Target CPA vs Target ROAS: Which Bidding Strategy and When
Target CPA and Target ROAS are the two most-used Smart Bidding strategies in Google Ads. Which one to use depends on whether your conversions carry meaningfully different values, how much conversion data you have, and whether your tracking reflects real revenue. Below, we explain both, when to use each, and the mistakes that cause accounts to underperform on both. The same target-setting logic carries over to AI Max campaigns.
- Target CPA works when all conversions are roughly equal in value; Target ROAS requires different conversion values.
- Start targets at current performance, not aspirational performance, and adjust in 10 to 15% increments.
- Below 30 conversions per month per campaign, Smart Bidding lacks enough signal to outperform manual.
- If the target is unrealistically low, impression share collapses while the metric appears to be met.
Why this matters
Smart Bidding decisions determine how Google allocates budget across every auction your account enters. Choosing the wrong strategy, or setting unrealistic targets on the right one, does not just reduce efficiency; it actively restricts delivery while appearing to work. An account running Target CPA at half its true achievable CPA will show a flattering cost-per-conversion on the small volume of conversions it captures and look fine until someone compares impression share to the prior period.
How each strategy works
Target CPA treats every conversion as equally valuable. The algorithm bids to achieve the target cost per conversion on average, sometimes bidding above it when confidence is high and below it when signals are weak. It is well-suited to lead generation where all qualified enquiries are worth roughly the same: a dentist booking a consultation, a trade business taking a call.
Target ROAS treats conversions as having different values and bids proportionally. An e-commerce account where some orders are £15 and some are £400 benefits from ROAS because the algorithm bids harder for signals that predict the £400 order and less for those that predict the £15 one. To do this it needs a value attached to each conversion event, either a fixed estimated value or a live revenue figure from the transaction.
When to use each
| Use Target CPA When | Use Target ROAS When |
|---|---|
| All conversions in the campaign are comparable in value | Conversion values vary meaningfully across orders or lead types |
| You are optimising to a lead or booking event | You have e-commerce revenue data or CRM deal values flowing back |
| Conversion volume is moderate and not widely variable | Volume is sufficient (50+ conversions per month) to learn value patterns |
| Your average order value or lead value is stable | You want the algorithm to prioritise higher-value customers automatically |
Both strategies require reliable conversion data to outperform manual bidding. An account with corrupted or incomplete tracking, covered in the measurement guide, will produce worse results under Smart Bidding than under manual because it scales the error efficiently.
Data requirements
Google’s documented minimums are 30 conversions in the last 30 days for Target CPA and 50 for Target ROAS per campaign. These are floors, not comfort zones. Below them the algorithm is essentially guessing from insufficient signal and will often underdeliver or overpay.
When accounts are below these thresholds, the practical options are to consolidate campaigns so the required volume comes from a smaller number of larger campaigns, or to optimise to a higher-volume proxy event upstream of the true conversion and use offline imports to add downstream quality signal. A service business that gets 15 booked consultations a month might optimise to phone calls and enrich bidding with closed-won data from the CRM.
Setting realistic targets
The starting target should reflect current performance, not desired performance. If the account is achieving a £60 CPA, setting a target of £35 immediately will cause the algorithm to restrict bids to the point of near-invisibility. The recommended approach: set the target at the actual current CPA, allow the algorithm to stabilise over two to four weeks, then reduce the target in 10 to 15% increments, pausing between adjustments to let the system adapt.
For ROAS, the same logic applies in reverse. Setting a target above what the account is currently achieving restricts delivery. Start at or slightly below current ROAS and adjust upward incrementally as the algorithm finds efficiency.
Common errors
- Setting targets based on goals, not data. The algorithm optimises to what is achievable in the auction. An aspirational target that the market cannot deliver produces restricted impressions, not improved efficiency.
- Switching strategies during a learning phase. Every significant change, including a target change, restarts the learning phase. Making multiple changes in quick succession keeps the campaign in a perpetual semi-learned state.
- Using Target ROAS with estimated conversion values. If every lead is assigned the same fixed value, ROAS bidding behaves identically to CPA bidding with extra steps. Only use ROAS when real value differences exist and are passed with the conversion event.
- Not monitoring the learning phase. During learning, performance is intentionally variable. The problem is that a genuinely broken campaign also looks variable; check impression share and conversion rate weekly rather than daily during the learning window.
Bidding strategy and account structure together determine what automated systems can do. The wider paid media context is in our performance marketing strategy guide and in the paid media management service.
Frequently Asked Questions
Target CPA instructs Google’s auction algorithm to set bids in order to achieve a specified cost per conversion on average across a campaign. It is suitable when all conversions in a campaign carry roughly the same value, such as qualified leads from a single service, and works best with adequate conversion volume.
Target ROAS instructs the algorithm to set bids aiming for a specific return on ad spend, meaning it weights bids higher for auction opportunities it predicts will result in higher-value conversions. It requires conversion values to be passed with each event, either fixed values or dynamic revenue data from a CRM or e-commerce system.
Google recommends at least 30 conversions in the past 30 days as a minimum for Target CPA, and more for Target ROAS because it also needs to learn value distributions. Below these thresholds the algorithm lacks the signal to outperform manual bidding reliably and may behave erratically.
If the target CPA is significantly lower than the account’s current cost per conversion, the algorithm restricts bids to the point where the campaign barely delivers impressions. If the target ROAS is set too high, the same thing happens. In both cases, impression share collapses while the target technically stays ‘met’ on the thin traffic that does show.
Ready to fix what’s costing you conversions?
We’ll review your acquisition funnel, show you exactly what’s underperforming, and hand you a clear, prioritised plan, whether or not you choose to work with us.
Get a clear diagnosis of your acquisition performance.
- 30-minute review
- Free, with no obligation
- Clear findings and priorities
- You keep the findings
30 minutes · Free · No obligation · You keep the findings