FlowsBook a callSign inCreate free accountSign up
AI employees and automation

An AI agent for insurance companies: risk around insured sites and new policy leads

Two jobs an insurer can hand to AI employees: watching storms and road incidents near the sites it covers, and finding new businesses that need commercial cover.

Part of the guide AI employees

By the xMap team · · 5 min read

An AI agent for insurance companies earns its keep on two fronts at once. On the risk side it checks the forecast and the roads around every property or business you insure and warns the claims team before a storm, not after. On the growth side it finds businesses that have just opened in your target areas and need commercial cover, and hands them to your agents or brokers. In xMap Flows each front is its own AI employee, built from blocks such as My data, Weather forecast, Traffic incidents and Nearby places.

This is written for claims managers, underwriting leads and distribution teams at carriers, managing general agents and larger brokerages. It explains what each employee checks, how to wire it, and where the data stops.

The risk side: what is coming towards your insured sites

Most insurers already hold a list of insured locations with addresses. Upload it once as a table in My data (a CSV with a latitude and longitude, or an address that can be placed) and every risk employee can work from it.

  • Weather forecast adds to each site the rain expected over the days you choose, how much, the next rainy day, the strongest wind and the hottest temperature, from national weather services. It takes up to 500 sites a run; sites within about 2 km share one forecast.
  • Traffic incidents checks around each site for road closures and roadworks right now, which matters when adjusters have to reach a cluster of claims after an event.
  • A Filter keeps only the sites that cross your thresholds, so the email lists twelve properties at risk, not twelve thousand that are fine.

A worked example on the risk side (made up)

Picture a home and small-business insurer in Auckland with a few thousand policies across the region. Its claims manager wants a warning whenever heavy rain is due near a group of insured addresses, so loss adjusters can be put on standby and policyholders reminded to clear gutters.

  1. Schedule: every day at 05:30, Pacific/Auckland time.
  2. My data: the insured locations table.
  3. Weather forecast: each incoming place, the next 3 days, rain and wind.
  4. Filter: rain total of 40 mm or more, or wind above the line the team agreed.
  5. xMap LLM, the whole list: group these sites by suburb, say which day looks worst for each group and how many policies it holds; never add figures that are not in the data.
  6. Email to the claims manager and the duty adjuster lead.

Most days the filter keeps nothing and the flow sends nothing, which is the right outcome. On a wet week the email is a short grouped list with a paragraph on top. The storm claims readiness template follows this pattern, and closures on the adjusters' routes adds the road check for the days after an event.

The growth side: new businesses that need cover

Commercial lines grow when a new cafe, clinic, workshop or shop opens and has not chosen an insurer yet. Nearby places searches xMap's places database (more than 300 million places, refreshed every month) for the kinds of business you write, and can keep only places opened in the last 6 or 12 months where the data has an opening date.

To give a sense of how much there is to work with in one city: xMap's places database lists 1,640 restaurants within 10 km of the centre of Sydney (November 2026), and of the 303 nearest the centre, 293 are independent. Independents rarely have a group insurance programme, which makes them the natural audience for a small-business policy.

  1. Schedule: every Monday morning.
  2. Area of interest: the cities or districts your agents cover.
  3. Nearby places: restaurants, beauty salons, dental clinics, or whatever your appetite allows, opened in the last 12 months, independents only.
  4. Only new, so each business is offered to an agent once.
  5. xMap LLM: score 0 to 100 how likely this business needs property and liability cover now, and say why in one line.
  6. Split by territory, so each agent receives the businesses in their own area.
  7. Email to each agent, or a row in a shared sheet in Google Sheets.

The new shops and restaurants for commercial insurance template is the ready-made form of this list, and independent brokers and agencies uses the same idea to find distributors instead of customers.

Mistakes to avoid

  • Mixing the two jobs in one flow. Risk warnings must go out every day without fail; lead lists can wait for Monday. Two employees, two timetables, two owners.
  • Thresholds nobody agreed. If the rain line is too low the claims team learns to ignore the email. Set it with them, then adjust after a month.
  • Treating the AI score as underwriting. The score says who to call first. Your own rules still decide what you can write.
  • Messaging policyholders without a check. If the flow writes to customers, put an Approval before the send so a person reads the first few batches.

What lands in the team's inbox

The claims manager gets the grouped risk email on mornings when something is due, and nothing otherwise. Agents get a weekly list of new businesses in their own area, each with its address, phone or website when the data has them, a score and one line on why. Both employees keep a thread in the Team inbox, where a reply such as add wind above 70 km/h or skip franchise outlets changes the flow once someone confirms it.

The pattern fits the field work around claims too, which the field operations guide covers in more depth. Insurers will find more examples on the insurance page.

Questions

Can it read our policy system directly?

It can read a table you upload to My data, a sheet in Google Sheets, or your own PostgreSQL, BigQuery or Snowflake database with the Database query block. It only reads; nothing is written back unless you add a block that saves.

Does it predict claims?

No. It reports forecasts and road incidents near your sites and leaves the judgement to your team. It is an early warning, not a loss model.

Which countries does the weather cover?

The forecast covers the whole world, for up to 16 days, with temperature, rain, wind and UV for each site that reaches the block.

Can it write to policyholders before a storm?

Yes, with an Email or a WhatsApp step after the filter. Keep an Approval in front of it at first so a person checks the wording and the list.

Templates that do this

IndustriesInsurance

Read next

More on ai employees and automation · Every article · Templates

Put it to work

An AI employee in xMap Flows does this on live data, on its own timetable. Start from a template or describe it in plain words.