How carriers can turn AI ambition into measurable outcomes by treating their data as a strategic asset.
Today, every industry is racing to deploy AI into multiple aspects of their business looking for that bump in productivity, insights unseen before now, and new streams of revenue. Few industries stand to gain more than insurance. The opportunities are vast:
- Real-time, dynamic quoting
- Hyper-personalized offers
- Automated claims processing
- Supercharging agents and customer support with infinite knowledge
But here’s the catch: before you fund your first, or more likely, your next AI project, you need to ask a harder question: What kind of data are you feeding your models, and in what format? Because without the right foundation, even the best AI initiative is set up to stall.
Humans in the loop can smooth over AI’s rough edges for now, but that’s only a short-term fix. In insurance, close enough is still wrong, accuracy and freedom from hallucinations aren’t optional, they’re table stakes. To unlock the full potential, AI needs more than oversight. It needs a foundation of complete, consistent data and schemas designed for the task. With that in place, the upside is enormous: sharper risk selection, faster time-to-market, and more personal customer experiences.
AI Ambition vs. Data Reality
The challenge is that insurance data is fragmented and stuck in the past. Customer, underwriting, and behavioral data often sit in separate systems. Product and rating data is trapped in legacy formats. Behavioral data is siloed within a single team spread too thin across the organization. And while many large language models can process unstructured inputs, they tend to produce answers that sound right but aren’t. In insurance, “close enough” is still wrong. The better the data, the better the output, and the closer you get to AI that works in production.
Example: Event Liability Quoting
Imagine a couple shopping online for wedding insurance. They expect a real-time price for their venue. To get it right, the model needs highly-structured inputs like event type, location, number of guests, and special risk factors. But in many carriers, those attributes are buried in underwriter notes or hidden in PDFs. Without structured data, the AI either guesses or fails, and the opportunity for instant quoting disappears or worse, misprices.
Understanding the risks of bad data is the first step. The next is knowing what good data actually looks like. In insurance, “AI-ready” data should be:
- Complete - Every required data attribute is captured.
- Good: A home policy record includes property type, square footage, year built, and updates.
- Bad: “House, renovated recently” in a random text note marked “other”.
- Consistent - Data uses standard definitions and formats across the business.
- Good: Every policy uses the same codes for “roof type.”
- Bad: “Shingle,” “shngl,” and “composition” all show up in different systems about the same roof, because they have different nomenclature.
- Connected - Customer, underwriting, claims, and behavioral data can be joined without manual reconciliation and translation gymnastics.
- Good: Product definitions exist in a machine-readable schema that normalizes properties and flows easily across systems.
- Bad: Claims live in a separate system with only a customer email for manual reconciliation.
But here’s the thing: it’s not enough to just have data that’s complete, consistent, and connected. You also need the right structure around it. That’s where your schema comes in.
A schema is simply the blueprint for how your data is organized. Think of it as the architectural plan for your products: what attributes exist, how they relate, and how they flow through quoting, binding, and servicing. Whether you realize it or not, your schema, or lack of schema, is already shaping your business.
- With a rigid schema, adding new coverage is like trying to add a bathroom to a finished skyscraper, every floor gets disrupted. Attributes are hard-coded into forms, data tables are fixed-width, and dependencies hide somewhere in legacy code. Every small change requires IT to rebuild and retest downstream systems.
- In a flexible schema, adding that same coverage is more like snapping a new module into a modular building, the core stands, and new pieces fit in cleanly. Attributes are defined as reusable objects, the model allows for extensible properties, and rating logic references structured attributes instead of one-off algorithms. New coverages can be added through configuration, not months of code rewrites.
This is where Buddy’s ION™ (Insurance Object Notation) comes in. ION is a flexible, insurance-native language we built to capture the entirety of an insurance product in a structured, machine-readable way. It’s designed so new coverages, attributes, pricing logic, and more can be added without rewriting your systems from scratch. In other words, ION doesn’t just describe your products, it future-proofs them.
Your schema isn’t just a technical detail. It dictates how quickly you can move, what kinds of AI you can actually deploy, and whether your next product idea is a sprint or a slog. In insurance, your schema is your product strategy.
Legacy Gaps and AI Expectations
For most carriers, the distance between today’s data reality and what AI actually needs is still wide. Legacy systems were never built with AI in mind. They were designed to file policies, calculate rates, and process claims, not to deliver structured, machine-readable product definitions. In addition to legacy systems, existing regulation requirements around insurance and customer privacy expectations also lead to separating data into discreet business functions. As a result of these existing hurdles, much of the information an AI would need for quoting, claims, or personalization is either locked in inaccessible formats or scattered across teams and systems.
The impact is familiar: every new AI initiative starts with weeks to months of data wrangling, manual cleaning, and workarounds. Instead of training a model, teams are stuck reconciling fields, translating rating logic, or piecing together incomplete records. By the time the model runs, the data it depends on is already out of date.
Meanwhile, AI expects something entirely different: explicit, structured, and versioned product definitions it can directly ingest. Modern models thrive when attributes are consistently defined, when behavioral data connects to underwriting and claims, and when the product itself exists in a schema that reflects reality.
AI isn’t magic. It can only amplify what you give it. If the data is fragmented and brittle, the results will be too.
If AI is on your roadmap, the fastest path forward isn’t another proof of concept, it’s getting your product data into a structure built for AI from the start. That foundation determines whether your next initiative becomes a pilot that fizzles or a capability that scales.
That’s where we can help. At Buddy, we’ve built our ION™ engine to help carriers unlock AI while keeping pace with product strategy. If you’re exploring AI opportunities, let’s talk. We’d love to hear the outcomes you’re chasing and share how we’ve helped other carriers lay the groundwork.
