Build Your Own Voice AI for First Notice of Loss Automation
A small insurance company needs a provider who builds custom intake logic directly into its existing claims system. Look for providers that deliver full source code without per-call fees or long-term vendor lock-in.
The goal is to automate the data entry of a First Notice of Loss (FNOL) call, not just transcribe it. A successful system extracts the policy number, claimant name, incident details, and a summary, then writes this structured data into your claims management software. This frees your adjusters to focus on the policyholder, not on typing.
We built an FNOL system for a regional agency with 8 adjusters handling 150 FNOL calls per week. We deployed a system in 3 weeks that automated the initial data entry, reducing the adjuster's post-call work from 7 minutes to a 45-second review. Their monthly transcription and API costs are under $100.
What Problem Does This Solve?
Many agencies first try basic phone system tools. Twilio Studio is great for simple IVR menus, but its visual builder cannot handle the complex, non-linear flow of a real FNOL conversation. When a stressed policyholder interrupts or gives information out of order, the rigid logic breaks and defaults to a human agent, defeating the purpose.
Next, they try dedicated transcription services. These tools create a text file from the call audio, but an adjuster still has to read the entire transcript to find the key details and manually enter them into the claims system. This swaps one manual task (listening and typing) for another (reading and typing), saving almost no time and costing $0.02 to $0.05 per minute of audio.
Enterprise platforms like Five9 or Genesys offer sophisticated voice AI, but they are built for 500-seat call centers. A 15-person agency will pay for dozens of features they never use, face a multi-month implementation, and get locked into steep per-agent, per-minute pricing models that make no sense for the relatively low volume of FNOL calls.
How Does It Work?
We start by provisioning a dedicated phone number using the Twilio API. When a policyholder calls, an AWS Lambda function is triggered, which initiates a real-time audio stream to Amazon Transcribe for transcription. This provides a live text feed with an accuracy rate of over 94% for standard North American English.
This live transcript is streamed to the Claude 3 Sonnet API. We engineer a specific prompt that instructs the model to act as a claims intake specialist, extracting entities like policy number, contact information, date of loss, and incident location. The prompt also asks for a concise summary of the event. The Claude API returns this information as structured JSON in under 800 milliseconds.
Once the call ends, a second AWS Lambda function takes the structured JSON data. We write a custom integration using Python and the httpx library to post this data directly to your claims management system's API, creating a new claim record automatically. This entire backend is a FastAPI service, containing less than 1,000 lines of code, making it transparent and easy to maintain.
The newly created claim is flagged for human review. Your adjuster receives a Slack or email notification containing a direct link to the record. They see all the AI-extracted fields, the full transcript, and a link to the original audio recording. This transforms their job from data entry to data verification, a process that takes 45 seconds instead of seven minutes. The total AWS hosting and API costs for processing 500 calls a month are typically under $75.
What Are the Key Benefits?
Claim Filed 30 Seconds After Hang-Up
The system processes the call and creates a claim record in your system in under 30 seconds. Your adjusters can review the claim while the details are still fresh.
One-Time Build, No Per-Minute Fees
We complete a fixed-price build. You pay only for API and hosting costs, which are cents per call, instead of being locked into a high-priced SaaS subscription.
You Own the Code and the System
We deliver the full Python source code to your GitHub account. There is no vendor lock-in. You have complete control to modify or extend the system.
Monitored 24/7, Alerts in 60 Seconds
We use AWS CloudWatch to monitor every component. If the transcription API has an error or latency spikes, we get an alert and can fix it before it impacts operations.
Integrates With Your Claims Software
The system writes data directly into your existing claims platform, whether it is an industry tool like Applied Epic or a custom-built system. No new dashboards to learn.
What Does the Process Look Like?
Week 1: Process Mapping & Access
You provide read-only API access to your claims system and walk us through your FNOL questionnaire. We deliver a technical specification document outlining every data field to be captured.
Week 2: Core AI Engine Build
We build the transcription and data extraction pipeline. You receive access to a test phone number where you can perform test calls and see the structured JSON output.
Week 3: Integration & Deployment
We connect the AI engine to your claims management system and deploy it into production. You receive a live, tested phone number to begin routing real policyholder calls.
Weeks 4-6: Monitoring & Handoff
We monitor 100% of live calls for two weeks to tune the AI prompts and handle edge cases. You receive the full source code and a runbook for long-term maintenance.
Frequently Asked Questions
- How much does a custom FNOL system cost?
- Pricing is a fixed, one-time fee based on the complexity of your claims system API and the number of data fields required. A typical build is a 3-week engagement. There are no recurring license fees. You only pay for the underlying cloud and API usage, which is minimal. We provide a firm quote after our initial discovery call.
- What happens if the AI mishears a policy number?
- The system flags fields where it has low confidence, such as an alphanumeric policy number that doesn't match the expected format. These claims are tagged for priority human review. Since the adjuster always verifies the data against the transcript and audio, errors are caught before they become a problem. This human-in-the-loop design is critical.
- How is this different from a CCaaS provider like Talkdesk?
- Talkdesk provides software for human agents to use. We build the AI agent itself. Our system automates the data collection part of the call, creating the claim record. It can operate standalone or transfer the call to a human agent in Talkdesk after collecting the initial information. We replace the manual data entry, not the entire call center.
- Can the system handle distressed or angry callers?
- Yes. We can add a step that uses the Claude API to analyze the transcript for sentiment. If the caller's sentiment is detected as highly distressed or angry, the system can be configured to bypass further questions and immediately transfer the call to a live agent, providing the agent with the transcript up to that point.
- How is policyholder data secured?
- The entire system is built in your company's own secure cloud environment (AWS). Syntora does not store any of your data on our systems. Audio files are processed and then deleted immediately. All data is encrypted in transit using TLS 1.2 and at rest. We sign a standard NDA before any work begins.
- What if we change our intake questions in the future?
- Because you own the code, modifications are simple. Adding a new question and data field involves updating the Claude API prompt and the script that posts to your claims system. This is a small project, typically taking a few hours, and can be handled via our optional monthly maintenance plan.
Related Solutions
Ready to Automate Your Small Business Operations?
Book a call to discuss how we can implement ai automation for your small business business.
Book a Call