I spent two years telling a home services client his CPA was excellent: $38 per lead, volume up 40% quarter over quarter, dashboard all green. Then he showed me close rate and average ticket by ZIP code and job type. Half those cheap leads were work he barely broke even on.
We were buying conversions, not profit.
CPA treats every conversion as equal. ROAS weights them by value. If every conversion is worth roughly the same, CPA is simpler, more stable, and often right. If values vary, as they do in most businesses, CPA can quietly fill your calendar with the wrong work while the report looks great.
Google has also made this needlessly confusing. The old tCPA and tROAS labels now sit inside Maximize Conversions and Maximize Conversion Value. A lot of advice still says to pick whichever sounds more advanced. Here is the practical version: what each metric measures, when each one wins, how much data each needs, and when a hybrid makes sense.
Both metrics answer the same question: how efficiently did spend turn into outcomes? They count the outcome differently. ROAS asks how much value came back per dollar spent; CPA asks what each conversion cost, regardless of value.
Spend $10,000 for 100 conversions worth $50,000 in revenue. Your CPA is $100. Your ROAS is 5.0, or 500%. Same spend. Same campaign. Two lenses.
Which lens you optimize toward determines what Google’s bidding chases.
ROAS is conversion value divided by cost. A 4:1 ROAS means $4 in tracked value for every $1 in ad spend.
For Google to optimize toward ROAS, you need to feed it values that mean something:
Without meaningful values, tROAS has nothing useful to weight. It will chase much the same volume CPA would, only with more volatility.
CPA is cost divided by conversions. Spend $10,000 for 200 leads and your CPA is $50.
That simplicity is why CPA is stable. It is also why it fails in specific cases. A dental practice may count a cleaning lead and an implant lead as one conversion. CPA will happily buy more cleanings because they are cheaper and more plentiful. A plumber can have the same problem with emergency calls and quote requests.
CPA works when conversions are worth similar amounts. When they are not, it can optimize toward cheap, low-value actions and call it efficiency.
I used to put almost every new account on CPA because it was easier to explain. The client got a lead goal, we hit it, everyone was happy. I was wrong about how often that was actually right.
CPA earns its keep when each conversion is worth about the same and you need volume at a predictable cost. Think single-offer lead gen with similar close rates, a SaaS trial with narrow plan mix, or a business where margin per sale sits in a tight band.
CPA is also more forgiving with thin data. It only needs to learn which clicks convert, not how much each conversion is worth.
Pick CPA when most of these are true:
My rule now is simple: if I cannot put a defensible dollar value on conversion types, I do not pretend ROAS will sort it out. Set a target CPA from real margins, not what felt affordable last month.
If values are uniform, CPA can buy the same outcome as ROAS with less volatility.
ROAS exists for the opposite problem: conversions are not equal, and treating them equally costs you money.
I learned this on an ecommerce account doing about $80k a month. The hero SKU converted well at an 18% margin, while high-margin bundles at 45% barely got impressions. CPA looked fine. Profit did not. We switched to value-based bidding, fed actual revenue, then moved to margin-adjusted values. Budget shifted away from cheap winners, and ROAS rose 31% in five weeks while revenue stayed flat because the mix improved.
Use ROAS when value spread is the story:
The catch is data. ROAS needs enough value-bearing conversions for the model to distinguish a $50 click that leads to a $600 sale from one that leads to a $60 sale.
My practical floor is 50 or more value-bearing conversions in 30 days per bidding campaign. Below that, ROAS can hunt, restrict spend, and starve volume. I have watched a 500% tROAS target reduce a good account to 10 conversions a week because it hit the target by refusing to spend.
Before touching a bidding setting, I run every account through two filters:
Get either filter wrong and you spend weeks fighting the bidding algorithm instead of competing in the auction.
| Business Model | Primary Bid Goal | Why It Wins | Tactical Implementation |
|---|---|---|---|
| Ecommerce: varied catalog | Target ROAS | Order values can swing from $30 to $400. CPA tends to favor the cheaper-to-acquire $30 buyers. | Use Maximize Conversion Value with a Target ROAS derived from gross margin, not top-line revenue. |
| Ecommerce: single product or flat pricing | Target CPA | Each transaction produces identical revenue. Value modeling adds little signal. | Set Target CPA comfortably below gross profit per unit to protect unit profitability. |
| Local and home services | Target CPA or hybrid ROAS | If jobs are similar in value, CPA keeps call volume steady. If emergency jobs pay 4x routine maintenance, ROAS has a job to do. | Start on Target CPA; move to ROAS using conversion value rules or offline conversion imports as data matures. |
| B2B SaaS and high-ticket lead gen | Target CPA early, ROAS later | Sales cycles run 30 to 90 days. Early revenue optimization can starve the algorithm. | Optimize to Target CPA on qualified pipeline stages such as SQLs or demos, then move to ROAS once CRM revenue feeds back reliably. |
Google Ads Help documentation explains how automated bidding works. In account management, I use stricter working thresholds: around 30 conversions in 30 days for target CPA, and 50 or more value-bearing conversions for target ROAS, per campaign.
These are not laws of physics. They are practical floors. Run Target ROAS on a campaign with 12 conversions a month and one $1,500 order can pull bidding in the wrong direction. A dry week can trigger sudden budget constriction.
When volume is thin, stay on Target CPA or broad Maximize Conversions until transaction frequency builds.
Google made this simpler in name and messier in practice. What used to appear as separate Target CPA and Target ROAS choices now appears as targets within two broader strategies: Maximize Conversions and Maximize Conversion Value.
Maximize Conversions with an optional target CPA maps to the old Target CPA. Maximize Conversion Value with an optional target ROAS maps to the old Target ROAS. Without a target, each strategy spends toward as much of its chosen outcome as it can get. With a target, it tries to hold efficiency while scaling, as Google’s guidance explains.
Here is the mapping that matters:
I see one mistake more than any other: people set a target because the field exists. If you had 22 conversions last month and set a 400% tROAS target, you have not made bidding smarter. You have told it to stop spending unless it is highly confident it can hit 4:1. With thin data, it will stop spending a lot.
No target with learning beats the wrong target with confidence. Start from trailing actuals, not the goal slide. See how groas approaches autonomous paid search management.
I have audited hundreds of accounts where someone swore Google Ads stopped working. Usually, they had chosen the wrong bid metric or choked it with bad inputs.
If your trailing 30-day ROAS is 280% and you suddenly set the target to 500%, the algorithm does not discover elite buyers. It can stop bidding in auctions where it cannot clear that hurdle. Spend drops. Lead volume follows.
Google Ads guidance recommends gradual target changes of 10% to 15%, rather than a dramatic reset every time someone updates a spreadsheet.
Assigning an arbitrary $50 value to every contact form and phone call means you are running Target ROAS on synthetic numbers. You get the volatility of value-based bidding without the weighting benefit.
Use Target CPA until you can pass back real margin or deal values through a CRM sync or Google offline conversion tracking.
A 600% ROAS on low-margin inventory can lose money. A 300% ROAS on private-label goods can generate substantial profit.
Never set ROAS targets from top-line revenue alone. Factor in product cost and operating overhead first.
The bid goal belongs at the campaign or portfolio level, not as a permanent label for the whole account. That is why mixed setups often make sense.
Yes. Bid strategies operate at the campaign and portfolio level, not across the entire account.
A common setup uses Maximize Conversion Value with Target ROAS on high-intent ecommerce catalogs with widely varying order values, while Maximize Conversions with Target CPA runs a showroom-booking campaign or top-of-funnel lead-generation push in the same account.
Google’s documentation allows Target ROAS on Search campaigns with as few as 15 conversions in 30 days. In practice, I want closer to 50 value-bearing conversions per month before relying on it.
Below that level, one unusually large purchase can skew bidding toward irrelevant search terms. A few dry days can severely restrict delivery.
Report contribution profit first. Then report the bidding metric that drives it.
CPA and ROAS are not status symbols. Neither pays the bills alone. If a $60 CPA delivers $400 in net gross profit while a $30 CPA delivers $50, the higher CPA wins every time.
Pick the bid goal that reflects how your business makes money, feed it honest data, and let the algorithm handle the mechanical work.
