Google Sheets

Our evaluation

Google Sheets is a reasonable starting point for a small invoice tracker when the formulas remain inspectable and one person owns the model. The published agent-executed test supports that narrow judgement: duplicate and invalid rows were separated, amount and date edits recalculated the expected totals, and formulas survived the XLSX export. I would keep raw imports separate from calculation columns, use a fixed reference date while checking, and reconcile totals against an independent ledger before relying on the sheet. The tradeoff is maintenance: a shared grid does not itself enforce your invoice process, approval rules or access discipline. This task did not establish multi-user behaviour, large-sheet performance or Gemini quality. Evaluate those separately before treating Sheets as a wider business system.

Editorial judgement from the official documentation and the published agent-executed invoice test. This is a scoped assessment, not a full-product rating or a claim of personal hands-on testing.

What the invoice test established

The fixture contained 20 fictional invoice rows, including duplicate IDs, invalid values, Unicode text and a quoted comma. An agent imported the rows, then wrote formulas through the authenticated Google Sheets connector. An independently calculated ledger supplied the expected results before native readback.

The formulas classified 12 valid unique rows, two duplicates and six quarantined rows. All 12 expected counts and GBP totals matched: outstanding invoices totalled GBP 1489.45, with GBP 1000.00 overdue at the fixed reference date of 9 October 2026.

Increasing an overdue invoice by GBP 50 increased both relevant totals by GBP 50. Moving the reference date forward one day moved the GBP 300 due-today invoice into the overdue total without changing the outstanding amount. The agent restored the original amount and reference date and checked the original results again.

The exported native XLSX retained all 20 raw rows, all 73 formulas and the 12 cached metric results. Independent ZIP/XML parsing checked those outputs against the fixture. The CSV export contained the Summary sheet's displayed values.

Where the evidence stops

This was one small synthetic invoice workflow. It does not establish large-file speed, ongoing reliability, security, multi-person collaboration or accounting compliance. The subscription or Workspace edition was not established, so the test cannot confirm a paid feature entitlement. Gemini was not used.

The CSV export was explicitly the first Summary sheet, not an export of the Invoices tab. The native export adapter did not select an arbitrary worksheet for CSV export. The XLSX contained the raw rows and formulas, but no wider compatibility benchmark was run.

Fewertools agents executed and documented the task. This page does not claim that Clinton personally tested the product, and it does not add an overall verdict, winner or quality score.

Price and billing scope

Selected official observations read 9 October 2026; this is an observation date, not a price-change date. Confirm eligibility, usage limits, tax and the final account offer before paying.

UK/GBP standard Workspace rates; no introductory discount or currency conversion used. Workspace is a bundle, not a standalone Sheets fee. Storage and bundle scope ↗ · Commitment and payment terms ↗.

These public rates do not establish the Workspace edition or paid entitlements of the account used in the invoice test. Gemini was not executed.

A useful trial before committing

Copy the published synthetic invoice fixture into a separate sheet. Keep raw rows separate from formulas, compare all expected counts and totals, edit one amount and the reference date, restore both, and inspect the XLSX export. Do not use private customer data for this trial.

Inspect or reproduce the result

• Download the 20-row synthetic invoice fixture.
• Inspect the independently calculated expected results.
• Download the native XLSX with raw rows and formulas.
• Download the Summary sheet CSV.

Use the complete test record for the formula checks, environment, exact reproduction steps and artifact hashes. Use synthetic data when repeating the task.

Related evidence