How to run a weekly ad performance review with an AI agent
The Monday morning numbers across every connected platform: spend, results, movers, and what explains them, in one pass instead of four exports.
IntentProduce the week's cross-platform picture with explanations, not just deltas.
- 5
- Steps
- 10
- Tools
- 4
- Rules
- 3
- Failure modes
When to use it
Use it when
- Every week, same day, same windows, so numbers are comparable.
- Before a weekly client or team meeting.
Do not use it when
- Mid-incident: use campaign-anomaly-investigation.
- For windows under 7 days, where platform attribution lag dominates the signal.
What it needs first
Preconditions
- At least one ad platform connected and ticked; the review covers whatever subset is connected and says which.
- Windows aligned: last full week against the week before, same attribution settings per platform.
Inputs
week- The week under review.Default: last full Monday-to-Sunday week.
kpi- The metric the account is steered by: CPA, ROAS, or volume.Default: the platform-reported conversion metric.
Procedure
In order. Every tool named is one the gateway ships, and links to its reference.
Pull spend, conversions, CPA/ROAS by campaign on each connected platform, both weeks.
Line the platforms up in one table, in account currency, and rank movers by absolute spend-weighted change.
For the top movers, read what changed: budgets, statuses, delivery.
If GA4 is connected, add the site's own view: sessions, key events and revenue by channel for both weeks.
If Search Console is connected, note organic movements that reframe paid ones.
Decision rules
The observable condition, what it lets you conclude, and what takes the conclusion back.
When
A mover's change is under 15% week over week
Conclude
Report it as noise-range unless it repeats a trend from previous weeks.
When
Spend moved but efficiency did not
Conclude
Attribute the result change to budget, not to performance.
When
A platform's conversions moved more than 25% while GA4's matching channel did not
Conclude
Flag probable measurement or attribution movement before any performance story.
When
The change log shows human edits on a mover
Conclude
Lead with the edit as the explanation candidate, dated and attributed.
Evidence every conclusion must carry
- Every mover carries both weeks' figures, not just the percentage.
- Every explanation cites its source: a change event, a delivery status, or is marked as hypothesis.
Where the agent stops
- The review proposes; budget or status changes it suggests go through Safe Writes with human confirmation.
How it goes wrong quietly
The cases where the analysis is wrong and still looks right. Read them before trusting a number.
- Comparing a week that includes a holiday against one that does not, without saying so.
- The most recent days under-report conversions; a Sunday-heavy week reads as a decline for 48 hours.
- Mixing attribution windows across platforms makes the cross-platform table look precise while being incomparable; label each column.
What the answer contains
- 01The table
- Platform × spend, conversions, efficiency, both weeks, one currency.
- 02Movers
- Top changes ranked by spend-weighted impact, each with its explanation or hypothesis.
- 03Watch list
- Trends one more week from being findings.