Monthly reports

Keep calculations deterministic

Prepare the source workbooks, expected columns, formulas and reporting period. Identify the steps that are ordinary data processing: joining tables, summing values, refreshing charts and preserving workbook structure. These often belong in scripts, spreadsheet formulas or Power Query.

Use a model where language interpretation helps, such as categorizing free-text explanations or drafting commentary from already verified totals.

Build the monthly flow

1. Validate input files, dates and column names before combining them.
2. Normalize and aggregate data with deterministic code. Retain reconciliation totals for each source.
3. Produce a compact table of verified metrics and exceptions.
4. Ask the model to draft the narrative from that table, with references to the relevant metric rows.
5. Review the narrative and write it into a copy of the reporting template. Recheck formulas and charts after export.

Verify the workbook

Test missing files, renamed sheets, formula cells and changing row counts. Compare totals before and after the transformation. Avoid sending an entire complex workbook to a model and treating a plausible summary as proof that every formula relationship was preserved.

An illustrative reconciliation can check that the sum of source-row counts equals the number of combined records after documented exclusions. Define those exclusions explicitly instead of hiding discrepancies in prose.

Account for data restrictions

Some organizations cannot upload reporting data to an external service. Check that condition before designing a hosted-model step. SeedRouter provides text model APIs; it is not a spreadsheet calculation engine or a replacement for your organization's data rules.

See business data queries for related validation patterns and text models for model selection. Measure model use only for the steps that need it, using current prices.