The myth of the data prerequisite
Every carrier we have ever spoken to has a data quality story. Most of them are true. None of them are a reason to delay AI deployment by twelve months. The reason is simple: AI workflows are scoped. You do not need clean data across the enterprise to deploy a motor FNOL agent - you need clean data on motor FNOL. That is a tractable problem.
What 'workflow-scoped data' looks like in practice
For a single workflow, the data you actually need is small, knowable, and usually already exists in the core or a downstream warehouse. The work is in the integration, the canonicalisation, and the quality gates - not in a multi-year enterprise data programme. We routinely take a workflow from data assessment to live agent in under eight weeks.
- Identify the 8–12 fields the workflow actually touches
- Build a quality gate that fails closed on the fields that matter
- Backfill or instrument the fields that are missing - not the warehouse
The cost of waiting
Every quarter a carrier delays AI deployment on the argument of data readiness, the carrier's competitors get a quarter of compounding learning. The data programme can run in parallel. It usually should. It is rarely the prerequisite the data team thinks it is.