Target CPA (tCPA) is the Smart Bidding strategy for accounts that count conversions rather than value each one individually: you set the average cost per acquisition you will accept, and Google bids per auction to hold that average. It is the natural fit for lead generation, where a form fill is a form fill until the CRM says otherwise. The setting is one field. The reason it so often "does not work" is never the field — it is that nobody in the account ever decided what a lead is worth.
How Target CPA works
tCPA is an average and a steering input, not a cap. Some conversions cost more than the target, some less; the model manages the aggregate over time. Given a target, it prices each auction by predicted conversion probability — bidding where the expected cost lands inside your promise, sitting out where it will not. Like every Smart Bidding strategy, it is downstream of your tracking: if conversion actions over-count or under-count, the bidder optimises the miscount faithfully.
The behaviour to internalise: the target moves volume. Tighten it and the model exits auctions it cannot win at your price — fewer conversions, cheaper each. Loosen it and participation widens. Neither is malfunction. The dial is the trade.
Deriving an honest target
The afternoon of arithmetic works backwards from money, in three steps:
- What is a customer worth? Average deal value — first order or lifetime; choose deliberately and remember which you chose.
- What fraction of conversions become customers? The close rate, from your CRM — the number this exercise usually reveals nobody actually knows.
- What share of customer value may acquisition consume? A margin decision, made by a human, once.
Multiply the three and you have the most a conversion can cost you. That is the ceiling; your target sits at or under it. Without step two, any tCPA is a dart throw — the same lead cost is brilliant at one close rate and ruinous at half of it.
Ramp targets in small steps
The learning system calibrates to the current target, so yanking it hard makes delivery lurch. The working discipline: move in small steps — on the order of ten to twenty percent — with a week or two between moves, so the system re-equilibrates and you can attribute what changed. The same patience applies to reading results: judge a target change on a window, not a weekend. With thin conversion volume, portfolio strategies pool learning across campaigns — and sometimes the honest interim answer is a targetless spell on Maximize Conversions while data accrues.
The standing health checks, run by prompt
tCPA health is exactly the kind of standing vigilance conversational account access runs best — on demand, cold and complete, in the time it takes to ask. The standing checks: delivered CPA versus target per campaign over the trailing thirty days, gaps flagged with direction; conversion volume trend since the last target change — the strangulation check; which targets changed recently, by whom, and what followed.
Your agent runs all three in minutes and lays the evidence in front of you. Where the maths says the target is dishonest, it works out the corrected number and stages it as a change you approve — reasoning attached, every edit logged. The afternoon of arithmetic still happens once, by you; keeping the target honest afterwards is exactly the vigilance worth handing to an agent that never looks away. If your conversions carry genuinely different values, the sibling strategy is Target ROAS.