You can improve ROAS by raising the revenue the advertising produces or by lowering the spend that produced it, and lowering the spend is faster, easier and available to anybody with a mouse. That is the trouble with the metric. A rising ratio is not evidence of anything by itself, so decide first whether you want a better number or a bigger business, because the shortest routes to the first one damage the second.
What follows is the order we work in on an account, and it is deliberately back to front. Almost every list like this opens with the ad account. Here the ad account is fifth.
Why does ROAS go up when you cut the budget?
Because spend is bought in order of quality. The first euro reaches the people most likely to buy anyway, and every euro after it reaches somebody slightly less interested. Halve the budget and you keep the good half, so the average return climbs while the total profit falls. Pausing a campaign is the most reliable ROAS improvement in the industry, which is why nobody puts it in a case study.
A ratio that improves while the total falls has not improved. It has been made smaller.
The test that settles it takes one line in a spreadsheet. Put gross profit after ad spend next to the ratio, week by week. If the ratio went up and that column went down, you did not find an efficiency. You found a smaller business.
First, which revenue are you dividing by the spend?
The store's, for the same dates, counted once. Every platform reports the revenue it believes it caused, on its own window, using its own rules, and the sum of those claims is reliably larger than the shop. Improving a ROAS built on a number that double counts means optimising towards whichever channel claims hardest. There is a whole note on the size of that gap and another on the email half of it.
Raise the order value before you touch the targeting
Order value moves the ratio directly and it moves it for every campaign at once, including the ones you are not looking at. Nothing in the ad account has that reach. On a store we built the whole conversion layer came to fourteen features in one theme: a three question skin quiz that routes to a set rather than to a single product, priced sets, a live cart total that shows the saving as it changes, and a cart drawer that never takes the visitor off the page. None of that is advertising, and all of it lands in the same ROAS. It is written up in the Portrelle case study.
Then the page the ad points at
Conversion rate multiplies everything upstream of it, which is why it is worth more than a bidding change. A store that converts at 1.4 per cent and gets to 1.8 per cent has bought itself a twenty nine per cent lift in orders on identical spend and identical traffic. No campaign edit available to you does that, and the gain does not decay when the auction gets more expensive. What we look at first is on the conversion page.
Make the second order arrive without paying for it
Repeat purchase is the only lever that lowers your break even instead of raising your return. On the brand we owned, 72.7 per cent of email revenue came from automated flows rather than from campaigns, and a flow only fires when somebody has already arrived, which means the flows are paid for by advertising that already happened. Build them in the order that matters and the first order is allowed to cost more. The Klaviyo page has the order we build them in.
Now the account
- One conversion action, valued properly, deduplicated. Most accounts we open are optimising towards two or three overlapping actions, so the bidding is chasing a number that counts some orders twice.
- Audience signals are a hint, not a target. In Performance Max they tell the system where to start looking. Treating them as targeting produces a campaign that quietly ignores you.
- Check what the catalogue actually shows. Approved is not the same as visible, and a campaign reads one feed label. That note is here.
- Budget lost impression share, per campaign. If a campaign is capped, its ratio is an artefact of the cap and not of the targeting.
- Enough creative to keep the auction fresh, pointed by a brief. We direct creative and say what to make next. We do not produce it, and a partner who does both has a reason to keep making more.
The improvement that is always real
Margin. Raise the price or lower the cost of goods and your break even ROAS falls, so the exact same campaign, unchanged, becomes profitable. It is the only item on this list that improves the ratio and the profit in the same movement, and it is the one that never appears in an advertising proposal, because it is not advertising.
How long before ROAS moves?
Give any change one full purchase cycle plus the attribution window before you read it, which for most stores means four to six weeks. Anything shorter is measuring the weather. The one exception is a broken conversion action, where the number is wrong rather than low, and that is worth fixing the same day.