How to compare ad performance across platforms with an AI agent
Spend, CPA and ROAS across Google, Meta, TikTok and Pinterest in one table, reconciled against GA4 so platform self-reporting is kept honest.
IntentAnswer 'which platform earns its budget' with comparable numbers and explicit caveats, not four self-graded report cards.
- 5
- Steps
- 7
- Tools
- 3
- Rules
- 3
- Failure modes
When to use it
Use it when
- Allocating next month's budget across platforms.
- A platform claims credit that totals more than the business saw.
Do not use it when
- Single-platform accounts.
- When GA4 is absent and the user needs settled truth: without a neutral referee this skill produces a labelled comparison, not a verdict.
What it needs first
Required
Preconditions
- Same window and currency across platforms.
- Each platform's attribution setting read and quoted in the output.
Inputs
window- Comparison window.Default: last 30 days.
kpirequired- CPA or ROAS.
Procedure
In order. Every tool named is one the gateway ships, and links to its reference.
Pull spend and conversions per platform, campaign-level, one window.
Normalise to one currency and one table; label each column with its attribution window.
Pull GA4 key events and revenue by channel for the same window.
Compute each platform's over-claim ratio: platform-reported conversions over GA4 channel conversions.
Rank platforms on the user's KPI twice: platform-reported and GA4-viewed.
Decision rules
The observable condition, what it lets you conclude, and what takes the conclusion back.
When
Platform conversions exceed GA4 channel conversions by more than 40%
Conclude
Flag heavy view-through or modelled credit on that platform; rank it on the GA4 view for decisions.
Unless
GA4 consent-mode gaps are known to depress that channel; then present both and say neither is settled.
When
The two rankings disagree on the top platform
Conclude
Lead with the disagreement itself: it is the finding, and budget moves should wait for an incrementality read.
When
A platform has under 50 conversions in the window
Conclude
Exclude it from ranking and say why; its CPA is an anecdote at that volume.
Evidence every conclusion must carry
- Both rankings shown, with the over-claim ratio per platform.
- Every column labelled with its attribution window and currency.
Where the agent stops
- Budget shifts decided on this table go through budget-reallocation and Safe Writes; nothing moves from here.
How it goes wrong quietly
The cases where the analysis is wrong and still looks right. Read them before trusting a number.
- Attribution windows differ silently across platforms; a table without labels is fiction with gridlines.
- GA4 last-click under-credits upper-funnel platforms; the GA4 view is a referee, not the truth.
- Currency conversion at today's rate distorts a month of spend; use the account's own reported currency totals where possible.
What the answer contains
- 01The table
- Platform × spend, conversions, KPI, both views, labelled.
- 02Over-claim ratios
- Per platform, with the likely mechanism named.
- 03Decision guidance
- What the table supports deciding, and what needs an incrementality test.