A model point file reconciliation checklist

A model point file is only as good as the evidence that it ties to the policy system. This checklist is how we would reconcile one, whatever software you use. It assumes a month-end or year-end extract and a reference file: production, or last period's.

1. Before you start

Agree the reference. Production or last period's file, and which version of it.
Fix the dates. The valuation date, the extract cut-off, and the dates of the unit prices and exchange rates.
Write the success criteria down, by product. For example: monetary totals within 0.5%, rates and indicators exact on at least 90% of rows, no unmatched policies without a reason.
Name the doer, checker and reviewer. Three different people, agreed before the work starts.

2. Check the files as they arrive

Every file against its interface layout: fields, types, allowed values and links between files.
Control totals from the extract: policy counts by product and status, premiums, fund values, sums assured.
Unit prices for every fund held, and exchange rates for every currency.
Each excluded record listed with its reason. Nothing disappears quietly.

3. Build, then check the totals

Rebuild the control totals from the model point file and agree them to the extract to the cent.
Record the lineage of each column: which file and field it came from, and every transformation applied.
Translate policy-system codes with a documented value map, and record any default used for a blank.

4. Reconcile field by field

Match policies on a stable key. List policies missing on either side.
Compare every field of every matched policy. Use exact matching for rates, dates and indicators, and a tolerance only for monetary fields.
Score each product against the success criteria, not just the book as a whole. A good total can hide a bad product.

5. Reconcile the movements

Opening count plus new business, less exits, plus or minus transfers, equals the closing count, by product.
Check that exits have a reason and a date in the period, and new business a start date in it.

6. Explain every exception

Classify each break: a data error, a timing difference, a change of method or a known limitation.
Record known data issues in a data limitations register, with their effect on the technical provisions and their treatment.
Don't sign off with unexplained breaks.

7. Sign off and keep the evidence

Doer, checker and reviewer each sign. Nobody checks or reviews their own work.
Lock the file. A correction becomes a new version with its reason, not an edit.
Keep an evidence pack: the inputs and outputs with fingerprints, the reconciliation report, the data-quality report and the sign-off.

How AutoImpact does it

TRACE does each of these steps in one place: interface checks, control totals, the build with lineage, field-by-field reconciliation scored by product, movements, exceptions, three-person sign-off and an evidence pack. Your team adds saved column mappings and data transformations, each applied once a second person approves it, and each run records how many rows every step changed. On AutoImpact's synthetic demo book, 2,000 model points are built and 99.8% match last year's file, with each exception listed. See model point file reconciliation.

This note is general information, correct as at 4 November 2026. It is not legal, accounting or actuarial advice. Check the sources for the current text.