Meta ROAS on your site: the setup, step by step
For the brand judged on ROAS: the tag or app, the feed, recommendations on every page, server-side events counted once, value-based audiences, and the numbers in your own ad account.
- What to prepare before day one
- What goes live, in order
- What to measure in week one and week four
Open the setup checklist
Enter your work email and the checklist opens on this page. We also send you the link, so you can come back to it.
Thanks. The checklist is open below.
We use your email to know who reads our research. Privacy policy
What to prepare
The tag or the app
One tag and your product feed, or the app on WooCommerce or Shopify. Stock, prices and images as they are now.
Read access to the ad account
Read-only. The agent reads spend and reported purchases to put next to the orders the store records.
A baseline in your own orders
Items per order and average order value for the last months, with and without a recommendation touch. This is what the test is read against.
One event id per purchase
Server-side purchase events carry the same id as the browser event, so Meta counts each purchase once.
What goes live, in order
Recommendations on every page
Product page, cart and category pages. The agent decides per visitor what to show next; the visible effect is the second product in the basket.
The same agent in the automations
Abandoned-cart and abandoned-browse emails built from the same recommendations, for each customer.
Server-side purchase events
Purchases sent from the server with the event id, deduplicated against the pixel. Check match quality in the ad account.
Value-based audiences
Audiences built from predicted customer value, synced to Meta, with existing customers kept out of retargeting.
Keep a holdout
Part of the visitors see the site without recommendations. A before-and-after test, as the bedding brand ran, is weaker than a holdout.
Change nothing in the buying
Same campaigns, creatives and budgets through the test, so the site is the only thing that moved.
What to look at, and when
| When | What to look at | What good looks like |
|---|---|---|
| Week 1 | Recommendations rendering on every page, purchase events deduplicated, match quality in the ad account | Each purchase counted once by Meta; every page shows picks for the visitor |
| Week 4 | Items per order and average order value, orders with a recommendation touch against the holdout | The direction the bedding brand saw in reverse: orders with a recommended product larger than ad-only orders |
| Month 3 | Return on ad spend inside your own ad account against the months before, and the share of orders with a recommended product | Return holds while spend grows, read in your own orders, not only in Meta's report |
Common mistakes
Changing creatives and budgets during the test. If three things move at once, nothing can be credited to the site.
Counting purchases twice. A browser event and a server event without one id become two purchases in Meta's report.
Judging on Meta's ROAS alone. Meta counts by its attribution settings; the store's own orders are the reference, with both reports kept.
Switching recommendations off in stages to see what happens. The bedding brand did, and the basket shrank first; a holdout gives the same answer without the loss.
Back to the use case, or see it in your store
See which of last week's visitors you missed
Thirty minutes, with your store and ad accounts open. Then 30 days free. A holdout group decides: if the agent doesn't add orders in 30 days, you don't pay.
Book 30 minutes
You'll talk to one of us.