Analyse your Google Ads campaigns with GPT Astra

The starting point

• Two comparable exports: campaign, type, date, currency, cost, impressions, clicks, conversions and conversion value
• Conversion definitions, attribution window and tracking changes
• Business goal, margin if known, promotions and budget changes

1. Prepare comparable exports

Separate campaign types and levels of detail. Do not add a total row to its underlying rows. Remove unnecessary personal data. Record currency, time zone and included conversions. Flag recent periods that may still receive attributed conversions as incomplete.

2. Validate calculations first

Recalculate CTR = clicks / impressions, CPC = cost / clicks, CPA = cost / conversions and ROAS = conversion value / cost. Sum underlying counts before calculating ratios; do not average campaign CPAs. Mark zero denominators as not calculable and reconcile totals with the source.

3. Separate observation, hypothesis and action

For each observation, request scope, supporting figures and absolute importance. A higher CPA may relate to traffic mix, tracking, seasonality or competition. The file may not distinguish these causes. Turn each hypothesis into a check, then a bounded test with an owner and success criterion.

4. Prepare a human decision

Select the three most useful investigations. Compare commercial value, volume and conversion quality before proposing changes. The analysis prepares an account review. Starting a test or changing spend remains a separate decision with an agreed scope.

A practical example

Analyse the attached Google Ads exports. First validate columns, periods, currency, duplicates and conversion definitions. Show totals and formulas. Calculate ratios from sums, never by averaging ratios. Flag zero denominators and insufficient data. Produce: five numerical observations linked to source rows; separate hypotheses; missing checks; and three proposed tests with goal, scope and stopping criteria. Do not invent causes or data. Do not modify campaigns.

How to assess progress

• The example does not establish higher profitability: conversion value and margin are unknown.
• Do not mix click-through and view-through conversions without documented definitions.
• Keep the file, calculation and context behind every recommendation.