Data & security

Exporting your book for machine learning: tables, joins and formats

How to get a model-ready dataset out of Soldek: the full export, which sheets to join on which keys, and when to use the Client API instead.

Updated 13 November 2026 · Written and reviewed by the Soldek team

A model needs tables, not reports. This article is the practical side: where the data is, how it comes out, and how to turn the sheets into one analysis frame.

Three ways out

Full export
Settings → Your data → Export. One Excel workbook with one sheet per table: products, field definitions, rate tables, underwriting rules, clients, quotes, policies, insured persons, claims, invoices, bordereaux and the activity log. Any administrator, any time, at no cost; each download is logged.
Per screen
Reports, bordereaux, statements and most lists export their rows as Excel or CSV. Useful for a quick look at one slice; use the full export for modelling so every table comes from the same moment.
Client API
Your products with their questions, and your submissions with their answers and status, as JSON, under an API key you issue in Settings. The right choice for a pipeline that pulls on a schedule into your warehouse.

How the answers come out

Application answers are stored per field on the quote. In the workbook they appear in one column as JSON (one object per quote, keyed by field name), because every product has different fields and a flat sheet cannot have a column for each. The first step of any pipeline is to expand that column into one column per field, which every spreadsheet tool, pandas and SQL engine can do in one call. The field definitions sheet tells you the type and the allowed options for each field.

Building the training frame

  1. 1Start from the policies sheet: one row per bound risk, with inception, expiry, premium, status and quote_id.
  2. 2Join quotes on quote_id to bring in product_id, the answers (expanded), sum insured, cover and the channel.
  3. 3Join field definitions on product_id to type the answer columns and to band options consistently.
  4. 4Compute exposure from inception, expiry and any cancellation effective date; for schedule products, multiply by insured persons from that sheet.
  5. 5Aggregate claims per policy_id: count, sum of paid plus reserve (or settled where closed), earliest date of loss.
  6. 6Keep the rate tables and rules sheets beside the frame: they are the tariff the book was priced on, which the model's output will be compared with.

Good habits

  • Export the whole environment in one go rather than table by table on different days; the keys then agree.
  • Keep each export with its date. A model should be reproducible from a named snapshot.
  • Exclude the fields you would not be allowed to price on before modelling, not after.
  • Treat the export as personal data under GDPR: store it where your policy allows, pseudonymise client identifiers before handing it to a third party, and record the transfer.

See it running.

Book a demo and we show you the complete platform live. Still stuck? Write to hello@soldek.io.