Attribution and true performance
Every ad platform grades its own homework: Meta, Google, and TikTok each claim credit for the same sale. AI can build your weekly true-performance sheet from raw spend and Shopify totals; a real holdout test is the only thing that tells you what was actually incremental.
Fastest path: one prompt, end to end
🤖 AI prompt — paste into ChatGPT / Claude
You are a DTC finance analyst. Here is my last [4] weeks by channel: [paste weekly spend and platform-reported revenue for Meta, Google, TikTok] and my Shopify total revenue for the same weeks: [paste]. New-customer orders per week: [paste if you have them].
Build ONE markdown table with columns: Week | Total spend | Shopify revenue | Blended MER (revenue / spend) | Sum of platform-reported revenue | Over-attribution gap (platform sum - Shopify revenue) | New-customer CPA. Add a one-line read of the MER trend and flag any week where platform-reported revenue exceeds actual Shopify revenue.
Rules:
1. Use ONLY the numbers I pasted. Do not invent spend, revenue, or a ROAS I did not give you. If new-customer counts are missing, leave nc-CPA blank and say so.
2. Do the arithmetic explicitly so I can check it. Round money to whole units.
3. Call the holdout result "directional", not proof, and do not claim causality from the sheet alone.
Or do it in 4 steps
- Build a weekly true-performance sheet. One row per week: blended MER (total revenue / total spend), new-customer CPA, and each platform's reported ROAS side by side. If Meta, Google, and TikTok each report 5x but total revenue is only 3x total spend, someone is double-counting the same customer.
- Decide on MER trends, not single numbers. Platform-reported ROAS overstates what each channel actually drove, and recent attribution-window changes only widen the gap, so a rising blended MER while spend holds flat is the efficiency signal no platform can fake. Trust the trend over any one dashboard.
- Run a simple 2-week regional holdout, described honestly as directional. Pause a suspect channel (often brand search or retargeting) in one region, keep it on elsewhere, and compare total revenue. Incremental ROAS almost always lands well below the platform-reported number. A 2-week test is a signal, not proof; a full geo holdout runs 6-8 weeks.
- Give upper-funnel channels credit last-click hides. Awareness channels like YouTube and CTV often drive real lift a last-click dashboard shows as zero; Haus's analysis of 190 YouTube incrementality tests is one body of evidence that these channels are undercredited (haus.io). Don't cut them on in-platform ROAS alone.
Done looks like: a weekly sheet with blended MER and nc-CPA, plus one holdout read per quarter before you scale or cut a channel.
Worked example: platform sum vs reality (illustrative)
| Channel | Platform-reported revenue |
|---|
| Meta | $30,000 |
| Google | $18,000 |
| TikTok | $9,000 |
| Platform sum | $57,000 |
| Actual Shopify revenue | $40,000 |
Blended MER = $40,000 / $12,000 spend = 3.3x. The platforms "credit" $57,000, but the store made $40,000, so $17,000 is the same customer counted more than once. Trust the MER, not the sum.
Review the sheet weekly; run a holdout each quarter.