Insurance agency AI training: a step-by-step guide

An insurance agency’s inbox is always filled with the same questions: "How much is comprehensive auto?", "When does my traffic policy expire?", "Is earthquake insurance mandatory, and how much?", "Does health insurance cover this?". Instagram posts, the WhatsApp line, the bottom of your website: it doesn’t matter; the same twenty questions repeat a hundred times a day. At 10 p.m., while the agent is dealing with a claims file, a customer writes "my policy expires tomorrow, can I get a quote right away?" and waits until morning for an answer. Every quote request answered late goes to another agency, and every customer whose renewal window you miss renews with a competitor.

inTusell takes over this load. But to do that, it first needs to learn your agency, the branches you work in and your quote process. This article explains how an insurance agency or broker trains the inTusell AI from scratch, step by step. This is the insurance chapter of our sector-by-sector "how to train your AI" series, and it is the pillar article of the series.

Quick answer

Why is training important?

inTusell is not a simple chatbot; it is an AI sales assistant and CRM that answers messages on WhatsApp, Instagram and web chat, books appointments, follows each customer through the sales pipeline and behaves like an experienced agency assistant. A good assistant doesn’t know your branches on day one either. They have to learn which types of insurance you work in, which coverages you offer and which questions to leave to the agent. Training makes the AI do two things right: give the correct information (coverage scope, process information, business hours) and stop at the right boundary (not inventing a premium or coverage it’s unsure about, not giving individual advice, leaving the exact quote to the agent).

An untrained assistant either gives an overly general answer or guesses on a topic it doesn’t know. In insurance, guessing is expensive. A wrong premium or a baseless sentence like "yes, this comprehensive policy covers everything" misleads the customer and creates liability for the agent. That is why the most critical part of training is teaching the AI what it knows, and teaching it to hand off to the agent when it doesn’t know, or when an exact quote or advice is required. A well-trained assistant stops when it doesn’t know.

Who is it for?

This guide is for insurance teams whose inboxes are crowded with quote questions and renewal requests, who answer the same questions over and over:

• Agencies working in traffic, comprehensive auto, earthquake and home branches
• Solo or small-team insurance agencies and brokers
• Teams selling health, life and complementary health insurance
• Agencies receiving dozens of quote requests a day via Instagram and WhatsApp
• Offices struggling to manually track many simultaneous renewals
• Brokers managing multi-step quote processes for SME and commercial insurance packages

You don’t need a technical team. All of the steps below are completed from sections in the panel, without writing code.

The structure of training: 5 core steps + 2 ongoing steps

Splitting training into two groups makes your job easier. The first 5 core steps get the assistant working; these are the one-time part of setup. The 2 ongoing steps that follow (reviewing replies and turning fixes into rules, and turning on channels) are the continuous part you sharpen as you use the assistant. You’ll find all of them in order below.

1. Load the policy and coverage knowledge base

The foundation of training is the knowledge base (RAG). In the panel, you upload all of your agency’s textual information into the Knowledge Base section. Every time the AI generates a response, it automatically searches these sources and uses only the information written there. In insurance, this means the branches you work in and your coverage processes.

Supported source types:

Every file you upload is automatically chunked and becomes searchable via pgvector. If you upload a "comprehensive auto coverage table.pdf" and a customer then asks "is glass breakage included in comprehensive cover?", the AI finds the right section and explains the coverage scope.

What must absolutely be loaded for an insurance agency: the branches you work in (traffic, comprehensive auto, earthquake, home, health, life, personal accident, business), each branch’s coverage scope and exclusions, the information required for a quote (such as license plate, registration details, age, address), claims and renewal process information, business hours and the most frequently asked questions. An important boundary: the knowledge base is a living source that must be kept up to date. If a coverage has changed or you’ve added a new branch, update the knowledge base; that is the foundation of a correct answer. The exact premium, though, is not a fixed value you write into the knowledge base. Because the premium changes by person and risk information, the AI does not state a premium; it collects the quote request (see step 4).

2. Set up quote meetings and business hours

The knowledge base explains "which coverages you offer"; the scheduling system makes it possible to give the customer a real meeting time. In insurance, this is most often a quote meeting, policy consultation or claims-file meeting. In the panel, you define three things:

Once these three are defined, the AI no longer offers times out of thin air. The AI collects the date, time and topic of the meeting. What happens next depends on your account: if the AI booking tools are on, it checks your business hours and booked times and books the free slot itself; if not, it passes the request to your team with those details, and you confirm the time or the customer picks a free slot on your online booking page. To see which applies, ask for a meeting in AI Test. The system prevents the same time from being booked twice (double-booking). If Google Calendar is connected, every new booking is also copied one-way to the agent’s calendar. A reminder goes out 1 day and 2 hours before the meeting; the customer can reschedule or cancel through the link in the message, so fewer hours go to waste. An event you add in Google, however, does not block the slot in inTusell.

The real weight of the quote process in insurance, though, lies less in scheduling and more in follow-up: collecting the quote request in full, sending follow-up messages to a lead who went quiet, and putting the renewal date in front of the team with a staff reminder on the contact card. inTusell does not track policy end dates itself. We’ve gathered every detail of this routine in a separate article: insurance quote and policy renewal automation. You can also see the quote and scheduling capabilities on the solutions page and the insurance solution page.

3. Set the persona and tone

In the first two steps you taught the AI what to say; now you’ll determine how it says it. There are two settings in the panel:

• Assistant name (ai_persona_name): The name it introduces itself to the customer with. Most agencies choose a name that fits their team, inspires trust and is clear.
• Tone (ai_tone): The typical tone for insurance is calm, reassuring and explanatory; the language of an advisor who tries to understand the customer’s need and risk and moves them toward a quote meeting without applying pressure. You can make the tone more corporate or more casual to match your own agency culture.

Multilingual note: The AI replies in whatever language the customer writes in. For customers writing from abroad in English, Arabic or Russian, you do nothing extra; the assistant detects the language automatically and replies in it while preserving your tone, so the foreign customer’s quote request gets collected instead of lost.

4. Set the response rules and boundaries

The persona defines "who it is"; the response rules define "how it behaves." In insurance, this step serves two purposes at once: it passes on your sales discipline and marks where the AI should stop and hand off to the agent. You add rules through the AI Manager Chat in plain language, just as you would brief a teammate.

Typical rules for insurance agencies:

• "When a quote is requested, don’t state the premium yourself; collect the information needed for the quote (license plate/registration/age/address) in full and forward it to the agent."
• "Explain coverage scope from the knowledge base; but don’t give personal advice like 'this policy is enough for you': leave the needs analysis to the agent."
• "Don’t state a coverage, exclusion or claims outcome you’re unsure about; say you’ll ask the agent and hand off."
• "Always pass claims files, payment and policy cancellation processes to the agent."

The most common mistake in insurance is the AI stating information it doesn’t have as if it were true: an exact premium, say, or a coverage promise like "definitely covers it." On top of that, insurance is a regulated field: the AI does not give individual insurance advice. It only informs and directs to the licensed agent, just as it routes to an expert in legal and health. In inTusell, two layers work against this. The main control is the system instruction, which constrains the AI to use only the information in the knowledge base from the start, not produce a premium or advice, and hand off when it’s unsure. The second layer is an output filter (a pattern scan plus a second-model check): it scans responses and flags misleading statements; it does not block everything. The real assurance comes from constraining the AI from the start, so that it won’t produce exaggerated promises or baseless premiums.

If you want to compare two different approaches, you can use the A/B testing feature: for example, you put a more informative tone next to a closing that calls the customer to a quote meeting more directly, at 50/50 traffic, and measure which one converts into more quote requests.

5. Teach from past conversations

This is the step that takes training from "good" to "specific to your agency." In the panel, you upload the audio recordings of your past sales conversations (MP3, MP4, WAV, M4A) and mark each one as Won or Lost, for example, did the customer buy/renew the policy, or did they go to another agency?

The system uses these recordings in two ways:

This way, the AI learns from your won calls how your best agent understands the customer’s need, how they move someone saying "let me think about it" to a quote meeting, and how they handle the "it’s cheaper at a competitor" objection. The KVKK side is protected: uploading requires explicit consent, and phone, email, ID and card numbers are masked in the transcript before processing. This step matters especially in insurance, where sensitive data such as ID and policy details is involved.

This step is not mandatory, but don’t skip it. The knowledge base teaches the AI "what" it knows; past conversations teach it "how you sell." When the two are combined, the assistant truly resembles an experienced insurance advisor.

6. Review and turn fixes into rules (ongoing step)

The first five steps get the AI working. Everything after that is the ongoing part that sharpens it over time. You check replies on the AI Test screen and in Conversations. Say the AI explained a coverage’s exclusion incompletely or miscounted the required documents for a branch: you add the right fact to the knowledge base or write a rule in AI Rules. You can switch the ready rule packs in the Behavior Library on or off, and write a plain instruction such as "ask about needs before pricing" in the AI Admin Chat. The way you explain coverages and your information-collection flow build up the same way.

The Label Manager is a separate quality record: customer satisfaction replies, the outcome of handed-off conversations, the won/lost classification and labels a team member adds by hand land there. You approve, reject or edit them; these corrections are kept as a record, but they do not change how the AI writes replies today. What changes behavior is the knowledge base and the rules. If you want to see every reply before it goes out, use Full approval mode or the conversation’s Approval mode.

In the first weeks, we recommend a short daily look at the replies. Review is the "live" part of training.

7. Turn on channels (ongoing step)

Once training is ready, you put the assistant in front of customers. inTusell gathers all channels into a single inbox: Instagram DM, Instagram comments, WhatsApp, Telegram and Web Chat.

Ways to turn on channels:

• Instagram: connects through Meta login, where you pick your Page and Instagram account; a Professional account linked to a Facebook Page is required.
• WhatsApp: connects by scanning a QR code or with an 8-character linking code sent for your phone number; your number stays the same and no Meta Business approval is required. The official WhatsApp line (Cloud API) opens after Meta approval.

In insurance, the first two channels to turn on are usually WhatsApp and Instagram DM, because that is where quote questions and "my policy is expiring" messages arrive most. Instagram comment automation is live too. DM replies are already on once Instagram is connected; comment replies have their own switch on the Instagram card ("Instagram comments"), and it is on by default. If you like, you can turn it off, answer the comments under posts by hand for a while, and switch it on when you are ready. If you connected Instagram before September 24, 2026, you need to reconnect it once for comments.

There are three account-wide modes (My business › Instructions › AI Settings › Advanced AI Settings): Intern (default; sends replies itself, asks you before a cancellation or refund), Senior (also handles cancellations and refunds itself) and Full approval (every reply on WhatsApp, Instagram and Telegram waits for your approval). In every mode the AI handles quote question traffic and hands claims files and conversations requiring an exact quote to the agent; the handoff lands on the Decisions page. You can also put a single conversation in Approval, Shadow or Paused mode. For the details of Instagram and WhatsApp automation: Instagram and WhatsApp quote automation for insurance.

How long does training take?

How long it takes depends mostly on whether your documents are ready. The steps of a working setup:

1. Loading branch and coverage information into the knowledge base (the easiest step if your coverage tables are ready)
2. Quote meeting types and business hours
3. Persona, tone and first response rules
4. Connecting channels

But there is no moment when "training is over." Over the first weeks, as you add the gaps you see to the knowledge base and AI Rules, the assistant sharpens to fit your agency. Keeping the coverage information up to date and tone tests are ongoing improvements too. Setup is fast; mastery is continuous.

What it isn’t

Putting inTusell in the right category matters, because a wrong expectation leads to a wrong setup.

• It is not an online policy sales/comparison site. It does not issue policies on its own or publish price comparisons; it learns your branches and runs the communication and quote intake with the customer.
• It is not an insurance advisor. It does not perform an individual risk assessment and say "buy this"; it informs about coverages and directs the final advice to the licensed agent.
• It is not a premium-calculation engine. It does not state the exact premium; it collects the information needed for a quote, and the agent finalizes the price.
• It is not a bot that invents information. It does not state coverage, exclusion or claims information it’s unsure about; it clearly says it doesn’t know and hands off to the agent.

In short: not a bot giving canned answers, but an AI sales assistant and CRM that represents your agency to the extent you train it, stops where it doesn’t know, gives no advice, and hands off to the agent.

Frequently asked questions

How long does AI training take for an insurance agency?

A working setup comes down to these steps: loading branch and coverage information into the knowledge base, defining quote meeting types and business hours, setting persona and tone, and adding a few response rules. How ready your documents are matters most for timing; the real refinement builds up over the first weeks as you add the gaps you see in AI Test and in real conversations to the knowledge base and AI Rules. Training is not one-off, it is continuous.

What happens if the AI states the wrong premium or coverage?

The AI only uses the coverage and process information you have loaded into the knowledge base; it will not invent a premium amount, a coverage scope or a claims outcome it is unsure about: it says it does not know and hands off to the agent. The exact premium and policy quote are always the agent’s responsibility; the AI collects the quote request, and the agent finalizes the price.

Do I need technical knowledge to train the AI?

No. All training is done from the panel; no code, API key or developer is required. You upload branch and coverage information as PDF/Excel, enter your quote meeting types and business hours, and write your response rules in plain language. You connect Instagram through Meta login and WhatsApp with a QR code or a linking code.

Does the AI give insurance advice?

No. inTusell informs about coverages and processes, collects the quote request and brings a suitable prospect to a meeting with the agent; but it does not give individual insurance/risk advice such as "buy this policy" or "this coverage is enough for you." A personalized risk assessment and the final quote are the licensed agent’s job; the AI protects this boundary from the start.

Does the AI answer quote and policy questions that come in at night?

Yes. It replies after hours too to questions like "how much is comprehensive auto, when does my traffic policy expire, is earthquake insurance mandatory" that arrive on Instagram and WhatsApp at night, collects the necessary information from a prospect requesting a quote and forwards it to the agent. When the team arrives in the morning, the complete quote requests are waiting in the inbox.

Is customer and policy data safe under KVKK?

Data is stored encrypted and each business’s data is kept separate from the others. You download your account data with "Download My Data (KVKK)" and manage the KVKK notice your customers receive, with your own privacy link, under "Legal notices". Uploading sales call recordings requires explicit consent; phone, email, ID and card numbers are masked in the transcript.

Next step

You’ve trained your assistant; next comes using it in daily operations. The next article in the series shows how the inTusell you’ve trained runs on a real workday: the quote question that comes in at night, quote request intake, renewal reminders and customer follow-up. Continue straight from there: how an insurance agency uses inTusell.

This was the insurance pillar article of our sector-by-sector series. We’ve set up the same structure in other sectors too; for example, for gyms see gym AI training, for tourism tour agency AI training, and for healthcare clinic AI training.

If you’d like to see it live before starting setup, use Get a demo so we can open your panel together in a short session, or write to hi@intusell.com. For package and quota details, see the pricing page, and for other guides, browse the all articles list.