Target ROAS for Shopping campaigns, explained
How tROAS bidding actually works in Google Shopping, what the algorithm optimises, how to pick a starting target, why product-level analysis beats campaign averages, and the ramp-down trap.
Target ROAS tells Google 'get me £X of revenue for every £1 of spend', and Google adjusts bids per auction to hit that average. Set it too high and Google stops spending; too low and you buy unprofitable clicks. The catch: one campaign-wide target flattens wildly different products into one average, which is why product-level tROAS analysis matters.
What Google actually does with your target
With tROAS, Google predicts conversion value for every individual auction (this query, this shopper, this device, this hour) and bids exactly as much as that predicted value justifies against your target. A 400% target means Google bids up to the point where predicted revenue ÷ cost = 4.
Two consequences follow:
- The target is an average, not a floor. Google will happily take a 200% auction if it's also winning 800% auctions elsewhere. Your reporting will show variance around the target, and that's the system working.
- The target is a throttle. Raising it doesn't "demand more efficiency": it shrinks the set of auctions Google considers worth entering. Raise it far enough and spend collapses. This is the single most common self-inflicted wound in Shopping accounts.
The averaging problem
A campaign-wide target treats a £9 accessory with 60% margin and a £600 hero product with 15% margin as the same business. Under one 400% target, the algorithm quietly starves whichever products can't clear the bar and overfeeds the ones that clear it easily: regardless of which products you actually want to grow.
The fix is granularity: group products whose economics genuinely match, and give each group its own target. Margin bands, price bands, and lifecycle stage (new vs proven) are the groupings that pay. Custom labels carry those groupings from your feed into campaign structure.
Deep dive Product-level tROAS and the thin-data problem
Even within a well-segmented campaign, per-product performance diverges: and Google's own reporting only shows you what happened, not what target each product could sustain. Product-level tROAS analysis works backwards from each product's conversion history, margin and price point to the target it could individually bear, then surfaces where the campaign target is leaving money on the table (products that could grow at a lower target) or burning it (products that never clear the current one).
The hard constraint is data thinness: a product with four conversions has no statistically usable ROAS. Standard treatments:
- Pooling: thin products inherit priors from their category/price-band siblings until they have their own signal.
- Incubation: new and thin-data products run in a separate Maximise Clicks campaign with its own budget, where the goal is buying data cheaply, not efficiency. They graduate to the tROAS campaign once they've proven conversion behaviour. This is the pattern BidSmart's incubator automates: including the graduation criteria and the move itself, staged for your approval.
- Ramp discipline: target changes limited to small relative steps so learning never resets. BidSmart's recommendations follow the same rule: staged adjustments, each one queued for explicit sign-off before anything executes.
When tROAS is the wrong tool
- Brand-new campaigns or products: no conversion history means the prediction engine is guessing. Incubate first.
- Hard cash-flow caps: tROAS spends unevenly (heavy on strong days). If a strict daily budget matters more than efficiency, budget-constrained tROAS behaves badly; fix the budget first.
- Lumpy, high-value conversions: a few large orders per month whipsaw the algorithm. Consider longer conversion windows or value rules.
The ramp-down trap (read this before touching the target)
The classic failure sequence: performance looks good → target raised 30% in one step → spend drops → panic → target dropped → learning resets → two lost weeks. The algorithm reads a big target jump as "most auctions are no longer acceptable" and exits them immediately, but the recovery when you reverse is slow because learning has to rebuild. Move in 10–20% relative steps, wait out the adjustment period, and judge on 2+ weeks of data: or let a system that watches per-product pacing daily propose the steps for you.
Frequently asked questions
What tROAS should I start with?
Start from your actual trailing 30-day ROAS, not your aspiration. If the campaign currently returns 320%, set 300–320% and tighten gradually. Jumping straight to a 500% target from a 320% reality tells Google to stop buying most of your traffic.
How much conversion data does tROAS need?
Google's formal minimum is low, but in practice tROAS behaves erratically below roughly 30–50 conversions per month per campaign. Below that, Maximise Clicks or Maximise Conversion Value (no target) builds data faster, this is exactly why new products belong in an incubator campaign.
How often can I change the target?
Small steps (10–20% relative), no more than once or twice a week. Every change triggers a re-learning period; large or frequent changes keep the algorithm permanently confused.