Use cases/Financial services/Underwriting submissions
AI underwriting for Australian insurers and brokers
AI underwriting means every broker submission is read when it arrives. Client details, schedules and loss history are set out in one file. The system checks the risk against your appetite and underwriting guidelines, pulls up similar past risks with their claims and lists what the broker still needs to send. The underwriter writes or declines every risk.
Talk to our team →- Supports
- Underwriter
- Reads
- Broker emails, Schedules, Loss history, Guidelines, Past risks
- Runs on
- Your servers, on premises
- Connects to
- Guidewire, WinBEAT, Sunrise Exchange, Steadfast INSIGHT
- Decides
- A person, every time
Architecture
Everything runs on your servers
We build it on an open-weight model such as Llama or Mistral, served with vLLM on NVIDIA GPUs in your server room, data centre or colocation. Submissions are indexed in PostgreSQL with pgvector, intake runs through your mailbox or Kafka, and the output is written back to the policy platform you already run. Brokers can use the same build to pull client details into each insurer format and set the returned quotes side by side.
- SourcesBroker emails / Schedules / Loss history / Guidelines / Past risks
- Your serversRead the submission → Check the appetite → Draft for underwriter
- ReviewUnderwriter
- OutputSubmissions read on arrival / Appetite and guideline check / Similar past risks and their claims / Missing information for the broker
It takes four jobs off the underwriter
- 01
Submissions read on arrival
Emails, PDFs and spreadsheets are read into one file with the client details, the schedule of risks and the loss history set out. The underwriter opens a file instead of an inbox.
- 02
Appetite and guideline check
Each risk is compared with your appetite and underwriting guidelines. The clause or rule that applies sits beside each result, so the underwriter can check it.
- 03
Similar past risks and their claims
Risks you have written before are pulled up with their claims history and the terms you quoted. Each one links to its file.
- 04
Missing information for the broker
The items a submission still needs are listed against what arrived. The list goes into a draft request to the broker, and the underwriter approves it before it goes.
It goes live in five steps
- Step 01
Collect past submissions
Submissions as they first arrived, with the quote or decline that followed and the claims on the risks you wrote. This is the test set and it decides the design.
- Step 02
Write down the appetite
We sit with the underwriters and list what they accept, refer and decline, and which documents each class needs. The rules are written down and encoded.
- Step 03
Test against past cases
The system reads submissions that already have a decision. Its appetite results are compared with what the underwriters decided, and the results set which classes go first.
- Step 04
Run in shadow
Underwriters work submissions as usual while the system reads the same ones beside them. They mark each miss, and those marks feed the evaluation set.
- Step 05
Underwriter writes or declines
The file reaches the underwriter with the checks, the similar risks and the missing items attached. Every quote and every decline stays with a person, and the log records each file the system read.
Software or a custom build
Broking systems such as WinBEAT carry the client and policy record, and Sunrise Exchange moves quotes and policy transactions between brokers and insurers. Steadfast INSIGHT carries the client record for Steadfast brokers. It became Steadfast Apps in September 2026. The Steadfast Client Trading Platform handles quoting and is now reached through Steadfast Apps. Use the intake and rating features in your policy platform or broking system first if submissions arrive in one format and your guidelines are already coded as rules. A custom build earns its cost when submissions arrive as emails with PDFs and spreadsheets in many formats, when appetite sits in guideline documents and in underwriters' heads, or when submission documents cannot leave your environment.
Guidewire / WinBEAT / Sunrise Exchange / Steadfast INSIGHT / Steadfast Client Trading Platform / Salesforce / SharePoint
In practice
We build this on servers the insurer owns. The model, the submission index and the logs sit inside your environment, and client and loss data is not sent to a public AI service. The retrieval pattern is covered in RAG on private infrastructure, and the build falls under LLM integration and RAG.
From 10 December 2026, APP 1.7 to 1.9 apply. An organisation covered by the Privacy Act must say in its privacy policy when a computer program makes, or does a key part of, a decision that could significantly affect a person's rights or interests and that uses their personal information. The OAIC lists computer programs involved in assessing an individual's eligibility for insurance among the decisions generally in scope of the transparency obligation. Your compliance team decides how that applies to each class you write. The system helps with a decision and a person makes it, which where to put the human covers. A person making the final call does not take a decision out of scope. The OAIC says a decision can be in scope even when a person reviews the program output.
Claims on the policies you write are handled under AI claims triage, and policy wordings and past files under document search. The wider picture for insurance has its own page.
Questions
What is AI underwriting?
It is the reading and checking that happens before an underwriter prices a risk. The system reads each broker submission, sets out the client details, schedule and loss history, checks the risk against your appetite and lists what is missing. The underwriter writes or declines the risk.
Does the AI quote or decline risks?
No. The underwriter writes or declines every risk. The system sets out the file, shows the appetite result with the rule behind it and drafts the request to the broker. Anything it is unsure of goes to the underwriter with the reason attached.
Can brokers use it too?
Yes. A broker build pulls client details from the broking system into each insurer format and sets the returned quotes side by side. The broker checks the comparison and chooses what to recommend.
We already run Guidewire or a broking system such as WinBEAT. Do we need to change?
No. We build beside the policy and broking systems you already run, reading from them and writing back. They stay the system of record.
Industries where this comes up
More financial services use cases
Document search / Credit file assembly / AML alert triage / Claims triage / Month-end close / Audit evidence matching / Advice file drafting / Lease abstraction / KYC and onboarding / Complaints handling / Regulatory reporting / Board and management reporting
Before you call
Tell us where your hours go.
We map the work, size the hardware and price what is worth automating.
Talk to our team →