Skip to content
← All notes

Notes

Set your tCPA from your real CPA, then stop touching it

Target CPA, or tCPA, is the most fiddled-with setting in a Google Ads account, and the fiddling is usually what breaks it. Two mistakes do most of the damage. You set the target off a number you wish were true, and then you change it every time the cost per conversion moves.

Set it from reality, not a wish

The target is not a guaranteed setting. It is a signal to Google about how strict you want to be, and where you want to sit between exploring for new demand, across audiences, keywords, times, and devices, and leaning hard on what has already converted. Both ends are valid configurations depending on your goal. It is not, however, a number you can wish below your real performance for free. Pull your actual cost per conversion over the last few weeks and set the target around there, or a little above if you want room to explore and grow volume.

Set it well below anything the account has ever achieved and you have not made acquisition cheaper. You have told the algorithm to bid only when it is almost certain a click will convert, so it holds back and your daily spend falls with it. The conversions you do get, you could have had without throttling the volume like that. The cost per conversion on the screen looks healthy while the account does far less than it could.

Setting the first target with no history

On a brand new campaign there is no trailing CPA to read, so you estimate. Take a plausible click cost from the Keyword Planner, toward the high end of the top-of-page bid range for your keywords, and divide it by the conversion rate you expect. As rough rules of thumb for that rate, use around 15% for lead gen, 3% for eCommerce and SaaS, and 20% for low-friction signup offers like newsletters or waitlists. A $6 click at a 15% lead-gen rate puts your starting target near $40. It is an estimate, not the answer, so once real conversions come in you reset the target off the realized CPA.

When a target can actually work

A target only works if the rate you want, conversions per day or ROAS per day, is reachable from the rate you are already realizing. It also needs enough data under it to mean anything. tCPA can start working from around fifteen conversions a month, and more is better, but much below that a campaign is on small numbers, so the cost per conversion swings on noise and a tight target laid over thin data swings with it.

When the cost per conversion does swing, look at it day by day, not just at the monthly average. A campaign that averages $100 can be hiding plenty of $20 days, and that spread could be a signal. There is cheap conversion in there that a $100 target is not leaning into, and it can be the moment to cut the tCPA hard rather than nudge it down. As you can see, it gets complex fast, because in some situations cutting a tCPA by 70% makes sense, while in others the same cut would kill the campaign.

Change it in steps, not in reactions

Smart Bidding is constantly balancing exploration against exploitation, testing which clicks convert at your target while spending most of the budget on what already works. Moving the target nudges that balance, and the damage comes from moving it badly. Big reactive swings, or a change every time the CPA wobbles, never let the system settle, so the cost per conversion stays noisy and the reflex is to change it again, which is the thing keeping it noisy. You become the volatility you are trying to fix.

Cost per conversion over time, left alone versus adjusted on every swing Two small charts from the same starting cost per conversion. On the left, left alone, the line works down and holds low. On the right, adjusted on every performance swing, it is knocked back up at each change and never settles. COST PER CONVERSION OVER TIME left alone adjusted on every swing
Same starting cost per conversion, two ways to manage it. Left alone, it works down and holds. Adjusted on every performance swing (each triangle is a change), it gets knocked back each time and never gets there.

How to actually move it

Done deliberately, moving the target is fine and often necessary. Experienced operators even use tCPA and tROAS as levers to shape how the algorithm allocates spend, tightening or loosening the explore-exploit balance on purpose, though that is advanced work. For most accounts the safe version is simple. Move the target in steps of around ten percent, week over week, and give it time to re-stabilize before the next one. Never move it in reaction to a single day.

Set it from real data, change it in small paced steps, and give it the conversions it needs between changes. Most tCPA problems are really patience problems.