A weekly report should help someone decide where to investigate or act. AI is useful for drafting the explanation after the data and calculations are trustworthy. Start with exports from your approved reporting system, metric definitions, comparison periods and a clear question.
Google Ads and Google Analytics can differ because their reporting definitions, timing and attribution differ. Treat those differences as something to investigate, not as a gap for AI to invent away. A change in reported conversions is an observation; an explanation of what caused it requires additional evidence.
What you will leave with
A short performance brief containing reconciled results, material changes, plausible hypotheses and owned next actions.
Step 1
Define a fair comparison
Record the period, timezone, currency, filters and conversion definitions for both datasets. Match the reporting grain: campaign per day is different from campaign per week. Note whether recent results are still processing and whether conversion credit uses click time or conversion time. Use the same interpretation on both sides of the comparison.
The comparison uses documented, compatible periods and definitions.
Step 2
Check and reconcile the inputs
Inspect row counts, duplicates, blank identifiers and unexpected totals. Reconcile summed spend, clicks and conversions to the source export. Do not join tables until you understand the key and expected match count. Keep a note of exclusions instead of silently deleting inconvenient rows.
Source totals and retained rows reconcile, with exclusions explained.
Step 3
Calculate with formulas or code
Calculate cost per recorded conversion (CPA) as total spend divided by total conversions, not an average of campaign CPAs. Use the same principle for other aggregate rates. A zero denominator is unavailable, not a meaningful zero rate. The AI Lab’s local KPI workbench provides a small example you can inspect and reproduce.
Try this instruction
Calculate totals and weighted rates from these validated rows. Show formulas and reconcile totals. Do not average campaign rates. Mark zero-denominator results unavailable. Return a calculation table before writing the narrative.
Calculations are reproducible and denominator rules are explicit.
Step 4
Separate observation from hypothesis
Ask the agent for the largest changes and the segments contributing to them. Label factual observations separately from possible explanations. Include sample size, tracking changes, mix shifts and missing context where relevant. Avoid treating a simultaneous change as proof of causality.
The narrative distinguishes measured changes from explanations to investigate.
Step 5
Make the review decision-ready
Finish with the question each change raises, the additional evidence needed, and an owner for the next step. A reviewer checks every material number against the calculated table. The final brief should make it easy to understand what changed, what remains uncertain and what happens next without repeating the entire export.
Every material number is checked and each next action has an owner.
When you get stuck
The AI explanation sounds certain despite thin data.
Require separate observation and hypothesis sections. Ask what additional evidence would confirm or contradict each explanation.
Ads and analytics totals do not match.
Compare definitions, dates, attribution, filters and freshness before diagnosing a tracking defect.
Terms explained
- Attribution
- The rule or model assigning credit for an outcome to marketing interactions.
- Weighted rate
- A rate calculated from summed numerators and denominators, such as total clicks divided by total impressions.
Sources and further reading
The workflow and templates are teaching guidance. The sources below support the underlying tools and concepts; product setup and availability can change.
Take it into your next task
Your completion checklist
0 / 5 checked
Checks are saved on this browser. Mark them only after you inspect the evidence. These are your checks, not an automated certification.