The term describes a property of how data is captured, not of any AI feature. A book is AI-ready when the facts that priced each risk (the answers to the application questions, the sum insured, the cover, the duration) are stored as columns rather than in documents; when the premium can be recomputed from stored rate rows; when the decision (accept, refer, decline, and what the underwriter then did) is explicit; and when every claim points to its policy with date, cause and amounts as numbers.
Most insurance data is not AI-ready because legacy systems stored the outcome of underwriting but not its inputs: proposal forms as PDFs, rating in a spreadsheet beside the system, claims in a separate application keyed by hand. The information cannot be recovered after the fact at any reasonable cost, which is why AI-readiness is decided by the platform a book is written on, from the first policy.
On Soldek
Soldek stores a book this way by construction: application questions are field definitions, answers are kept per field on the quote, rate tables and underwriting rules are data keyed on those fields, claims sit on policies with fixed status transitions, and the whole environment exports as one workbook from Settings → Your data. The engine itself stays deterministic; the data is the MGA's to model.
Your dataQuestions
- Is AI-ready data the same as having AI in the software?
- No. AI-ready describes the data: structured, labelled, joined, exportable. A platform can be AI-ready without any model in it, and a platform with a chatbot can sit on data that cannot train anything. For an MGA the first is worth far more than the second.
- How do I check whether our data is AI-ready?
- Ask for the application answers of a policy from two years ago as fields, not as a PDF; ask which rate factors produced its premium; ask for its claims with amounts, joined by key. If each of those is one query, the data is AI-ready.