exceed, already high client expectations. This is where FRANK has positioned itself: as a supporting team member for every insurance broker, so that together we can protect people better.
Which tasks can AI actually do in an insurance agency today?
Today, an AI Operator can reliably support insurance brokers with:
- Reading client requests from email and WhatsApp and flagging what’s missing
- Logging into insurer portals and filling in the data
- Collecting quotes and analyzing proposals across insurers
- Opening claims and following their status
- Answering questions about the book of business, such as which policies renew next month
- Many other tasks
Quotes are where the difference is easiest to see. Brokers tend to quote fewer insurers than they could, because every extra insurer means another portal and another questionnaire. When the quotes run in parallel, the client gets a wider comparison, not just the insurers that are easiest to work with.
AI is also getting better at unusual cases, such as a handwritten form or a policy with special conditions. It can help there, but human verification and judgment are usually still recommended. That is also why, at FRANK, every result goes back to the broker before anything reaches a client.
One example of what this looks like in practice: after Storm Kristin hit Leiria, Portugal, in early 2026, we opened a free platform to every broker in Portugal to speed up home claims. The broker sent the client a form, and FRANK passed the data to each insurer and followed the claim status. We did not replace anyone; we built the infrastructure that helped everyone involved work and connect better.
How does AI change the relationship between brokers and their clients?
That depends more on the agency than on the technology. Time freed from the back office does not go to clients on its own. It can easily disappear into more admin.
The agencies we see getting the most out of it decide in advance what the hours are for. The renewal review they never had time to prepare, with proposals from several insurers already compared. The call to the client who just bought a new car. Following up on a claim before the client has to ask.
The client still talks to the broker. What changes is how prepared the broker is for that conversation, and how much faster, better and easier it becomes with FRANK by their side. Some of our clients have started saying good morning and thank you to FRANK, the way they would to a colleague, which tells us more about how the work is shared than any feature list.
Where should an insurance agency start with AI?
There are many AI systems, from smaller open-source LLMs and model-agnostic AI harnesses to the largest proprietary AIs, such as Claude or ChatGPT. All of them can work well in many processes. Most of them are general-purpose. FRANK was built from the ground up for insurance and its processes.
We use several of these high-end models, and on top of them we have built several layers of proprietary software, platforms and automation tools. Whatever tool an agency chooses, the rollout matters as much as the tool itself.
How to start depends on the agency. Some prefer to go one process at a time, chosen carefully, learning and integrating each one well. Others, more tech-savvy, prefer a broad rollout, with as much automation as possible from day one. At FRANK we can adapt to different rollouts.
One step we recommend is to spend one or two weeks measuring first, so you know your benchmarks and where your time goes: how many quotes, how many portals per quote, how many claims and policy changes. This makes it easier to measure the improvement once any automation is in place. Many clients then start with a process that repeats every week and takes the most time. Quotes are often the easiest place to start, because the time saved is simple to measure. Agencies with back-office teams usually know where the hours go, and those teams are typically the most eager to see that work automated.
Expect the first weeks to need attention. Portal credentials have to be kept up to date, and someone should check every result until the team trusts the output. Judging the tool on its first few days usually says more about the setup than about what it can do.
Before choosing any tool, it is worth checking a few things. Does it work with the insurer portals you already use? Does it connect to your management system or ERP? Can it receive requests from email and WhatsApp? And can anything reach a client without your approval? For reference, FRANK connects to the broker ERPs Libax, Gemese, MPM and i2S, as well as Gmail, Outlook, Google Calendar, WhatsApp and Salesforce. We also build tailor-made solutions, as many of our enterprise clients have very specific needs.
Is it safe to use AI with client data?
It can be, but “AI” covers very different things, and the risks are often not where people look.
Pasting client data into a public chatbot is the most common one. Once it is sent, it has left the agency’s control. Sharing portal credentials with aggregators is another: those logins can be used to take over accounts or commit payment fraud, a risk we describe in our whitepaper “Pipes or Harmony”. And bots that hit insurer portals at high speed overload them, which is why insurers block them.
Questions worth asking any provider: where portal credentials are stored and who can access them, whether each agency’s data is kept separate, whether every action is recorded, whether a person approves everything before it reaches the client, where data is hosted, and whether client data is used to train AI models.
At FRANK, our answers are these. Portal credentials are encrypted in a dedicated vault, and each agency’s data is kept separate. FRANK works inside insurer portals with the agency’s own access and sends requests one at a time, so insurer servers are not overloaded. Every action is recorded and can be reviewed. Nothing reaches a client without the broker’s approval. Data is hosted in the EU, client data is not used to train AI models, and a GDPR data processing agreement is available.
Regulation is moving too. Since January 2025, the EU’s DORA regulation has required insurers to manage technology risk, including risk that comes from their partners. The largest brokers also fall under it directly. How insurers will apply it to their broker networks is still taking shape, but it is reasonable to expect more questions about the tools brokers use.
How do you know if it’s worth it for your agency?
Start with your own numbers, not a vendor’s. For one or two weeks, track the time that goes into work that is mostly copying data between systems: quotes across several insurers, opening claims, chasing status updates, preparing renewals.
Then look at where that time sits. If it is a few minutes spread across the day, automation will change little. If it is concentrated in a few people or a few processes that repeat every week, that is where a tool can make a visible difference.
Two other things matter as much as the hours. The first is how many insurer portals your team works with, because every extra portal multiplies the copying. The second is whether someone in the agency will own the first weeks: keeping credentials current, checking results and deciding what the freed time is for. Without that person, even a good tool tends to stall.
Agencies using FRANK report around 10 hours freed per broker, per week. Treat that as a reference point, not a promise. It varies a lot with volume, with the insurers involved and with how the team works.