Target ROAS is the most misused setting in google ads smart bidding. Set it too tight and the algorithm starves your campaigns of volume. Set it too loose and you hand margin back to Google. The right number comes from your margin math, and this is how to derive it.
What Google Ads smart bidding does with a ROAS target (and what it ignores)
When you set a Target ROAS, you are telling the algorithm the average conversion value per unit of spend it must defend. It then bids per auction based on predicted conversion value: aggressive where it expects high-value buyers, minimal or absent where it does not.
What it does not do: know your margins, your shipping costs, or which products you would rather sell. It optimizes toward measured conversion value and nothing else. If your tracking undercounts revenue, it bids on bad information. If your target sits above what the market allows, it simply stops entering auctions, and spend dries up without warning or explanation.
That last behavior is the throttling effect, and it is why an overambitious target looks exactly like a broken campaign.
Derive the target from margins, not platform averages
The wrong way: copy an industry benchmark or pick a round number that feels safe. A 4x target is profit on one product line and a loss on another inside the same store.
The right way starts with break-even ROAS: divide one by your contribution margin after product cost, shipping, payment fees, and fulfillment. A store keeping 40% of each order breaks even at 2.5x. From there, decide what ads must deliver on the first purchase. If repeat purchase is strong, you can run near break-even and let lifetime value carry profitability. If most customers buy once, set the target where first-order economics work on their own.
The number that comes out of this math is yours. It survives arguments, audits, and platform reps suggesting you lower it.
The learning phase: touching targets early costs you weeks
Smart bidding learns from conversion data, and every meaningful target change partially resets that learning. The expensive habit is reacting to three slow days by moving the target, which restarts the exact process that would have stabilized performance.
The working rule across my accounts: after launch or a major change, leave the target alone long enough for real conversion volume to accumulate, then judge on full weeks of data rather than days. When you do adjust, move in small steps. Each small step keeps most of the learning; a big swing throws it away.
Patience here is not passivity. It is protecting the data asset the algorithm runs on. The profitability timeline FAQ covers what a realistic ramp looks like.
Low ROAS: a bidding problem or a product and feed problem?
Before touching the target, diagnose which kind of problem you have, because the responses are opposite.
Bidding problems look like this: decent conversion rate once people land, sensible search terms, but unprofitable click costs or wild swings in volume. The auction strategy is the issue.
Product and feed problems look different: clicks are cheap and plentiful, but conversion is weak, or the search terms are wrong for what you sell. No ROAS target fixes a product page that does not convert or a feed that enters the wrong auctions.
The bathroom products account I rebuilt finished 146% up on ROAS, and the bidding changes only worked because the inputs underneath were fixed first. Targets amplify whatever sits below them. If yours keeps drifting down, the ROAS dropping FAQ walks the same diagnostic.
Segment targets by product tier, not account averages
A single account-wide target forces your hero products and your long tail to live under the same economics, which suits neither.
Split campaigns or asset groups by margin tier and give each its own number: hero SKUs with strong margins and conversion history can carry an aggressive growth target; mid-tier products run at the standard derived number; clearance and thin-margin lines get a defensive target or stay out of paid entirely.
The segmentation does not need to be elaborate. Even two tiers, heroes and everything else, puts budget where the economics are best instead of letting averages make the decision for you.
Scaling: phase the target as volume grows
Targets and volume trade against each other. A higher target means fewer, choicer auctions. A lower target opens volume at thinner returns. Scaling is the discipline of trading target for volume at a controlled pace.
The phased approach: once a campaign holds its target on consistent volume, either raise the budget while holding the target, or lower the target slightly to expand reach. One change at a time, with enough days between moves to read the result cleanly. The accounts that scale well are the ones where every step looks boring.
What breaks accounts is doubling the budget and tightening the target in the same week, then reading the chaos as a platform problem. The scaling without killing ROAS FAQ covers the sequence in more detail.
When to override smart bidding, and when to leave it alone
Manual intervention earns its keep in a few places: seasonality adjustments ahead of sale periods the algorithm has not seen, data exclusions after tracking outages so corrupted days do not poison learning, and structural moves like splitting a product tier out to its own target.
Where manual loses: shading bids by hour, device tweaks on gut feel, and daily target nudges. The algorithm sees auction-level signals you cannot, and second-guessing it at that altitude burns learning for no edge.
The division of labor that works: you own the economics, meaning targets, structure, and inputs. The machine owns the auctions.
The target is a business decision, not a platform setting
Everything above comes down to one shift: stop treating Target ROAS as a dial to fiddle with and start treating it as the encoded version of your unit economics. Get the math right, protect the learning phase, segment where margins differ, and scale in boring increments.
This is the day-to-day of my Google Ads management work. If you want your current targets sanity-checked against your margins, the free 48-hour audit does that math for you.