The Challenge: "There's Gold in That Inbox"
Referrals in delegated authority insurance happen when risks exceed a coverholder's binding limits and need internal underwriter sign-off. Referral rates are a critical efficiency metric and excessive referrals suggest poorly calibrated authority or appetite drift, eroding the efficiency gains that make delegation valuable in the first place. Despite representing a small proportion of premium, referrals take up a significant amount of the underwriters' time.
At Carbon, we created a dedicated referral inbox for our coverholders, which rapidly transformed into a treasure trove of untapped insight. Each day it brought a fresh wave of emails: policy details nested in PDFs, risk descriptions fragmented across Word documents, key data points concealed in Excel files. The gold was there, but digging through hundreds of emails manually, was simply unrealistic.
The solution, in theory, seemed obvious: "use AI." But how, and would it provide us with the meaningful insights we were after?
In just over a month, our team at Carbon built an AI-powered pipeline that pulls referrals from our inbox every thirty minutes, makes sense of them, and transforms the data into an internal dashboard on Graphene Insights, our analytics platform.
Now we can finally answer questions like: Which partners send us the best referral business? What is our referral loss ratio? Do referrals perform better or worse than our standard book?
The Problem We Needed to Solve
Referrals arrive in every format imaginable. Often, critical information is spread across multiple attachments and multiple emails. We needed to condense this and create a consistent structure, before we could unearth that gold!
How We Built It
Step 1: Capturing the Emails
We built a tool that automatically pulls referral emails directly from Outlook and stores everything: email content, attachments, sender details and, timestamps, into our data warehouse.
Step 2: Making Sense of the Content
To unlock the meaningful insights, we applied Gemini, Google's AI model, to read lengthy email narratives, PDFs, and spreadsheets, across conversation history. Gemini was able to extract key insights and put useful information into our prescribed structured format. All the key information was there for us to analyse. Now, we need to integrate it with our existing data!
Step 3: Linking referrals to outcomes.
Using our data transformation tools, we combined the AI-processed referral summaries with our existing policyholder and claims data. This gave us the complete picture; not just what was referred, but how those referrals actually performed.
Step 4: Driving profitable decisions.
The final piece was building dashboards that our underwriting and business development teams could actually use. The data team can now explore referral performance themselves, looking at things like:
- Claims patterns by referral
- How quickly we are converting referrals to policies
- Whether referred business performs differently to our standard book
What's Changed
This solution has genuinely transformed how we work with referrals:
- Speed: What used to take weeks of manual review now happens automatically within hours Our mechanism is scalable, currently processing up to fifty referrals every half hour, but with room for more!
- Clarity: We finally know which referral partnerships are truly valuable.
- Scale: We can handle hundreds of emails monthly, and the system grows with us. GCP allows us to scale the system with ease.
- Immediate: Teams across the business can access insights whenever they need them.
What's Next
This is just the start. We are already thinking about what else becomes possible:
- Smarter referral decisions: Using historical patterns to assess new referrals instantly.
- Saving underwriter time: Automatically flagging referrals that match previously rejected profiles.
The referral inbox is no longer an opaque box. It's become one of our most valuable data sources, and we're only just beginning to tap into what that means for our business.
To find out more about Carbon, or to book a demo of Graphene Insights, our analytics platform transforming underwriting, reach out to us at www.carbonuw.com/contact
