Garbage in, garbage out. That is a motto insurers are focused on as AI adoption accelerates. No matter how sophisticated an AI system is, if it is working from inconsistent, incomplete, or siloed data, the insights it produces will be worthless.
Insurers don't lack data. In fact, most have mountains of data that keep growing—from policy systems, claims files, third-party risk scores, IoT devices, digital customer interactions, and more. The challenge is how this data is managed. Many carriers rely on highly centralized data teams to clean, interpret, standardize, and provide data for the entire organization. This team often becomes a bottleneck by necessity: they are asked to translate data they did not create and often do not deeply understand. Requests for data access or analytics pile up, creating delays and frustration, while business units learn to work around centralized processes to move faster.
This is not a sustainable model in an AI-first environment. It's important that managing data enables speed, accountability, and trust at scale. Increasingly, a data mesh model meets those parameters.
Data becomes a shared responsibility
Data mesh is not just a technology shift. It is a structural and cultural shift in how organizations manage and collaborate around data. It is built on the idea that data should be managed by the experts closest to it—not solely by a central data team removed from daily operations
Rather than funneling data from every department into a warehouse managed by IT, data mesh distributes responsibility across business domains. In insurance, many of these domains already exist: underwriting, claims, billing, actuarial, distribution, as well as other groups. Data mesh aligns data ownership to these domains and asks each to treat the data it produces as a high-quality product that the rest of the business can easily use.
The people who use the data every day are responsible for ensuring it is correct, clear, and understandable, while the central data team provides the tools, guardrails and support so the different domains do not have to reinvent the wheel.
The four principles of data mesh
Data mesh is not just about technology – it is a socio-technical shift that requires rethinking how organizations perceive and manage data. It is built on four key principles that are essential to solving the scalability issues of centralized data platforms.
1. Domain-oriented ownership
In the traditional model, business teams generate data but expect IT or analytics teams to clean, define, and interpret it. This leads to mismatched definitions—such as claims and finance using different interpretations of "paid loss"—and to slow turnaround times when insights are needed quickly.
With domain-oriented ownership, the team that creates the data is responsible for it. They own, maintain, and serve their data as a product to others. For example, the claims department maintains a clear, documented, and continuously updated claims loss history dataset. Underwriting does not need to file a request or wait weeks for data cleanup; it can access that dataset directly and trust that the definitions and logic behind it are accurate and consistent.
This is a cultural shift from "we send data to IT to fix" to "we own our data and stand behind its quality."
2. Data as a product
Treating data as a product means designing it for ease of use and with the end user in mind. Data should be understandable, discoverable, and reliable. It should have: service level expectations meaning the organization knows how often data is updated and how accurate it should be; versioning that will notify users if there are changes to how the data is calculated; and, a feedback loop enabling users to request improvements or ask questions. Insurance data products might include written premium histories, loss run reports, producer performance insights, or rate plan documentation.
3. Self-serve data infrastructure
This is the backbone of data mesh. It reduces bottlenecks and gives business users direct access to the data they need, when they need it. The role of the central data team evolves from performing custom data pulls and cleanings tasks to providing shared tools and platforms that allow domains to publish and consume data themselves. This may include a searchable data catalog, analytics workspaces, visualization tools, and automated data quality checks.
4. Federated governance
Insurance is a regulatory-driven industry, and data governance cannot be decentralized entirely. Federated governance establishes standard definitions, security rules, and compliance requirements across the enterprise, while allowing each domain to apply those standards in the context of their work. The goal is consistency without reintroducing the logjam of centralized control.
Instilling a data mesh culture
Adopting data mesh is not simply an IT re-architecture—it is an organizational transformation. Success requires thoughtful change management. Leaders need to clearly define ownership roles, invest in training, and empower teams to make decisions about their data. Starting with one or two domains—often claims and underwriting—helps demonstrate value quickly.
Collaboration must also be encouraged. Many insurers still operate with siloed teams that rarely share data or context. Data mesh requires open communication and shared priorities across departments. This takes leadership reinforcement, not just new tooling.
As AI becomes core to the insurance operating model, carriers can no longer afford data systems that are slow, centralized, and fragmented. Data mesh provides a path forward—one that aligns data responsibility with business expertise, strengthens trust in shared information, and enables faster, more confident decisions. For insurers seeking to modernize, data mesh is not just a new architecture. It is a new way of working.
