How tCPA actually works
tCPA takes your cost-per-conversion goal and turns it into a per-auction bidding rule. At each auction the model estimates how likely the click is to convert, then bids up when it can win the conversion below your target and down when it cannot. The target is an average the system aims for across many conversions, not a cap on any single one, so you will see individual conversions above and below it. It treats every conversion as equally valuable, which is the assumption that decides whether tCPA is the right tool at all.
tCPA vs tROAS: equal conversions vs variable value
tCPA cares only about cost per conversion and treats a 20 dollar order and a 2,000 dollar order identically. tROAS cares about revenue and will chase the high-value order harder. Use tCPA when conversions really are interchangeable, like a lead, a call, or a fixed-price signup. Use tROAS when order values vary widely, which is most eCommerce. Running tCPA on a store with a wide range of basket sizes quietly optimises toward cheap, low-value conversions and away from the expensive orders that actually fund the business.
Why a target set too low kills volume
A tCPA target is a request, not a guarantee, and there is a floor the auction will not go below without giving something up. Set a target far under what the account has ever achieved and the model does the only thing it can: it stops bidding on all but the surest, cheapest conversions, and volume collapses. The number looks great on the few conversions that survive while total conversions crater. The right target is anchored to proven performance, not ambition, then tightened gradually as tracking and volume earn the model's trust.
How to set a tCPA that holds volume
Start from what the campaign already delivers under maximize conversions, set the first tCPA at or slightly above that proven cost, and only then tighten. Move it 10 to 15 percent at a time and let each change run through a full conversion cycle before the next, so the model is never yanked back into learning. iClick derives the ceiling from margin first, using the maximum cost a conversion can bear and stay profitable, then approaches it in steps rather than demanding it on day one.
